# Fileset

[nanomaterials-3261714-Revision_T.pdf](https://mdr.nims.go.jp/filesets/0500b5ae-96fc-4838-9523-4effc300749e/download)

## Creator

[Guohai Chen](https://orcid.org/0000-0001-8481-0972), [Dai-Ming Tang](https://orcid.org/0000-0001-7136-7481)

## Rights

[Creative Commons BY Attribution 4.0 International](https://creativecommons.org/licenses/by/4.0/)

## Other metadata

[Machine Learning as a “Catalyst” for Advancements in Carbon Nanotube Research](https://mdr.nims.go.jp/datasets/d18e1667-ca64-48e3-b496-33d57e8876f1)

## Fulltext

Type of the Paper (Article     Nanomaterials 2024, 14, x. https://doi.org/10.3390/xxxxx www.mdpi.com/journal/nanomaterials Review 1 Machine Learning as a “Catalyst” for Advancements in Carbon 2 Nanotube Research 3 Guohai Chen 1,* and Dai-Ming Tang 2,3,* 4 1 Nano Carbon Device Research Center, National Institute of Advanced Industrial Science and Technology 5 (AIST), Tsukuba Central 5, 1-1-1 Higashi, Tsukuba, Ibaraki 305-8565, Japan 6 2 Research Center for Materials Nanoarchitectonics (MANA), National Institute for Materials Science (NIMS), 7 Tsukuba, 305-0044, Japan 8 3 Institute of Pure and Applied Sciences, University of Tsukuba (Collaborative University), Tsukuba, 305-8571 9 Japan 10 * Correspondence: guohai-chen@aist.go.jp (G.C.), TANG.Daiming@nims.go.jp (D.T.) 11 Abstract: The synthesis, characterization, and application of carbon nanotubes (CNTs) have long 12 posed significant challenges due to the inherent multiple complexity nature involved in their pro- 13 duction, processing, and analysis. Recent advancements in machine learning (ML) have provided 14 researchers with novel and powerful tools to address these challenges. This review explores the role 15 of ML in the field of CNT research, focusing on how ML has enhanced CNT research by: (1) revo- 16 lutionizing CNT synthesis through the optimization of complex multivariable systems, enabling 17 autonomous systems, and reducing reliance on conventional trial-and-error approaches; (2) improv- 18 ing the accuracy and efficiency of CNT characterizations; and (3) accelerating the development of 19 CNT applications across several fields such as electronics, composites, and biomedical fields. The 20 review concludes by offering perspectives on the future potential of integrating ML further into 21 CNT research, highlighting its role in driving the field forward. 22 Keywords: carbon nanotube; machine learning; data-driven; chemical vapor deposition; synthesis; 23 characterization; application 24  25 1. Introduction 26 1.1. Machine Learning 27 Recently, machine learning (ML), a new paradigm, has demonstrated great potential 28 as a tool for materials research [1-3]. ML is a subset of artificial intelligence (AI) concerned 29 with the development and study of algorithms and statistical models that can learn from 30 data and generalize to unseen data and thus perform tasks without explicit instructions. 31 Instead of following predetermined instructions, ML systems learn from data to recognize 32 patterns, make decisions, and improve their performance over time [4-10]. 33 ML has shown a wide range of applications across various domains, such as materi- 34 als research, healthcare, finance, retail, autonomous vehicles, language processing, etc. 35 Specifically, in the field of materials science, ML has been applied to various aspects and 36 demonstrated great success [11]. Some of them are briefly discussed here (Figure 1).  37 (1) Materials discovery. ML has significantly accelerated the discovery of new mate- 38 rials, showing great advantages compared with traditional approaches which has been 39 slow and labor-intensive, involving extensive experimentation and trial-and-error pro- 40 cesses [12, 13]. ML offers a data-driven alternative, enabling the prediction of new mate- 41 rials with desired properties. With the assistance of ML, high-throughput screening al- 42 lows thousands or even millions of potential compounds to be evaluated computationally 43 using trained ML models before selecting a few candidates for real experimental testing. 44 Citation: To be added by editorial staff during production. Academic Editor: Firstname Last-name Received: date Revised: date Accepted: date Published: date  Copyright: © 2024 by the authors. Submitted for possible open access publication under the terms and conditions of the Creative Commons Attribution (CC BY) license (https://creativecommons.org/licenses/by/4.0/). Nanomaterials 2024, 14, x FOR PEER REVIEW 2 of 22   Additionally, inverse design becomes feasible, where researchers begin with the desired 45 material properties and use ML algorithms to identify the structures or compositions that 46 can achieve those properties. For instances, Rao et al. used an iterative scheme that com- 47 bines ML, density functional theory, experiments, and thermodynamic calculation to find 48 two new invar alloys with extremely low thermal expansion out of millions of candidates. 49 The alloys are both compositionally complex, high entropy materials, thus demonstrating 50 the power of this approach for materials discovery [14].  51 (2) Process optimization. ML is also revolutionizing the optimization of manufactur- 52 ing/synthesizing processes in materials science. These processes often involve multiple 53 complex variables, such as temperature, pressure, gas, and time, which must be carefully 54 controlled to achieve optimal results, however, the variable space is far too vast to ex- 55 haustively explore experimentally. ML enables more efficient optimization by analyzing 56 complex datasets and identifying the most critical parameters. For example, in the field of 57 chemical vapor deposition (CVD) synthesis of single-walled carbon nanotubes (SWCNTs), 58 Rao et al. utilized ML to find optimal synthesis conditions within one hundred experi- 59 ments for selectively growing SWCNTs within a narrow diameter range [15]. Lin et al. 60 predicted growth conditions toward addressing the synthetic trade-off between crystal- 61 linity and growth efficiency of SWCNT forests by training a ML regression model and 62 validated in less than 50 tests [16]. ML significantly advanced the process optimization, 63 especially for complex multivariable systems.  64 (3) Materials characterization. ML is being integrated into the characterization of ma- 65 terials, where it can reveal hidden patterns in complex datasets and automate the analysis 66 of experimental results [17, 18]. ML, particularly deep learning techniques such as convo- 67 lutional neural networks, is used to analyze microscopy images to automatically classify 68 microstructural features in the images, such as defects, phase boundaries, grain structures, 69 and particles. etc. ML is also used to analyze data from various spectroscopic techniques, 70 including Raman [19, 20], infrared [21, 22], and X-ray photoelectron spectroscopy (XPS) 71 [23-26]. ML can deconvolute complex spectra, identify chemical species, and quantify ma- 72 terial composition. For example, Dee et al. performed quantitative analysis of catalyst na- 73 noparticles for CNT synthesis using high-resolution, high-rate video capture of environ- 74 mental transmission electron microscopy experimentation coupled with automated image 75 processing, involving computer vision ML algorithms and convolutional neural networks 76 [27]. The automation using ML accelerates the analysis process and provides more con- 77 sistent and objective results [28]. 78 (4) Property Prediction. Predicting the properties of materials, such as mechanical, 79 thermal, electronic, and optical properties, based on their composition and structure is 80 one of key applications of ML in materials science. This is particularly useful for materials 81 where experimental property measurement is challenging or time-consuming. For in- 82 stance, Hajilounezhad et al. predicted mechanical properties, including stiffness and 83 buckling properties of CNT forests, using an image-based ML classifier model, showing 84 an accuracy of >91% [29].  85 (5) Design of experiments (DoE). ML is greatly enhancing the DoE by enabling more 86 efficient exploration of the experimental parameter space. Traditionally, DoE involves 87 systematically varying a few parameters to identify optimal conditions. However, this ap- 88 proach can be limited when dealing with high-dimensional spaces with many variables. 89 Leveraging ML analysis alongside the DoE approach allows for using limited resources, 90 particularly time, more efficiently and increases the likelihood of achieving a true opti- 91 mum. For example, Cao et al. successfully optimized the materials and devices of organic 92 photovoltaics via the combination of ML and DoE [30]. 93 In summary, ML is profoundly impacting various aspects of materials science, from 94 the discovery of new materials and the optimization of manufacturing/synthesizing pro- 95 cesses to the prediction of material properties and the automation of materials character- 96 ization. These advancements are accelerating the pace of innovation, reducing the time 97 and cost associated with materials development, and enabling the design of materials 98 Nanomaterials 2024, 14, x FOR PEER REVIEW 3 of 22   with unprecedented precision and desired functionality. As ML techniques continue to 99 evolve, their integration into materials science will likely become even more pervasive, 100 driving further breakthroughs and opening up new possibilities in this dynamic field. 101  102 Figure 1. Machine learning-assisted materials research in various aspects. 103 1.2. Carbon Nanotubes 104 Carbon nanotubes (CNTs) represent a sub-realm of materials science and a cross- 105 discipline field that intersects with chemistry and applied physics. CNTs, a specific allo- 106 trope of carbon, are cylindrical structures characterized by diameters in the nanometer 107 range and lengths that can span from nanometers to centimeters. Due to their unique one- 108 directional structures, CNTs exhibit exceptional mechanical, electrical, and thermal prop- 109 erties. Specifically, CNTs possess remarkable tensile strength, excellent electrical conduc- 110 tivity, and superior thermal stability. These qualities make CNTs highly promising for a 111 broad spectrum of applications, including electronics, composites, energy storage, and bi- 112 omedical engineering. In electronics, CNTs are being explored for their potential in field- 113 effect transistors [31, 32], touch panels [33], and field emitters [34, 35] due to their high 114 conductivity and electron mobility. In energy-related applications, CNTs have demon- 115 strated significant promise in supercapacitors [36, 37], hydrogen storage [38, 39], and lith- 116 ium-ion batteries [40, 41], where their high surface area and electrical properties improve 117 energy density and charge capacity. In the biomedical field, CNTs are being researched 118 for biosensors/bioelectrodes [42, 43] and drug delivery systems [44], taking advantage of 119 their biocompatibility and structural versatility. Additionally, through-silicon-via inter- 120 posers [45] and CNT-based dry adhesives [46, 47] are being developed for use in advanced 121 manufacturing and semiconductor technologies. 122 ML has also proven to be highly beneficial across numerous facets of research and 123 development of CNTs, given the nature of the involvement of a vast array of intricate and 124 multifaceted parameters. As a special one-dimensional molecule with a unique tubular 125 structure, controlling the atomic structures, characterizing the distinct chirality, and es- 126 tablishing the relationship between properties, structures and device performance for 127 CNTs are significantly more challenging. The complexity of CNT synthesis, which in- 128 cludes factors such as catalyst composition/structure, temperature, synthesis ambient, car- 129 bon feedstock, growth enhancers, and reaction conditions, presents additional significant 130 challenges. ML can address them by identifying patterns and optimizing processes that 131 would be difficult or time-consuming to achieve through traditional experimentation 132 methods. Moreover, ML has facilitated the interpretation of large datasets from CNT 133 Nanomaterials 2024, 14, x FOR PEER REVIEW 4 of 22   characterizations, enabling researchers to uncover subtle correlations and predict proper- 134 ties with greater accuracy. This ability to manage and analyze complex data sets has 135 proven indispensable in advancing CNT applications, where precise control over struc- 136 tural properties can lead to innovations in fields ranging from electronics to materials sci- 137 ence. Consequently, the role of ML plays in CNT research continues to grow, offering the 138 potential to unlock new possibilities and drive further progress in this cutting-edge area 139 of nanotechnology. 140 This short review briefly highlights the significant contributions of the ML approach 141 to CNT synthesis, characterizations, and applications, demonstrating how ML has sub- 142 stantially advanced the CNT research. Finally, the review concludes with a perspective, 143 offering insights into potential future developments and directions for deeper integrating 144 ML into CNT research to drive the field forward. 145 2. Machine Learning-Assisted CNT Synthesis 146 Synthesis of CNTs involves complex multivariable parameters. A typical CVD syn- 147 thesis process of CNTs includes catalyst preparation, catalyst nanoparticle formation, and 148 CNT growth, targeting the desired CNT structural properties (Figure 2a) [48]. As shown 149 in Figure 2b, all these processes are highly sensitive to a vast range of parameters includ- 150 ing, but not limited to, catalyst type (composition: commonly used elements such as Fe, 151 Co, Ni or their alloys; structure: single-layer or multi-layer; preparation method: sputter- 152 ing, evaporation, casting, spin coating, etc.), temperature, pressure, carbon feedstock (type 153 and concentration), growth enhancer (type and concentration), ambient (carrier gas type 154 and concentration), time, etc. [49-51]. These parameters must be meticulously controlled 155 to achieve CNTs with desired structural properties [52], typically including length [53], 156 crystallinity [54], diameter [55], wall number [56], surface area [57], alignment [58], den- 157 sity [59], etc.. Traditionally, optimizing these parameters has relied heavily on trial-and- 158 error approaches, which are time-consuming, labor-intensive, and may not fully capture 159 the complex interactions between these variables. ML has emerged as a powerful tool to 160 address these challenges, offering data-driven strategies for efficiently optimizing CNT 161 synthesis parameters and even enabling the development of autonomous synthesis sys- 162 tems. 163  164 Figure 2. (a) A typical CNT CVD synthesis process. Reprinted with permission from Reference [48]. 165 (b) Complex multivariable parameters in CNT synthesis process. 166 Nanomaterials 2024, 14, x FOR PEER REVIEW 5 of 22   2.1. Data-Driven Optimization of CNT Synthesis 167 Data-driven optimization generally refers to the use of ML algorithms to analyze 168 large datasets generated from CNT synthesis experiments and focuses on optimizing the 169 multitude of variables involved in the production process. By identifying patterns and 170 correlations within the dataset, ML algorithms can model the complex relationships be- 171 tween synthesis parameters and the resulting properties of CNTs. These models can pre- 172 dict how changes in synthesis parameters will affect the properties of the resulting CNTs, 173 facilitating the identification of optimal synthesis conditions, significantly reducing the 174 need for trial-and-error approaches and accelerating the discovery of targeted CNTs with 175 desired structural properties. This approach can also provide deeper insights into the un- 176 derlying mechanisms governing CNT growth. 177 In CNT synthesis, the parameter space is vast and multidimensional, as listed in Fig- 178 ure 2b. These parameters may interact in non-linear ways to influence the structural prop- 179 erties of CNTs. ML algorithms such as support vector machines (SVM), random forests 180 (RF), XGBoost, multilayer perceptron (MLP), Bayesian optimization, and artificial neural 181 network (ANN) are particularly effective in exploring this parameter space. They can 182 identify complex patterns and correlations that might be missed by traditional statistical 183 methods. These models have been previously applied in material synthesis and showed 184 excellent performance [16, 60-65]. In the field of CNT synthesis, successful applications of 185 data-driven approaches have also been demonstrated [15, 16, 29, 66-70].  186 For instance, Lin et al. demonstrated a route for addressing the synthetic trade-off 187 between CNT forest crystallinity and growth efficiency through the assistance of data- 188 mining approach [16]. Specifically, synthetic trade-offs in the synthesis of SWCNT forests 189 are very challenging as growing certain desired properties can often come at the expense 190 of other desirable characteristics. One example is the one between crystallinity and growth 191 efficiency, resulting in the difficult of achieving both high crystallinity and tall forest sim- 192 ultaneously (Figure 3a). To address this, Lin et al. used data mining approach to train a 193 machine learning regression model (XGBoost) using nine input feature descriptors with a 194 set of 585 experimental synthesis data (Figure 3b). Subsequently, 16000 exploratory “vir- 195 tual” experiments were performed by the trained model to examine potential routes to- 196 ward addressing the current crystallinity−height (growth efficiency) trade-off limitation, 197 and possible growth conditions were predicted (Figure 3c). Finally, validation experi- 198 ments showed very good agreement with the predictions, highlighting the effectiveness 199 and accuracy of the predictive capability of the ML model, which achieved improved re- 200 sults in less than 50 validation tests. Importantly and fundamentally, the trained model 201 revealed the surprising importance of the nature of the carbon feedstock, as a route for 202 indirectly maintaining a high level of crystallinity at a high growth efficiency through the 203 maintenance of the activity of the catalysts to overcome the trade-off. This data-driven 204 approach represents a significant advance in complex multivariable CNT synthesis sys- 205 tems. 206 Another example is that Ji et al. used ML for high-throughput screening of the effi- 207 cient growth of high quality SWCNTs [68]. A database of 1280 experiments was built with 208 4 input parameters of thickness of cobalt (Co) catalyst film, growth temperature, reduction 209 time, and carbon precursor flow rate and one output of Raman IG/ID ratio. ML models 210 including linear regression (LR), random forest regression (RFR), support vector regres- 211 sion (SVR), and ANN were trained and RFR was finally chosen due to its highest predic- 212 tion performance (Figure 3e). The trained model was used to further explore the parame- 213 ter space for the optimum growth conditions for high-quality SWCNTs with 38016 com- 214 binations of growth parameters (Figure 3f-h). Finally, the predicted optimum growth pa- 215 rameters were validated to achieve a high IG/ID value of 138 in conventional CVD growth 216 process. This efficient modeling, predicting, and learning ability of the combined high- 217 throughput and ML method shows great potential to speed up the controlled synthesis of 218 SWCNTs with specific structures and properties. 219 Nanomaterials 2024, 14, x FOR PEER REVIEW 6 of 22   In the data-driven approach, the size of the dataset plays a crucial role in determining 220 the performance and reliability of ML models. The general thinking is that larger datasets 221 are beneficial for training ML models because they can provide more information for the 222 model to learn from, reducing overfitting, leading to more accurate and robust predictions. 223 However, they also require more computational resources. Sufficient unbiased data is cru- 224 cial for successful ML training and validation [71]. The appropriate dataset size depends 225 on several factors, including the problem type, the level of complexity, the number of fea- 226 tures, data quality and error tolerance. This makes it a case-by-case scenario. The most 227 widely used rule-of-thumb is that the dataset size should be at least 10 times the number 228 of weights in the model. However, other guidelines also exist, such as requiring 50 to 1000 229 times the number of prediction classes or 10 to 100 times the number of features [72]. 230 Krasnikov et al. examined the implementation of ML techniques and discussed features 231 of the optimal dataset size and density for aerosol synthesis of SWCNTs with a complex 232 carbon source [69]. They originally employed a dataset of 369 points comprising 4 inputs 233 and 3 target parameters and assessed the performance of six ML methods. They showed 234 that even a dataset of 250 points with the inhomogeneous distribution of input parameters 235 is sufficient to reach an acceptable performance of the ANN model, wherein the error is 236 most likely to arise from experimental inaccuracy and hidden uncontrolled variables. 237 Therefore, careful consideration of dataset quality and input feature selection, rather than 238 just dataset size, seems to be more important. 239 In summary, the use of ML in data-driven optimization of CNT synthesis parameters 240 not only reduces the time and resources required for CNT synthesis experimentations, but 241 also enhances the reproducibility of results. By providing a deeper understanding of the 242 synthesis process, ML enables researchers to design experiments more strategically, lead- 243 ing to the discovery of new synthesis pathways and the production of CNTs with tailored 244 properties. 245  246  247 Figure 3. Data-driven optimization for CNT synthesis. (a-d) Data-mining assisted SWCNT forest 248 synthesis overcoming the synthetic trade-off. Reprinted with permission from Reference [16]. (a) 249 Trade-off between G/D ratio and CNT height. (b) Heat map of Pearson correlation coefficient matrix 250 for the nine inputs and two outputs. (c) Plot of virtual experiment data with prediction points be- 251 yond the boundary highlighting the successful access to the inaccessible region. (d) Validation ex- 252 periment results. (e-h) ML assisted SWCNT high-throughput screening for high-quality SWCNTs. 253 Reprinted with permission from Reference [68]. (e) IG/ID values from experimental measurements 254 and predictions from the RFR model. Dependence of predicted Raman IG/ID on the thickness of Co 255 catalyst film as a primary variable and (f) growth temperature, (g) reduction time, and (h) flow rate 256 through ethanol as secondary variables. 257 Nanomaterials 2024, 14, x FOR PEER REVIEW 7 of 22   2.2. Autonomous CNT Synthesis 258 Autonomous synthesis represents a paradigm shift in materials science, where ML 259 and automation technologies are integrated to enable fully automated production/synthe- 260 sis processes. In the context of CNT synthesis, this involves the development of systems 261 that can autonomously adjust synthesis parameters in real-time based on continuous feed- 262 back from ML algorithms, ensuring the production of CNTs with desired, precise and 263 consistent properties. This concept goes beyond traditional methods that rely on manual 264 adjustments and trial-and-error experimentation. Instead, autonomous systems can dy- 265 namically adjust synthesis parameters on-the-fly to optimize the quality and yield of 266 CNTs, significantly improving efficiency and reproducibility. 267 Due to the multivariable nature of CNT synthesis, a closed-loop autonomous synthe- 268 sis system is uniquely suitable for optimizing the synthesis process continuously and au- 269 tonomously by integrating real-time data monitoring, ML algorithms, and automated pro- 270 cess controls. As conceptually diagrammed in Figure 4a, a simple such system generally 271 starts with a CVD reactor to perform the synthesis process and collect the growth param- 272 eters. The system also needs to be equipped with in-situ monitoring and evaluation units, 273 such as Raman spectroscopy, thermogravimetric analysis (TGA), and mass spectrometry, 274 to provide real-time monitoring of the process. Then all these data are sent into a ML 275 control unit to analyze the data and predict the optimal synthesis conditions. Finally, the 276 decisions of new synthesis conditions are set and fed back into the CVD process. 277 Autonomous CNT synthesis has been successfully demonstrated in several experi- 278 mental setups [73-76]. For instance, a proof of concept was exemplified by Maruyama et 279 al. with adaptive rapid experimentation and in situ spectroscopy (ARES) system (Figure 280 4b). This system was developed to increase the experimentation rate by 100-fold, to 100 281 runs per day, with results analyzed in situ and in real time via Raman spectroscopy. Such 282 system was used to synthesize CNTs on the surface of micropillars and linear regression 283 modeling was used to map regions of selectivity toward SWCNT and multiwall CNT 284 (MWCNT) growth in the complex parameter space of 534 CVD experiments [73]. Recently, 285 the same group demonstrated a closed-loop system using Bayesian optimization (BO) as 286 an efficient and robust ML algorithm integrated with the ARES setup (Figure 4b). The 287 CNT growth rates were extracted from the ARES and fed into the BO algorithm. The al- 288 gorithm then generated a new set of conditions, run by ARES, and the new output data 289 was sent to the BO planner to update the existing dataset and plan a new experiment. 290 Such closed-loop system can significantly improve the predictive power, show good con- 291 sistency performance, exploit a complex parameter space, and achieve a 5-fold faster ex- 292 perimentation speed than before [75]. 293 More recently, Zhang’s group introduced an AI-driven platform, Carbon Copilot 294 (CARCO), which integrates transformer-based language models tailored for carbon ma- 295 terials, robotic CVD, and data-driven ML models. Using CARCO, a catalyst discovery was 296 demonstrated by predicting a superior Titanium-Platinum bimetallic catalyst for high- 297 density horizontally aligned CNT (HACNT) array synthesis, validated through over 500 298 experiments. With the assistance of millions of virtual experiments, an unprecedented 299 56.25% precision in synthesizing HACNT arrays with predetermined densities was 300 achieved, within just 43 days [76]. This work exemplifies a great advance towards the 301 integration of AI/ML with human expertise to overcome the limitations of traditional ex- 302 perimental approaches. 303 In summary, despite of some facing challenges, the implementation of autonomous 304 synthesis systems could revolutionize the CNT research by drastically reducing the time 305 and cost associated with traditional synthesis methods. This approach not only enhances 306 the scalability of CNT production but also allows for the precise control of CNT properties, 307 which is crucial for their integration into commercial applications such as electronics, 308 composites, and biomedical devices. 309  310 Nanomaterials 2024, 14, x FOR PEER REVIEW 8 of 22    311 Figure 4. Autonomous CNT synthesis. (a) A conceptual diagram of a closed-loop autonomous CNT 312 synthesis system. (b) A closed-loop autonomous system with adaptive rapid experimentation and 313 in situ spectroscopy (ARES) and Bayesian optimization ML algorithm. Reprinted with permission 314 from Reference [75]. 315 3. Machine Learning-Assisted CNT Characterization 316 The most essential question in materials science including CNT research is to estab- 317 lish the relation of the microstructures with synthesis conditions, physical and chemical 318 properties. Materials characterization is to extract information about the structure, includ- 319 ing imaging in real space and spectroscopy in energy space, respectively. By establish the 320 relationship between structure and properties, characterization is essential to understand 321 growth mechanism, design microstructures for desired properties. 322 For CNTs, structural information extracted from characterization includes purity, 323 morphology, diameter, length, conductivity type, and distribution of chiralities, and so 324 on. Transmission electron microscopy (TEM) was the key for the discovery of CNTs [77], 325 and for the precise characterization of individual CNTs, including the chiral indices [78]. 326 On the other hand, Raman spectroscopy has been widely used for quick and statistical 327 analysis of CNT samples, including purity, distribution of diameter and chirality [79, 80]. 328 In conventional materials characterizations, images and spectra are acquired and an- 329 alyzed manually one-by-one. It is challenging to reach the optimum acquisition parame- 330 ters because of the trade-offs in temporal resolution, spatial resolution, field of view, data 331 volume, acquisition speed, stability, irradiation damage, and signal-to-noise ratio (SNR). 332 For CNTs, from images in real space, diameters and lengths are measured. From diffrac- 333 tion patterns in reciprocal space, crystalline structures could be analyzed by measuring 334 the distances and angles of diffraction spots. And from spectra, peak positions are com- 335 pared with existing spectra in databases to determine the substances. The conventional 336 analysis process is time consuming and biased by human factors. 337 In recent years, ML has been applied to materials characterizations to automate the 338 analysis process and to recognize the hidden patterns in data [81-86]. It has been demon- 339 strated that ML can enhance the efficiency, increase the accuracy and most importantly to 340 handle large datasets through high-throughput analysis. In this section, ML -assisted CNT 341 characterization will be introduced with examples of rapid identification of Raman spec- 342 troscopy and automated chirality determination from TEM Images. 343 3.1. Machine Learning for Rapid Identification of Raman Spectroscopy 344 Raman spectroscopy is based on the inelastic scattering of photons with phonons. It 345 is widely used to analyze vibrational modes and to identify molecular structures with the 346 fingerprints in the position, width, shape of the peaks. Raman spectroscopy has been one 347 of the most important methods to characterize CNTs [80]. The purity and quality of CNTs 348 could be extracted from the ratio of G and D bands. In addition, diameter and chirality 349 distribution could be obtained from the radial breathing modes (RBM) modes [79, 87]. 350 One of the challenges for Raman spectroscopy is the low SNR, since inelastic Raman 351 Nanomaterials 2024, 14, x FOR PEER REVIEW 9 of 22   scattering cross-section is much smaller than the cross-section for elastic Rayleigh scatter- 352 ing. In recent years, ML algorithms have been applied for Raman spectroscopy-based clas- 353 sification and recognition [19, 68, 88, 89]. 354 For instance, Zhang et al. used deep learning to automate and speed up the analysis 355 of Raman spectra to identify the structure features of suspended CNTs, including position, 356 number and metallicity (Figure 5a-f) [88]. The deep learning model was based on a con- 357 volutional neural network (CNN) with four convolutional layers, a softmax layer, and a 358 cross-entropy layer for defining the loss function. A database was consisted of 62130 Ra- 359 man spectra from suspended CNTs. And the samples were labeled as semiconducting (S- 360 CNTs), metallic (M-CNTs), and empty trenches (no CNTs). The CNN was trained on the 361 dataset and tested on a separate validation set of 48887 spectra. After training the CNN, 362 softmax thresholding was applied to refine the classification, to reduce the number of false 363 positives, and to enhance the accuracy. The model achieved a classification accuracy of 364 90% for spectra with the SNR values as low as 0.9, and the accuracy was up to 98% for 365 higher SNRs. The accuracy was found to increase with longer integration times (>30 ms) 366 and higher laser power, with a maximum of 99% at 1 mW laser power. The deep learning- 367 assisted rapid identification of CNT structures from Raman spectroscopy can be inte- 368 grated into industrial production lines for real-time monitoring of CNTs growth. 369 In another example, Kajendirarajah et al demonstrated a rapid diagnosis and analysis 370 of tip-enhanced Raman spectroscopy (TERS) mappings of individual CNTs using deep 371 learning neural networks [89]. In this work, TERS maps were measured on SWCNTs. 372 TERS spatial resolution together with spectral selectivity is used to facilitate discrimina- 373 tion of semiconductive versus metallic characteristics of SWCNTs, the identification of 374 bundles, and the identification of defect areas. Two multi-purpose ANNs were created 375 and utilized to characterize SWCNTs. Using Raman spectra gathered from TERS experi- 376 ments, ANN Model 1 classified each spectrum as background or CNT with an associated 377 confidence percentage. ANN Model 2 provided clear and rapid filtering of the three vi- 378 brational modes of CNTs. For any given CNT Raman spectrum, ANN Model 2 accurately 379 identified the existence of these modes. Both ANN Model 1 and Model 2 were used to 380 generate TERS maps with enhanced contrast. ANN Model 1 and ANN Model 2 worked 381 synergistically to identify the number and the assignments associated with the different 382 SWCNTs vibrational modes. In addition, the developed approach enables enhanced vis- 383 ualization of CNT defect areas. The methodologies used to create the deep learning ANNs 384 have resulted in a 98% accuracy for ANN Model 1 and 96% accuracy for ANN Model 2. 385 The approach performed a diagnosis of unseen raw TERS hyperspectral data in only 4 to 386 6 hours for a spectral cube composed of 8000–10 000 set of spectra with 1600 points each. 387 In summary, the utilization of ML has enabled a fast and accurate identification and 388 classification of CNTs combing with Raman spectroscopy, resulting in a powerful and 389 comprehensive analysis tool. 390  391 Nanomaterials 2024, 14, x FOR PEER REVIEW 10 of 22    392 Figure 5. Deep learning-assisted rapid identification of Raman spectra from suspended CNTs. (a) 393 Raman imaging. (b) Data collection. (c) Raman spectra database labeling. (d) CNN deep learning 394 model. (e) Training and classification. (f) CNT identification. Reprinted with permission from Ref- 395 erence [88]. 396 3.2. Machine Learning for Automated Chirality Determination from TEM Images 397 TEM is a key instrument for characterizing nanomaterials such as CNTs. It combines 398 multiple functions, ranging from imaging with atomic resolution, chemical analysis of in- 399 dividual atom, mapping of local physical properties, to fabrication with nanometer or 400 even atomic precision. 401 For CNTs, it is quite a challenge to operate a modern TEM to precisely characterize 402 the atomic structures, because of the small atomic mass, small diameter, and complex cir- 403 cular geometry. It takes years to train a professional microscopist to master the variety of 404 skills, including high-resolution TEM (HRTEM) imaging, nano-beam electron diffraction 405 for chiral indices, STEM imaging and 4D-STEM for probing local structures and proper- 406 ties.  407 In recent years, ML methods have been applied to electron microscopy, ranging from 408 data modelling and data analysis to atomic fabrication [27, 90-94]. For example, Lin et al. 409 quantified the fluctuations in catalyst carbon content by using an automated, atomic-scale 410 structural analysis of the time-resolved ETEM images. Such fluctuations in the composi- 411 tion of catalysts were correlated with the SWCNT growth rate [28]. Förster et al. developed 412 a deep learning approach for determining the chiral indices of CNTs from HRTEM images 413 (Figure 6a) [93]. Since the chiral indices entirely determine the atomic structures, elec- 414 tronic and optical properties, it is critical to identify the chiral indices of individual CNTs 415 and more important for the statistical distribution of chirality in a CNT sample. Electron 416 diffraction is the most precise method to determine CNTs’ chiral indices, however, indi- 417 vidual CNTs as long as 100 nm are required. In this work, Förster decided to use HRTEM 418 images to determine the CNTs chiralities, since sub-angstrom resolution is achievable for 419 modern Cs-corrected TEMs.  420 Due to the lack of enough high-quality experimental images of CNTs, a database was 421 constructed by high-throughput atomistic structural simulation and followed by TEM 422 simulations to generate HRTEM images. To increase the diversity of the data and gener- 423 ality of the model, variability was included, such as position, orientation, magnification, 424 defocus, aberration, noise, and so on. In total, all possible 261 chiral indices for the CNTs 425 with diameter of 0.48 nm to 2.30 nm were considered. And for each chirality, 5000 images 426 were simulated to general a database of 1.3 million images. A CNN based on LeNet-5 was 427 Nanomaterials 2024, 14, x FOR PEER REVIEW 11 of 22   used to model and classify simulated TEM images to identify the chiral indices. To en- 428 hance the modelling precision, two CNNs were used for the diameter and chirality, re- 429 spectively. The first CNN could determine the diameter with an accuracy of 99 %. And an 430 overall accuracy of 90.5% was found for the second CNN in classifying chiral indices. The 431 CNN based chirality classification system was evaluated by using experimental HRTEM 432 images. When images of sufficiently high quality were used, 71% of the results were con- 433 sistent for classification done manually and automatically. 434 In this pioneering work, ML has demonstrated enhanced efficiency to identify the 435 chiral indices, which typically takes 15-30 minutes to analyze one picture manually. And 436 the model showed a high robustness, for the nanotubes with defects. This method opens 437 the door for high-throughput analysis of HRTEM images to get insight into the distribu- 438 tion of chirality to understand the growth mechanism of CNTs. 439 As is well known, the structures of CNTs are predominantly determined by the char- 440 acteristics of the catalyst. Therefore, analyzing catalyst nanoparticles is essential to under- 441 standing the CNT growth mechanisms. However, due to their dynamic nature and na- 442 noscale size, quantitatively studying nanoparticle morphologies and conducting statisti- 443 cal analyses on them is extremely challenging. Recently, Lee et al. developed a mass- 444 throughput method for analyzing nanoparticle morphologies by applying a genetic algo- 445 rithm to an image analysis technique (Figure 6b) [94]. This approach enabled the analysis 446 of over 150000 nanoparticles with a high precision of 99.75% and a low false discovery 447 rate of 0.25%. The study also introduced clustering techniques to group nanoparticles with 448 similar morphological shapes for extensive statistical analysis. It was determined that an- 449 alyzing at least 1500 nanoparticles is required to represent the total nanoparticle popula- 450 tion at a 95% credible interval. Additionally, the number of TEM images and the average 451 number of nanoparticles per image should be considered to ensure a representative sam- 452 ple. The statistical distribution of polydisperse nanoparticles was also found to be critical 453 in accurately estimating their optical properties.  454 In summary, combing ML with TEM techniques not only enhances the efficiency and 455 accuracy of CNT analysis but also enables precise characterization of catalyst nanoparticle 456 morphology. This integration provides a powerful tool to significantly advance the field 457 of CNT research. 458  459  460 Figure 6. (a) Architecture of the CNNs for determining the chiral indices of CNTs from HRTEM 461 images. Reprinted with permission from Reference [93]. (b) Statistical characterization of the mor- 462 phologies of nanoparticles through ML-assisted TEM image analysis. Reprinted with permission 463 from Reference [94]. 464 4. Machine Learning-Assisted CNT Applications 465 Besides the application in CNT synthesis, ML also plays a crucial role in enhancing 466 the CNT applications across various fields to improve their integration into devices and 467 materials and predict their performance. This section explores how ML aids CNT appli- 468 cations in electronics and sensors, composite materials, and the biomedical field. 469 Nanomaterials 2024, 14, x FOR PEER REVIEW 12 of 22   4.1. CNT Applications in Electronics and Sensors 470 CNTs possess exceptional electrical properties, such as high conductivity, ballistic 471 transport, and excellent current-carrying capacity, making them ideal for electronic de- 472 vices and sensors [31]. However, integrating CNTs into practical devices requires precise 473 control over their properties, processing, and placement, which can be challenging due to 474 their nanoscale dimensions and variability in structural properties. ML can help overcome 475 the challenges by providing data-driven insights and optimization strategies [95-101].  476 For instance, Tadokoro et al. demonstrated a simple and AI-assisted method for the 477 more efficient fabrication of a massive CNT-based nanocantilever (Figure 7a), compared 478 to traditional fabrication methods using CVD and/or dielectrophoresis, which generally 479 contain manual, time-consuming processes such as the placing of additional electrodes 480 and careful observation of single-grown CNTs [97]. They trained deep neural network to 481 recognize the randomly positioned single CNTs on the substrate, measure their positions, 482 and determine the edge of the CNT on which an electrode should be clamped to form a 483 nanocantilever. This AI assistance significantly improved the recognition and measure- 484 ment processes, 2 sec versus 12 h by manual processing. 485 Another example is that Aliyana et al. utilized ML approach to accurately correlate 486 the impedance variations in zinc oxide/MWCNT nanocomposite (F-MWCNT/ZnO-NFs) 487 to NH4+ ion concentrations [98]. The proposed NH4+ sensor along with the decision-mak- 488 ing ML model can identify and operate at specific operating frequencies to continuously 489 collect the most relevant information from a system (Figure 7b). In addition, Bian et al. 490 utilized ML techniques (LR, RF) to create a calibration method for Hg2+ sensors based on 491 CNT field-effect transistor (FET) devices (Figure 7c) to solve sensor response saturation 492 [99]. Such application of ML techniques to investigate which features in the FET signal 493 maximally correlate with concentration changes provide valuable insight into the CNT 494 sensing mechanism and assist in the rational design of future nanosensors. Furthermore, 495 Fan et al. proposed an efficient framework for optimizing the design of CNT FET through 496 the integration of device physics, ML, and multi-objective optimization to expedite the 497 early-stage development of advanced transistors [100]. Moreover, Kelich et al. trained ML 498 models (convolutional neural network, SVN) to predict DNA sequences with strong opti- 499 cal response to neurotransmitter serotonin (Figure 7d). They discovered five DNA- 500 SWCNT sensors with higher fluorescence intensity response than those using only man- 501 ual screening method [101]. 502 In summary, these studies collectively highlight the transformative potential of ML 503 approaches in advancing the development of CNT applications in electronics and sensors. 504 These examples underscore the potential of ML to not only solve specific technical chal- 505 lenges but also to revolutionize the entire process of CNT-based device development. By 506 automating complex analyses, optimizing sensor design, and providing rapid, data- 507 driven insights, ML approaches are poised to play a pivotal role in accelerating the inno- 508 vation and deployment of CNT technologies in various electronic and sensing applica- 509 tions. 510  511 Nanomaterials 2024, 14, x FOR PEER REVIEW 13 of 22    512 Figure 7. (a) AI-assisted framework for fabricating nanocantilevers, using a neural network to rec- 513 ognize CNTs and a Python code to generate electrode patterns, and the final massive fabrication 514 process. Reprinted with permission from Reference [97]. (b) NH4+ selective impedance sensors fab- 515 ricated by embedding F-MWCNT/ZnO-NF active layers on interdigitated arrays. Reprinted with 516 permission from Reference [98]. (c) Schematic diagram for the composition of sensing material and 517 liquid-gated CNT FET. Reprinted with permission from Reference [99]. (d) ML-assisted approach 518 to identify DNA sequences in DNA−SWCNT conjugates with high serotonin response. Reprinted 519 with permission from Reference [101]. 520 4.2. CNT Applications in Composite Materials 521 CNTs are widely used in composite materials to enhance their mechanical, thermal, 522 and electrical properties [45]. However, achieving uniform dispersion and proper align- 523 ment of CNTs within the composite matrix is challenging due to the strong van der Waals 524 forces between the nanotubes and their tendency to agglomerate. ML provides powerful 525 tools to optimize these processes and predict the resulting composite properties, thereby 526 enhancing the performance of CNT-based composites across various applications [102- 527 112]. 528 For instance, Yu et al. utilized ML techniques to explore the structure-property link- 529 ages of CNT-reinforced AlSi10Mg nanocomposites [102]. The proposed processing frame- 530 work pipeline starts with SEM image processing of cellular microstructural features, fol- 531 lowed by training of ML models (AdaBoost, gradient tree boosting, K-nearest neighbors, 532 decision tree, and extra trees regressors), and property prediction (Figure 8a). The devel- 533 oped models demonstrated satisfactory performance, with the extra trees regression 534 model predicting hardness with a 2.47% error and the decision tree regression model pre- 535 dicting relative mass density with a 1.42% error. This framework can be applied to process 536 optimization and mechanical property manipulation for designing new alloys or compo- 537 sites. 538 In another example, Ranaiefar et al. investigated structures and mechanical property 539 predictions of CNT-reinforced acrylonitrile butadiene styrene (ABS) composites fabri- 540 cated by 3D-printing using ML approach [103]. Multiple regression algorithms were eval- 541 uated, including ridge regressor, linear regression, k-neighbors regressor, gradient boost- 542 ing regressor, random forest regressor, extra trees regressor, decision tree regressor, and 543 lasso regression. The predictive classification and regression supervised ML models sup- 544 ported the experimental results with 0.92 accuracy and a 0.96 coefficient of determination 545 (Figure 8b). This approach utilizing ML techniques can help inform and guide the design 546 of 3D-printed structures for targeted performance. 547 Furthermore, Jalal et al. demonstrated a concept of ‘Big Data’ analytics in composite 548 structure with focusing on functionally graded CNT-reinforced composites (FG-CNTRC) 549 Nanomaterials 2024, 14, x FOR PEER REVIEW 14 of 22   using a mesh-free method and an optimized neural network (ONN) approach to study 550 the effect of structural parameters on vibrational frequence [104]. They built a big data 551 containing 15625 entries with six parameters and developed an optimized ONN model 552 for predictive modeling. The ONN model demonstrated computation speeds thousands 553 of times faster than the mesh-free method while keeping a simulation error as low as 1%, 554 suggesting it is an efficient and reliable approach for handling big data in the optimization 555 and design of composites. 556 Moreover, different ML techniques have been utilized to study various CNT-based 557 composites for property improvements either in combination with experimental investi- 558 gations, such as CNT-reinforced cementitious composites [105-107], CNT-reinforced pol- 559 ymeric composites [108-111], or alongside computer simulations [112]. 560 In summary, ML is revolutionizing the development and application of CNT-based 561 composite materials by providing powerful tools for optimizing fabrication processes, 562 predicting composite properties, and guiding the design of composites with tailored per- 563 formance characteristics. By leveraging large datasets and advanced ML algorithms, re- 564 searchers can overcome the challenges associated with CNT dispersion and alignment, 565 enabling the production of high-performance composites for a wide range of applications. 566 As ML technologies continue to evolve, their integration with composite material research 567 and manufacturing processes will likely lead to even greater advancements in this field. 568  569  570 Figure 8. (a) Schematic of the proposed processing framework pipeline for AlSi10Mg nanocompo- 571 sites microstructure-properties linkages. Reprinted with permission from Reference [102]. (b) Opti- 572 cal microscopy cross-sections, in-plane with the print direction, of CNT reinforced ABS honeycomb 573 composite and predicted ultimate compressive strength from trained linear regression model eval- 574 uated against experimental measurements. Reprinted with permission from Reference [103]. 575 4.3. CNT Applications in the Biomedical Field 576 CNTs also possess unique properties, such as high surface area, biocompatibility, and 577 the ability to penetrate cell membranes, making them ideal for various biomedical appli- 578 cations [42, 113]. ML enhances these applications by providing tools for optimizing the 579 design of CNT functionalization, predicting their interactions with biological molecules, 580 enabling high-throughput selection of targeted DNA sequences, accelerating the discov- 581 ery of effective biomolecular constructs, and even predicting their genotoxicity [114-118]. 582 For instance, Ouassil et al. developed a random forest classifier (RFC, Figure 9a) to 583 investigate the relationship between a protein’s amino acid sequence and a protein’s bind- 584 ing propensity to SWCNTs [114]. The classifier aimed to predict which protein-SWCNT 585 interactions to expect in biological environments and to predict high-affinity protein bind- 586 ers to SWCNTs, along with protein features associated with such binding affinity to im- 587 prove the process of protein-nanoparticle construct design (Figure 9b). The model was 588 validated with a new set of proteins by performing quantitative protein adsorption 589 Nanomaterials 2024, 14, x FOR PEER REVIEW 15 of 22   experiments. The ML classifier can serve as a useful tool for understanding how protein 590 sequences influence nanotube binding. 591 Additionally, the use of single-stranded DNA (ssDNA) as a wrapping surfactant for 592 SWCNTs is a promising approach to functionalize and modify the SWCNT surface, ena- 593 bling precise control over SWCNT alignment. This approach has shown great potential in 594 biomedical applications, including drug delivery, gene therapy, and biosensors. Lee et al. 595 employed high-throughput systematic selection of high-affinity ssDNA sequences using 596 ML models (RF, MLP, CNN, etc.) from a vast random library (Figure 9c). The model ac- 597 curately distinguished high-affinity ssDNA sequences and provided predictive capabili- 598 ties for binding affinity, supporting the design of tailored DNA-SWCNT constructs. The 599 stability of ssDNA conformations on SWCNTs were validated by molecular dynamics 600 simulations[115]. 601 Furthermore, the prerequisite of utilizing DNA in sequence-dependent applications, 602 such as biomedical sensor, is to search specific sequences, which represents a significant 603 challenge due to the countless possible sequence combinations. Lin et al. demonstrated an 604 ML assisted experimental method for systematically searching DNA sequences based on 605 sequence-dependent recognition between DNA and SWCNTs (Figure 9d). Compared 606 with empirical search methods, their approach significantly improved both the number 607 of resolvable sequences and the success rate of finding them, from ∼102 to ∼103 and from 608 ∼10% to >90%, respectively [116]. Moreover, ML approaches have also been successfully 609 used to investigate CNT genotoxicity prediction [117] and discover molecular recognition 610 based on SWCNT corona phases [118]. 611 In summary, ML techniques enhance CNT applications in the biomedical field by 612 enabling precise and targeted functionalization of CNTs, improving their biocompatibil- 613 ity, and optimizing their interaction with biological systems. Additionally, ML-driven ap- 614 proaches provide powerful predictive tools to accelerate the discovery and development 615 of CNT-based biomedical constructs, expanding the possibilities for innovative therapies 616 and diagnostics. 617  618  619 Figure 9. (a) RFC workflow and (b) classifier performance results on different biofluid training da- 620 tasets and with varied protein feature inputs. Reprinted with permission from Reference [114]. (c) 621 Selection of high-affinity ssDNA sequences on SWCNT surfaces. Reprinted with permission from 622 Reference [115]. (d) ML assisted systematic search of DNA sequences. Reprinted with permission 623 from Reference [116]. 624  625 Nanomaterials 2024, 14, x FOR PEER REVIEW 16 of 22   5. Conclusion and Perspective 626 This review demonstrates the transformative impact of ML on CNT research, span- 627 ning optimization of growth conditions, automation of structure characterization, to ap- 628 plication development. ML algorithms have proven highly effective in modeling the com- 629 plex relationships between synthesis parameters and the resulting structures and proper- 630 ties of CNTs, by identifying patterns and correlations within the dataset. As a result, the 631 height of CNT forests and the quality of CNT thin films were effectively improved. In 632 addition, the integration of autonomous systems has further enhanced the growth pro- 633 duction, reducing reliance on manual trial-and-error processes. 634 ML has been widely adopted for materials characterization and has started to revo- 635 lutionize the CNT field by automating traditionally manual tasks. Successful applications 636 include determination of the chiral indices through HRTEM images and rapid identifica- 637 tion of suspended CNTs using Raman spectroscopy, both of which used deep learning for 638 automation. ML has demonstrated the enhanced efficiency to extract information to auto- 639 mate the analysis process, markedly improving the speed and accuracy of data analysis, 640 facilitating high-throughput studies and more consistent results. 641 Moreover, ML has been accelerating the innovation and deployment of CNT-based 642 applications in electronics and sensors, through automating complex analyses, optimizing 643 sensor design, and providing rapid, data-driven insights. ML has provided valuable guid- 644 ance to design composites with desired properties, through optimizing fabrication pro- 645 cesses, predicting composite properties. In the biomedical field, ML is playing a crucial 646 role in optimizing CNT functionalization and predicting their interactions with biological 647 molecules, furthering CNT applications in drug delivery, biosensors, and diagnostics. 648 Despite the significant progress, the integration of ML in CNT research is still in its 649 early stages and holds immense potential for exponential growth in the near future. Up 650 to now, application examples of ML in CNT characterizations are still very few. A search 651 using the key words “machine learning” and “carbon nanotube” in the “Web of Science 652 Core Collection” reveals only 131 publications, with the number of publications increases 653 in a linear manner from 2 in 2015 to >30 in 2024.  654 One of the main limitations is the lack of organized data, which is indispensable for 655 training complex ML models. Currently, there is no centralized database for CNT research 656 to include experimental data from various groups, with most advances being driven by 657 individual groups accumulating years of experimental data. The hunger for data is a com- 658 mon challenge in data-driven science. Typically, a scale of tens of thousands of data points 659 is required for high accuracy modelling of deep learning. However, it is time consuming 660 to do experiments. It takes hours to grow, seconds to take TEM images, and milliseconds 661 to obtain a Raman spectrum. Collaboration in CNT research community, among experi- 662 mentalists, theorists, and data scientists will be the key to construct shared CNT databases 663 and establish protocols for data storage, sharing, using, and importantly protecting the 664 precious data. 665 As a subset of ML, generative AI has gained a lot of attention, since it not only learns 666 but also generates new content, such as text and images, based on training data. In recent 667 years, natural language processing (NLP) models like GPT-4 can generate human-like text 668 and engage in conversations. Generative AI has been started to be applied to scientific 669 research including materials science, though the applications in CNT research are still few. 670 The potential for such tools is immense and should not be overlooked. Generative AI 671 models, such as generative adversarial networks (GANs) and variational autoencoders 672 (VAEs), could be highly useful in future CNT research for tasks like optimizing synthesis 673 pathways, predicting novel CNT structures, and enabling property-driven material dis- 674 covery. The integration of generative AI into CNT research has the potential to further 675 accelerate innovation by automating more creative aspects of material design and discov- 676 ery. As this area develops, we anticipate that generative AI will play a pivotal role along- 677 side ML, opening new frontiers in CNT research. 678 Nanomaterials 2024, 14, x FOR PEER REVIEW 17 of 22   Looking ahead, another important direction will be the development of integrated 679 AI systems that combine growth optimization, structural characterization, and property 680 measurements into a single closed-loop framework. By leveraging ML techniques, these 681 systems could continuously optimize the entire CNT research process, significantly accel- 682 erating discovery and innovation. As human researchers shift their focus to designing AI 683 systems and making key scientific judgments, this integrated AI-driven approach is ex- 684 pected to lead to revolutionary breakthroughs in CNT growth mechanisms, novel struc- 685 tures, and unprecedented applications. 686  687 Author Contributions: “Conceptualization, G.C. and D.T.; investigation, G.C. and D.T.; writing— 688 original draft preparation, G.C. and D.T.; writing—review and editing, G.C. and D.T.; visualization, 689 G.C.; funding acquisition, G.C and D.T. All authors have read and agreed to the published version 690 of the manuscript. 691 Funding: This work was supported by JSPS KAKENHI Grant Number JP23K04552. D.T. discloses 692 support from JSPS Kakenhi (grants JP25820336, JP20K05281, JP23H01796), JST-FOREST Program 693 (grant JPMJFR223T, Japan), WPI-MANA ‘Challenging Research Program (CRP)’, NIMS ‘Support 694 system for curiosity-driven research’, and "Advanced Research Infrastructure for Materials and 695 Nanotechnology in Japan (ARIM)" of the Ministry of Education, Culture, Sports, Science and Tech- 696 nology (MEXT). Proposal Number JPMXP1224NM5238. 697 Data Availability Statement: Not applicable. 698 Conflicts of Interest: The authors declare no conflicts of interest. 699 References 700 1. Mjolsness, E.; DeCoste, D. Machine learning for science: State of the art and future prospects. Science 2001, 293, 2051-2055, 701 https://doi.org/10.1126/science.293.5537.2051. 702 2. Ramprasad, R.; Batra, R.; Pilania, G.; Mannodi-Kanakkithodi, A.; Kim, C. Machine learning in materials informatics: recent 703 applications and prospects. npj Comput Mater 2017, 3, 54, https://doi.org/10.1038/s41524-017-0056-5. 704 3. Jose, R.; Ramakrishna, S. Materials 4.0: Materials big data enabled materials discovery. Appl Mater Today 2018, 10, 127-132, 705 https://doi.org/10.1016/j.apmt.2017.12.015. 706 4. Jordan, M.I.; Mitchell, T.M. Machine learning: Trends, perspectives, and prospects. Science 2015, 349, 255-260, 707 https://doi.org/10.1126/science.aaa8415. 708 5. Butler, K.T.; Davies, D.W.; Cartwright, H.; Isayev, O.; Walsh, A. Machine learning for molecular and materials science. Nature 709 2018, 559, 547-555, https://doi.org/10.1038/s41586-018-0337-2. 710 6. Huang, G.N.; Guo, Y.N.; Chen, Y.; Nie, Z.W. Application of Machine Learning in Material Synthesis and Property Prediction. 711 Materials 2023, 16, 5977, https://doi.org/10.3390/ma16175977. 712 7. Xu, P.C.; Ji, X.B.; Li, M.J.; Lu, W.C. Small data machine learning in materials science. npj Comput Mater 2023, 9, 42, 713 https://doi.org/10.1038/s41524-023-01000-z. 714 8. Wei, J.; Chu, X.; Sun, X.-Y.; Xu, K.; Deng, H.-X.; Chen, J.; Wei, Z.; Lei, M. Machine learning in materials science. InfoMat 2019, 1, 715 338-358, https://doi.org/10.1002/inf2.12028. 716 9. Choi, R.Y.; Coyner, A.S.; Kalpathy-Cramer, J.; Chiang, M.F.; Campbell, J.P. Introduction to Machine Learning, Neural Networks, 717 and Deep Learning. Transl Vis Sci Techn 2020, 9, 14, https://doi.org/10.1167/tvst.9.2.14. 718 10. Sarker, I.H. Machine Learning: Algorithms, Real-World Applications and Research Directions. SN Computer Science 2021, 2, 160, 719 https://doi.org/10.1007/s42979-021-00592-x. 720 11. Rickman, J.M.; Lookman, T.; Kalinin, S.V. Materials informatics: From the atomic-level to the continuum. Acta Mater. 2019, 168, 721 473-510, https://doi.org/10.1016/j.actamat.2019.01.051. 722 12. Raccuglia, P.; Elbert, K.C.; Adler, P.D.F.; Falk, C.; Wenny, M.B.; Mollo, A.; Zeller, M.; Friedler, S.A.; Schrier, J.; Norquist, A.J. 723 Machine-learning-assisted materials discovery using failed experiments. Nature 2016, 533, 73-76, 724 https://doi.org/10.1038/nature17439. 725 13. Tshitoyan, V.; Dagdelen, J.; Weston, L.; Dunn, A.; Rong, Z.Q.; Kononova, O.; Persson, K.A.; Ceder, G.; Jain, A. Unsupervised 726 word embeddings capture latent knowledge from materials science literature. Nature 2019, 571, 95-98, 727 https://doi.org/10.1038/s41586-019-1335-8. 728 14. Rao, Z.Y.; Tung, P.Y.; Xie, R.W.; Wei, Y.; Zhang, H.B.; Ferrari, A.; Klaver, T.P.C.; Körmann, F.; Sukumar, P.T.; da Silva, A.K.; 729 Chen, Y.; Li, Z.M.; Ponge, D.; Neugebauer, J.; Gutfleisch, O.; Bauer, S.; Raabe, D. Machine learning-enabled high-entropy alloy 730 discovery. Science 2022, 378, 78-84, https://doi.org/10.1126/science.abo4940. 731 https://doi.org/10.1126/science.293.5537.2051https://doi.org/10.1038/s41524-017-0056-5https://doi.org/10.1016/j.apmt.2017.12.015https://doi.org/10.1126/science.aaa8415https://doi.org/10.1038/s41586-018-0337-2https://doi.org/10.3390/ma16175977https://doi.org/10.1038/s41524-023-01000-zhttps://doi.org/10.1002/inf2.12028https://doi.org/10.1167/tvst.9.2.14https://doi.org/10.1007/s42979-021-00592-xhttps://doi.org/10.1016/j.actamat.2019.01.051https://doi.org/10.1038/nature17439https://doi.org/10.1038/s41586-019-1335-8https://doi.org/10.1126/science.abo4940Nanomaterials 2024, 14, x FOR PEER REVIEW 18 of 22   15. Rao, R.; Carpena-Nunez, J.; Nikolaev, P.; Susner, M.A.; Reyes, K.G.; Maruyama, B. Advanced machine learning decision policies 732 for diameter control of carbon nanotubes. npj Comput Mater 2021, 7, 157, https://doi.org/10.1038/s41524-021-00629-y. 733 16. Lin, D.; Muroga, S.; Kimura, H.; Jintoku, H.; Tsuji, T.; Hata, K.; Chen, G.H.; Futaba, D.N. Addressing the Trade-Off between 734 Crystallinity and Yield in Single-Walled Carbon Nanotube Forest Synthesis Using Machine Learning. ACS Nano 2023, 17, 22821- 735 22829, https://doi.org/10.1021/acsnano.3c07587. 736 17. Burnett, T.L.; Withers, P.J. Completing the picture through correlative characterization. Nat. Mater. 2019, 18, 1041-1049, 737 https://doi.org/10.1038/s41563-019-0402-8. 738 18. Pagan, D.C.; Phan, T.Q.; Weaver, J.S.; Benson, A.R.; Beaudoin, A.J. Unsupervised learning of dislocation motion. Acta Mater. 739 2019, 181, 510-518, https://doi.org/10.1016/j.actamat.2019.10.011. 740 19. Qi, Y.P.; Hu, D.; Jiang, Y.C.; Wu, Z.P.; Zheng, M.; Chen, E.X.; Liang, Y.; Sadi, M.A.; Zhang, K.; Chen, Y.P. Recent Progresses in 741 Machine Learning Assisted Raman Spectroscopy. Adv Opt Mater 2023, 11, 2203104, https://doi.org/10.1002/adom.202203104. 742 20. Guo, S.X.; Popp, J.; Bocklitz, T. Chemometric analysis in Raman spectroscopy from experimental design to machine learning- 743 based modeling. Nat. Protoc. 2021, 16, 5426-5459, https://doi.org/10.1038/s41596-021-00620-3. 744 21. Menevseoglu, A.; Entrenas, J.A.; Gunes, N.; Dogan, M.A.; Pérez-Marín, D. Machine learning-assisted near-infrared 745 spectroscopy for rapid discrimination of apricot kernels in ground almond. Food Control 2024, 159, 110272, 746 https://doi.org/10.1016/j.foodcont.2023.110272. 747 22. Windarsih, A.; Jatmiko, T.H.; Anggraeni, A.S.; Rahmawati, L. Machine learning-assisted FT-IR spectroscopy for identification 748 of pork oil adulteration in tuna fish oil. Vib. Spectrosc. 2024, 134, 103715, https://doi.org/10.1016/j.vibspec.2024.103715. 749 23. Penfold, T.; Watson, L.; Middleton, C.; David, T.; Verma, S.; Pope, T.; Kaczmarek, J.; Rankine, C. Machine-learning strategies 750 for the accurate and efficient analysis of x-ray spectroscopy. Mach Learn-Sci Techn 2024, 5, 021001, https://doi.org/10.1088/2632- 751 2153/ad5074. 752 24. Drera, G.; Kropf, C.M.; Sangaletti, L. Deep neural network for x-ray photoelectron spectroscopy data analysis. Mach Learn-Sci 753 Techn 2020, 1, 015008, https://doi.org/10.1088/2632-2153/ab5da6. 754 25. de Curtò, J.; de Zarzà, I.; Roig, G.; Calafate, C.T. Large Language Model-Informed X-ray Photoelectron Spectroscopy Data 755 Analysis. Food Control 2024, 5, 181-201, https://doi.org/10.3390/signals5020010. 756 26. Pielsticker, L.; Nicholls, R.L.; DeBeer, S.; Greiner, M. Convolutional neural network framework for the automated analysis of 757 transition metal X-ray photoelectron spectra. Anal. Chim. Acta 2023, 1271, 341433, https://doi.org/10.1016/j.aca.2023.341433. 758 27. Dee, N.T.; Schneider, M.; Zakharov, D.N.; Kidambi, P.R.; Hart, A. Automated processing of environmental transmission 759 electron microscopy images for quantification of thin film dewetting and carbon nanotube nucleation dynamics. Carbon 2022, 760 192, 249-258, https://doi.org/10.1016/j.carbon.2022.02.019. 761 28. Lin, P.A.; Gomez-Ballesteros, J.L.; Burgos, J.C.; Balbuena, P.B.; Natarajan, B.; Sharma, R. Direct evidence of atomic-scale 762 structural fluctuations in catalyst nanoparticles. J. Catal. 2017, 349, 149-155, https://doi.org/10.1016/j.jcat.2017.03.009. 763 29. Hajilounezhad, T.; Bao, R.; Palaniappan, K.; Bunyak, F.; Calyam, P.; Maschmann, M.R. Predicting carbon nanotube forest 764 attributes and mechanical properties using simulated images and deep learning. npj Comput Mater 2021, 7, 134, 765 https://doi.org/10.1038/s41524-021-00603-8. 766 30. Cao, B.; Adutwum, L.A.; Oliynyk, A.O.; Luber, E.J.; Olsen, B.C.; Mar, A.; Buriak, J.M. How To Optimize Materials and Devices 767 via Design of Experiments and Machine Learning: Demonstration Using Organic Photovoltaics. ACS Nano 2018, 12, 7434-7444, 768 https://doi.org/10.1021/acsnano.8b04726. 769 31. Tang, D.-M.; Erohin, S.V.; Kvashnin, D.G.; Demin, V.A.; Cretu, O.; Jiang, S.; Zhang, L.; Hou, P.-X.; Chen, G.H.; Futaba, D.N.; 770 Zheng, Y.; Xiang, R.; Zhou, X.; Hsia, F.-C.; Kawamoto, N.; Mitome, M.; Nemoto, Y.; Uesugi, F.; Takeguchi, M.; Maruyama, S.; 771 Cheng, H.-M.; Bando, Y.; Liu, C.; Sorokin, P.B.; Golberg, D. Semiconductor nanochannels in metallic carbon nanotubes by 772 thermomechanical chirality alteration. Science 2021, 374, 1616-1620, https://doi.org/10.1126/science.abi8884. 773 32. LeMieux, M.C.; Roberts, M.; Barman, S.; Jin, Y.W.; Kim, J.M.; Bao, Z.N. Self-sorted, aligned nanotube networks for thin-film 774 transistors. Science 2008, 321, 101-104, https://doi.org/10.1126/science.1156588. 775 33. Feng, C.; Liu, K.; Wu, J.S.; Liu, L.; Cheng, J.S.; Zhang, Y.Y.; Sun, Y.H.; Li, Q.Q.; Fan, S.S.; Jiang, K.L. Flexible, Stretchable, 776 Transparent Conducting Films Made from Superaligned Carbon Nanotubes. Adv. Funct. Mater. 2010, 20, 885-891, 777 https://doi.org/10.1002/adfm.200901960. 778 34. Choi, W.B.; Jin, Y.W.; Kim, H.Y.; Lee, S.J.; Yun, M.J.; Kang, J.H.; Choi, Y.S.; Park, N.S.; Lee, N.S.; Kim, J.M. Electrophoresis 779 deposition of carbon nanotubes for triode-type field emission display. Appl. Phys. Lett. 2001, 78, 1547-1549, 780 https://doi.org/10.1063/1.1349870. 781 35. Chen, G.H.; Shin, D.H.; Iwasaki, T.; Kawarada, H.; Lee, C.J. Enhanced field emission properties of vertically aligned double- 782 walled carbon nanotube arrays. Nanotechnology 2008, 19, 415703, https://doi.org/10.1088/0957-4484/19/41/415703. 783 36. Izadi-Najafabadi, A.; Yamada, T.; Futaba, D.N.; Yudasaka, M.; Takagi, H.; Hatori, H.; Iijima, S.; Hata, K. High-Power 784 Supercapacitor Electrodes from Single-Walled Carbon Nanohorn/Nanotube Composite. ACS Nano 2011, 5, 811-819, 785 https://doi.org/10.1021/nn1017457. 786 37. Yalovega, G.E.; Brzhezinskaya, M.; Dmitriev, V.O.; Shmatko, V.A.; Ershov, I.V.; Ulyankina, A.A.; Chernysheva, D.V.; Smirnova, 787 N.V. Interfacial Interaction in MeOx/MWNTs (Me-Cu, Ni) Nanostructures as Efficient Electrode Materials for High-Performance 788 Supercapacitors. Nanomaterials 2024, 14, 947, https://doi.org/10.3390/nano14110947. 789 https://doi.org/10.1038/s41524-021-00629-yhttps://doi.org/10.1021/acsnano.3c07587https://doi.org/10.1038/s41563-019-0402-8https://doi.org/10.1016/j.actamat.2019.10.011https://doi.org/10.1002/adom.202203104https://doi.org/10.1038/s41596-021-00620-3https://doi.org/10.1016/j.foodcont.2023.110272https://doi.org/10.1016/j.vibspec.2024.103715https://doi.org/10.1088/2632-2153/ad5074https://doi.org/10.1088/2632-2153/ad5074https://doi.org/10.1088/2632-2153/ab5da6https://doi.org/10.3390/signals5020010https://doi.org/10.1016/j.aca.2023.341433https://doi.org/10.1016/j.carbon.2022.02.019https://doi.org/10.1016/j.jcat.2017.03.009https://doi.org/10.1038/s41524-021-00603-8https://doi.org/10.1021/acsnano.8b04726https://doi.org/10.1126/science.abi8884https://doi.org/10.1126/science.1156588https://doi.org/10.1002/adfm.200901960https://doi.org/10.1063/1.1349870https://doi.org/10.1088/0957-4484/19/41/415703https://doi.org/10.1021/nn1017457https://doi.org/10.3390/nano14110947Nanomaterials 2024, 14, x FOR PEER REVIEW 19 of 22   38. Züttel, A.; Sudan, P.; Mauron, P.; Kiyobayashi, T.; Emmenegger, C.; Schlapbach, L. Hydrogen storage in carbon nanostructures. 790 Int. J. Hydrogen Energy 2002, 27, 203-212, https://doi.org/10.1016/S0360-3199(01)00108-2. 791 39. Brzhezinskaya, M.; Belenkov, E.A.; Greshnyakov, V.A.; Yalovega, G.E.; Bashkin, I.O. New aspects in the study of carbon- 792 hydrogen interaction in hydrogenated carbon nanotubes for energy storage applications. J. Alloys Compd. 2019, 792, 713-720, 793 https://doi.org/10.1016/j.jallcom.2019.04.107. 794 40. Landi, B.J.; Ganter, M.J.; Cress, C.D.; DiLeo, R.A.; Raffaelle, R.P. Carbon nanotubes for lithium ion batteries. Energy & 795 Environmental Science 2009, 2, 638-654, https://doi.org/10.1039/b904116h. 796 41. Lahiri, I.; Oh, S.W.; Hwang, J.Y.; Cho, S.; Sun, Y.K.; Banerjee, R.; Choi, W. High Capacity and Excellent Stability of Lithium Ion 797 Battery Anode Using Interface-Controlled Binder-Free Multiwall Carbon Nanotubes Grown on Copper. ACS Nano 2010, 4, 3440- 798 3446, https://doi.org/10.1021/Nn100400r. 799 42. Chen, G.H.; Dodson, B.; Hedges, D.M.; Steffensen, S.C.; Harb, J.N.; Puleo, C.; Galligan, C.; Ashe, J.; Vanfleet, R.R.; Davis, R.C. 800 Fabrication of High Aspect Ratio Millimeter-Tall Free-Standing Carbon Nanotube-Based Microelectrode Arrays. ACS Biomater. 801 Sci. Eng. 2018, 4, 1900-1907, https://doi.org/10.1021/acsbiomaterials.8b00038. 802 43. Yang, N.; Chen, X.P.; Ren, T.L.; Zhang, P.; Yang, D.G. Carbon nanotube based biosensors. Sensor Actuat B-Chem 2015, 207, 690- 803 715, https://doi.org/10.1016/j.snb.2014.10.040. 804 44. Zhang, W.X.; Zhang, Z.Z.; Zhang, Y.G. The application of carbon nanotubes in target drug delivery systems for cancer therapies. 805 Nanoscale Res. Lett. 2011, 6, 555, https://doi.org/10.1186/1556-276x-6-555. 806 45. Chen, G.H.; Sundaram, R.; Sekiguchi, A.; Hata, K.; Futaba, D.N. Through-Silicon-Via Interposers with Cu-Level Electrical 807 Conductivity and Si-Level Thermal Expansion Based on Carbon Nanotube-Cu Composites for Microelectronic Packaging 808 Applications. ACS Appl. Nano Mater. 2021, 4, 869-876, https://doi.org/10.1021/acsanm.0c03278. 809 46. Xu, M.; Du, F.; Ganguli, S.; Roy, A.; Dai, L.M. Carbon nanotube dry adhesives with temperature-enhanced adhesion over a large 810 temperature range. Nat. Commun. 2016, 7, 13450, https://doi.org/10.1038/ncomms13450. 811 47. Qu, L.T.; Dai, L.M.; Stone, M.; Xia, Z.H.; Wang, Z.L. Carbon nanotube arrays with strong shear binding-on and easy normal 812 lifting-off. Science 2008, 322, 238-242, https://doi.org/10.1126/science.1159503. 813 48. Chen, G.H.; Futaba, D.N.; Kimura, H.; Sakurai, S.; Yumura, M.; Hata, K. Absence of an Ideal Single-Walled Carbon Nanotube 814 Forest Structure for Thermal and Electrical Conductivities. ACS Nano 2013, 7, 10218-10224, https://doi.org/10.1021/nn404504f. 815 49. Shibuya, A.; Chen, G.H.; Miyoshi, A.; Futaba, D.N. Improving the synthetic efficiency of single-wall carbon nanotube forests 816 using a gas-analysis-designed mixed carbon feedstock. Carbon 2020, 170, 59-65, https://doi.org/10.1016/j.carbon.2020.08.001. 817 50. Chen, G.H.; Tsuji, T.; Yamada, M.; He, J.; Shimizu, Y.; Sakakita, H.; Hata, K.; Futaba, D.N.; Sakurai, S. Multi-step chemical vapor 818 synthesis reactor based on a microplasma for structure-controlled synthesis of single-walled carbon nanotubes. Chem. Eng. J. 819 2022, 444, 136634, https://doi.org/10.1016/j.cej.2022.136634. 820 51. Sharma, P.; Pavelyev, V.; Kumar, S.; Mishra, P.; Islam, S.S.; Tripathi, N. Analysis on the synthesis of vertically aligned carbon 821 nanotubes: growth mechanism and techniques. J Mater Sci-Mater El 2020, 31, 4399-4443, https://doi.org/10.1007/s10854-020- 822 03021-6. 823 52. Chen, G.H.; Davis, R.C.; Futaba, D.N.; Sakurai, S.; Kobashi, K.; Yumura, M.; Hata, K. A sweet spot for highly efficient growth 824 of vertically aligned single-walled carbon nanotube forests enabling their unique structures and properties. Nanoscale 2016, 8, 825 162-171, https://doi.org/10.1039/C5NR05537G. 826 53. Sugime, H.; Sato, T.; Nakagawa, R.; Hayashi, T.; Inoue, Y.; Noda, S. Ultra-long carbon nanotube forest via in situ supplements 827 of iron and aluminum vapor sources. Carbon 2021, 172, 772-780, https://doi.org/10.1016/j.carbon.2020.10.066. 828 54. Lee, S.H.; Park, J.; Park, J.H.; Lee, D.M.; Lee, A.; Moon, S.Y.; Lee, S.Y.; Jeong, H.S.; Kim, S.M. Deep-injection floating-catalyst 829 chemical vapor deposition to continuously synthesize carbon nanotubes with high aspect ratio and high crystallinity. Carbon 830 2021, 173, 901-909, https://doi.org/10.1016/j.carbon.2020.11.065. 831 55. Chen, G.H.; Seki, Y.; Kimura, H.; Sakurai, S.; Yumura, M.; Hata, K.; Futaba, D.N. Diameter control of single-walled carbon 832 nanotube forests from 1.3-3.0 nm by arc plasma deposition. Sci. Rep. 2014, 4, 3804, https://doi.org/10.1038/srep03804. 833 56. Zhao, B.; Futaba, D.N.; Yasuda, S.; Akoshima, M.; Yamada, T.; Hata, K. Exploring Advantages of Diverse Carbon Nanotube 834 Forests With Tailored Structures Synthesized by Supergrowth from Engineered Catalysts. ACS Nano 2009, 3, 108-114, 835 https://doi.org/10.1021/Nn800648a. 836 57. Hiraoka, T.; Izadi-Najafabadi, A.; Yamada, T.; Futaba, D.N.; Yasuda, S.; Tanaike, O.; Hatori, H.; Yumura, M.; Iijima, S.; Hata, K. 837 Compact and Light Supercapacitor Electrodes from a Surface-Only Solid by Opened Carbon Nanotubes with 2200 m2 g-1 Surface 838 Area. Adv. Funct. Mater. 2010, 20, 422-428, https://doi.org/10.1002/adfm.200901927. 839 58. Xu, M.; Futaba, D.N.; Yumura, M.; Hata, K. Alignment Control of Carbon Nanotube Forest from Random to Nearly Perfectly 840 Aligned by Utilizing the Crowding Effect. ACS Nano 2012, 6, 5837-5844, https://doi.org/10.1021/nn300142j. 841 59. Ci, L.; Vajtai, R.; Ajayan, P.M. Vertically aligned large-diameter double-walled carbon nanotube arrays having ultralow density. 842 J. Phys. Chem. C 2007, 111, 9077-9080, https://doi.org/10.1021/jp072123c. 843 60. Tang, B.J.; Lu, Y.H.; Zhou, J.D.; Chouhan, T.; Wang, H.; Golani, P.; Xu, M.Z.; Xu, Q.; Guan, C.T.; Liu, Z. Machine learning-guided 844 synthesis of advanced inorganic materials. Mater Today 2020, 41, 72-80, https://doi.org/10.1016/j.mattod.2020.06.010. 845 https://doi.org/10.1016/S0360-3199(01)00108-2https://doi.org/10.1016/j.jallcom.2019.04.107https://doi.org/10.1039/b904116hhttps://doi.org/10.1021/Nn100400rhttps://doi.org/10.1021/acsbiomaterials.8b00038https://doi.org/10.1016/j.snb.2014.10.040https://doi.org/10.1186/1556-276x-6-555https://doi.org/10.1021/acsanm.0c03278https://doi.org/10.1038/ncomms13450https://doi.org/10.1126/science.1159503https://doi.org/10.1021/nn404504fhttps://doi.org/10.1016/j.carbon.2020.08.001https://doi.org/10.1016/j.cej.2022.136634https://doi.org/10.1007/s10854-020-03021-6https://doi.org/10.1007/s10854-020-03021-6https://doi.org/10.1039/C5NR05537Ghttps://doi.org/10.1016/j.carbon.2020.10.066https://doi.org/10.1016/j.carbon.2020.11.065https://doi.org/10.1038/srep03804https://doi.org/10.1021/Nn800648ahttps://doi.org/10.1002/adfm.200901927https://doi.org/10.1021/nn300142jhttps://doi.org/10.1021/jp072123chttps://doi.org/10.1016/j.mattod.2020.06.010Nanomaterials 2024, 14, x FOR PEER REVIEW 20 of 22   61. Yuan, R.H.; Liu, Z.; Balachandran, P.V.; Xue, D.Q.; Zhou, Y.M.; Ding, X.D.; Sun, J.; Xue, D.Z.; Lookman, T. Accelerated Discovery 846 of Large Electrostrains in BaTiO3-Based Piezoelectrics Using Active Learning. Adv. Mater. 2018, 30, 1702884, 847 https://doi.org/10.1002/adma.201702884. 848 62. Lu, S.H.; Zhou, Q.H.; Ouyang, Y.X.; Guo, Y.L.; Li, Q.; Wang, J.L. Accelerated discovery of stable lead-free hybrid organic- 849 inorganic perovskites via machine learning. Nat. Commun. 2018, 9, 3405, https://doi.org/10.1038/s41467-018-05761-w. 850 63. Xue, D.Z.; Balachandran, P.V.; Hogden, J.; Theiler, J.; Xue, D.Q.; Lookman, T. Accelerated search for materials with targeted 851 properties by adaptive design. Nat. Commun. 2016, 7, 11241, https://doi.org/10.1038/ncomms11241. 852 64. Kaushal, A.; Alexander, R.; Rao, P.T.; Prakash, J.; Dasgupta, K. Artificial neural network, Pareto optimization, and Taguchi 853 analysis for the synthesis of single-walled carbon nanotubes. Carbon Trends 2021, 2, 100016, 854 https://doi.org/10.1016/j.cartre.2020.100016. 855 65. Shields, B.J.; Stevens, J.; Li, J.; Parasram, M.; Damani, F.; Alvarado, J.I.M.; Janey, J.M.; Adams, R.P.; Doyle, A.G. Bayesian reaction 856 optimization as a tool for chemical synthesis. Nature 2021, 590, 89-96, https://doi.org/10.1038/s41586-021-03213-y. 857 66. Chu, D.M.; Ji, Z.C.; Zhang, X.J.; Zhao, X.Y.; He, Y.; Bai, W.J. Machine learning for the regulation strategy and mechanism of the 858 integrated growth of carbon nanotube arrays. New J. Chem. 2023, 47, 21883-21896, https://doi.org/10.1039/d3nj04124g. 859 67. Bulmer, J.S.; Sloan, A.W.N.; Glerum, M.; Carpena-Nunez, J.; Waelder, R.; Humes, J.; Boies, A.M.; Pasquali, M.; Rao, R.; 860 Maruyama, B. Forecasting carbon nanotube diameter in floating catalyst chemical vapor deposition. Carbon 2023, 201, 719-733, 861 https://doi.org/10.1016/j.carbon.2022.08.001. 862 68. Ji, Z.H.; Zhang, L.L.; Tang, D.M.; Chen, C.M.; Nordling, T.E.M.; Zhang, Z.D.; Ren, C.L.; Da, B.; Li, X.; Guo, S.Y.; Liu, C.; Cheng, 863 H.M. High-throughput screening and machine learning for the efficient growth of high-quality single-wall carbon nanotubes. 864 Nano Res. 2021, 14, 4610-4615, https://doi.org/10.1007/s12274-021-3387-y. 865 69. Krasnikov, D.V.; Khabushev, E.M.; Gaev, A.; Bogdanova, A.R.; Iakovlev, V.Y.; Lantsberg, A.; Kallio, T.; Nasibulin, A.G. Machine 866 learning methods for aerosol synthesis of single-walled carbon nanotubes. Carbon 2023, 202, 76-82, 867 https://doi.org/10.1016/j.carbon.2022.10.044. 868 70. Shin, S.; Song, H.; Shin, Y.S.; Lee, J.; Seo, T.H. Bayesian Optimization of Wet-Impregnated Co-Mo/Al2O3 Catalyst for Maximizing 869 the Yield of Carbon Nanotube Synthesis. Nanomaterials 2024, 14, 75, https://doi.org/10.3390/nano14010075. 870 71. Ding, R.; Chen, J.; Chen, Y.; Liu, J.; Bando, Y.; Wang, X. Unlocking the potential: machine learning applications in electrocatalyst 871 design for electrochemical hydrogen energy transformation. Chem. Soc. Rev. 2024, https://doi.org/10.1039/D4CS00844H. 872 72. Alwosheel, A.; van Cranenburgh, S.; Chorus, C.G. Is your dataset big enough? Sample size requirements when using artificial 873 neural networks for discrete choice analysis. Journal of Choice Modelling 2018, 28, 167-182, 874 https://doi.org/10.1016/j.jocm.2018.07.002. 875 73. Nikolaev, P.; Hooper, D.; Perea-Lopez, N.; Terrones, M.; Maruyama, B. Discovery of Wall-Selective Carbon Nanotube Growth 876 Conditions via Automated Experimentation. ACS Nano 2014, 8, 10214-10222, https://doi.org/10.1021/nn503347a. 877 74. Nikolaev, P.; Hooper, D.; Webber, F.; Rao, R.; Decker, K.; Krein, M.; Poleski, J.; Barto, R.; Maruyama, B. Autonomy in materials 878 research: a case study in carbon nanotube growth. npj Comput Mater 2016, 2, 16031, 879 https://doi.org/10.1038/npjcompumats.2016.31. 880 75. Chang, J.; Nikolaev, P.; Carpena-Nunez, J.; Rao, R.; Decker, K.; Islam, A.E.; Kim, J.; Pitt, M.A.; Myung, J.I.; Maruyama, B. Efficient 881 Closed-loop Maximization of Carbon Nanotube Growth Rate using Bayesian Optimization. Sci. Rep. 2020, 10, 9040, 882 https://doi.org/10.1038/s41598-020-64397-3. 883 76. Li, Y.; Wang, S.; Lv, Z.; Wang, Z.; Zhao, Y.; Xie, Y.; Xu, Y.; Qian, L.; Yang, Y.; Zhao, Z. Transforming the Synthesis of Carbon 884 Nanotubes with Machine Learning Models and Automation. arXiv preprint arXiv:2404.01006 2024, 885 https://doi.org/10.48550/arXiv.2404.01006. 886 77. Iijima, S. Helical Microtubules of Graphitic Carbon. Nature 1991, 354, 56-58, https://doi.org/10.1038/354056a0. 887 78. Qin, L.C. Electron diffraction from carbon nanotubes. Rep Prog Phys 2006, 69, 2761-2821, https://doi.org/10.1088/0034- 888 4885/69/10/R02. 889 79. Jorio, A.; Saito, R.; Hafner, J.H.; Lieber, C.M.; Hunter, M.; McClure, T.; Dresselhaus, G.; Dresselhaus, M.S. Structural (n, m) 890 determination of isolated single-wall carbon nanotubes by resonant Raman scattering. Phys. Rev. Lett. 2001, 86, 1118-1121, 891 https://doi.org/10.1103/PhysRevLett.86.1118. 892 80. Dresselhaus, M.S.; Dresselhaus, G.; Saito, R.; Jorio, A. Raman spectroscopy of carbon nanotubes. Phys. Rep. 2005, 409, 47-99, 893 https://doi.org/10.1016/j.physrep.2004.10.006. 894 81. Belianinov, A.; Vasudevan, R.; Strelcov, E.; Steed, C.; Yang, S.M.; Tselev, A.; Jesse, S.; Biegalski, M.; Shipman, G.; Symons, C.; 895 Borisevich, A.; Archibald, R.; Kalinin, S. Big data and deep data in scanning and electron microscopies: deriving functionality 896 from multidimensional data sets. Advanced Structural and Chemical Imaging 2015, 1, 6, https://doi.org/10.1186/s40679-015-0006-6. 897 82. Aguiar, J.A.; Gong, M.L.; Unocic, R.R.; Tasdizen, T.; Miller, B.D. Decoding crystallography from high-resolution electron 898 imaging and diffraction datasets with deep learning. Sci Adv 2019, 5, eaaw1949, https://doi.org/10.1126/sciadv.aaw1949. 899 83. Cui, A.Y.; Jiang, K.; Jiang, M.H.; Shang, L.Y.; Zhu, L.Q.; Hu, Z.G.; Xu, G.S.; Chu, J.H. Decoding Phases of Matter by Machine- 900 Learning Raman Spectroscopy. Phys Rev Appl 2019, 12, 054049, https://doi.org/10.1103/PhysRevApplied.12.054049. 901 https://doi.org/10.1002/adma.201702884https://doi.org/10.1038/s41467-018-05761-whttps://doi.org/10.1038/ncomms11241https://doi.org/10.1016/j.cartre.2020.100016https://doi.org/10.1038/s41586-021-03213-yhttps://doi.org/10.1039/d3nj04124ghttps://doi.org/10.1016/j.carbon.2022.08.001https://doi.org/10.1007/s12274-021-3387-yhttps://doi.org/10.1016/j.carbon.2022.10.044https://doi.org/10.3390/nano14010075https://doi.org/10.1039/D4CS00844Hhttps://doi.org/10.1016/j.jocm.2018.07.002https://doi.org/10.1021/nn503347ahttps://doi.org/10.1038/npjcompumats.2016.31https://doi.org/10.1038/s41598-020-64397-3https://doi.org/10.48550/arXiv.2404.01006https://doi.org/10.1038/354056a0https://doi.org/10.1088/0034-4885/69/10/R02https://doi.org/10.1088/0034-4885/69/10/R02https://doi.org/10.1103/PhysRevLett.86.1118https://doi.org/10.1016/j.physrep.2004.10.006https://doi.org/10.1186/s40679-015-0006-6https://doi.org/10.1126/sciadv.aaw1949https://doi.org/10.1103/PhysRevApplied.12.054049Nanomaterials 2024, 14, x FOR PEER REVIEW 21 of 22   84. Trejo, O.; Dadlani, A.L.; De La Paz, F.; Acharya, S.; Kravec, R.; Nordlund, D.; Sarangi, R.; Prinz, F.B.; Torgersen, J.; Dasgupta, 902 N.P. Elucidating the Evolving Atomic Structure in Atomic Layer Deposition Reactions with in Situ XANES and Machine 903 Learning. Chem. Mater. 2019, 31, 8937-8947, https://doi.org/10.1021/acs.chemmater.9b03025. 904 85. Kaufmann, K.; Zhu, C.Y.; Rosengarten, A.S.; Maryanovsky, D.; Harrington, T.J.; Marin, E.; Vecchio, K.S. Crystal symmetry 905 determination in electron diffraction using machine learning. Science 2020, 367, 564-568, https://doi.org/10.1126/science.aay3062. 906 86. Spurgeon, S.R.; Ophus, C.; Jones, L.; Petford-Long, A.; Kalinin, S.V.; Olszta, M.J.; Dunin-Borkowski, R.E.; Salmon, N.; Hattar, 907 K.; Yang, W.C.D.; Sharma, R.; Du, Y.G.; Chiaramonti, A.; Zheng, H.M.; Buck, E.C.; Kovarik, L.; Penn, R.L.; Li, D.S.; Zhang, X.; 908 Murayama, M.; Taheri, M.L. Towards data-driven next-generation transmission electron microscopy. Nat. Mater. 2021, 20, 274- 909 279, https://doi.org/10.1038/s41563-020-00833-z. 910 87. Uchida, T.; Tazawa, M.; Sakai, H.; Yamazaki, A.; Kobayashi, Y. Radial breathing modes of single-walled carbon nanotubes in 911 resonance Raman spectra at high temperature and their chiral index assignment. Appl. Surf. Sci. 2008, 254, 7591-7595, 912 https://doi.org/10.1016/j.apsusc.2008.01.107. 913 88. Zhang, J.; Perrin, M.L.; Barba, L.; Overbeck, J.; Jung, S.; Grassy, B.; Agal, A.; Muff, R.; Brönnimann, R.; Haluska, M.; Roman, C.; 914 Hierold, C.; Jaggi, M.; Calame, M. High-speed identification of suspended carbon nanotubes using Raman spectroscopy and 915 deep learning. Microsyst Nanoeng 2022, 8, 19, https://doi.org/10.1038/s41378-022-00350-w. 916 89. Kajendirarajah, U.; Avilés, M.O.; Lagugné-Labarthet, F. Deciphering tip-enhanced Raman imaging of carbon nanotubes with 917 deep learning neural networks. Phys. Chem. Chem. Phys. 2020, 22, 17857-17866, https://doi.org/10.1039/d0cp02950e. 918 90. Luo, Q.X.; Holm, E.A.; Wang, C. A transfer learning approach for improved classification of carbon nanomaterials from TEM 919 images. Nanoscale Adv 2021, 3, 206-213, https://doi.org/10.1039/d0na00634c. 920 91. Kalinin, S.V.; Ziatdinov, M.; Spurgeon, S.R.; Ophus, C.; Stach, E.A.; Susi, T.; Agar, J.; Randall, J. Deep learning for electron and 921 scanning probe microscopy: From materials design to atomic fabrication. MRS Bull. 2022, 47, 931-939, 922 https://doi.org/10.1557/s43577-022-00413-3. 923 92. Govind, K.; Oliveros, D.; Dlouhy, A.; Legros, M.; Sandfeld, S. Deep learning of crystalline defects from TEM images: a solution 924 for the problem of 'never enough training data'. Mach Learn-Sci Techn 2024, 5, 015006, https://doi.org/10.1088/2632-2153/ad1a4e. 925 93. Förster, G.D.; Castan, A.; Loiseau, A.; Nelayah, J.; Alloyeau, D.; Fossard, F.; Bichara, C.; Amara, H. A deep learning approach 926 for determining the chiral indices of carbon nanotubes from high-resolution transmission electron microscopy images. Carbon 927 2020, 169, 465-474, https://doi.org/10.1016/j.carbon.2020.06.086. 928 94. Lee, B.; Yoon, S.; Lee, J.W.; Kim, Y.; Chang, J.; Yun, J.; Ro, J.C.; Lee, J.S.; Lee, J.H. Statistical Characterization of the Morphologies 929 of Nanoparticles through Machine Learning Based Electron Microscopy Image Analysis. ACS Nano 2020, 14, 17125-17133, 930 https://doi.org/10.1021/acsnano.0c06809. 931 95. Chen, D.; Gao, J.; He, D.; He, J.; Li, Y.; Zhang, M.; Li, W.; Chen, X.; He, X.; Fu, T. Plasmonic Bridge Sensor Enabled by Carbon 932 Nanotubes and Au-Ag Nano-Rambutan for Multifunctional Detection of Biomechanics and Bio/Chemical Molecules. ACS Appl. 933 Mater. Interfaces 2023, 15, 8783-8793, https://doi.org/10.1021/acsami.2c22634. 934 96. Kelich, P.; Adams, J.; Jeong, S.; Navarro, N.; Landry, M.P.; Vukovic, L. Predicting Serotonin Detection with DNA-Carbon 935 Nanotube Sensors across Multiple Spectral Wavelengths. J. Chem. Inf. Model. 2024, 64, 3992-4001, 936 https://doi.org/10.1021/acs.jcim.4c00021. 937 97. Tadokoro, Y.; Funayama, K.; Kawano, K.; Miura, A.; Hirotani, J.; Ohno, Y.; Tanaka, H. Artificial-intelligence-assisted mass 938 fabrication of nanocantilevers from randomly positioned single carbon nanotubes. Microsyst Nanoeng 2023, 9, 32, 939 https://doi.org/10.1038/s41378-023-00507-1. 940 98. Aliyana, A.K.; Kumar, S.K.N.; Marimuthu, P.; Baburaj, A.; Adetunji, M.; Frederick, T.; Sekhar, P.; Fernandez, R.E. Machine 941 learning-assisted ammonium detection using zinc oxide/multi-walled carbon nanotube composite based impedance sensors. 942 Sci. Rep. 2021, 11, 24321, https://doi.org/10.1038/s41598-021-03674-1. 943 99. Bian, L.; Wang, Z.H.; White, D.L.; Star, A. Machine learning-assisted calibration of Hg2+ sensors based on carbon nanotube field- 944 effect transistors. Biosens. Bioelectron. 2021, 180, 113085, https://doi.org/10.1016/j.bios.2021.113085. 945 100. Fan, G.; Low, K.L. Physics-Integrated Machine Learning for Efficient Design and Optimization of a Nanoscale Carbon Nanotube 946 Field-Effect Transistor. ECS Journal of Solid State Science and Technology 2023, 12, 091005, https://doi.org/10.1149/2162-8777/acfb38. 947 101. Kelich, P.; Jeong, S.; Navarro, N.; Adams, J.; Sun, X.Q.; Zhao, H.H.; Landry, M.P.; Vukovic, L. Discovery of DNA-Carbon 948 Nanotube Sensors for Serotonin with Machine Learning and Near-infrared Fluorescence Spectroscopy. ACS Nano 2022, 16, 736- 949 745, https://doi.org/10.1021/acsnano.1c08271. 950 102. Yu, T.Y.; Mo, X.D.; Chen, M.J.; Yao, C.F. Machine-learning-assisted microstructure-property linkages of carbon nanotube- 951 reinforced aluminum matrix nanocomposites produced by laser powder bed fusion. Nanotechnol Rev 2021, 10, 1410-1424, 952 https://doi.org/10.1515/ntrev-2021-0093. 953 103. Ranaiefar, M.; Singh, M.; Halbig, M.C. Additively Manufactured Carbon-Reinforced ABS Honeycomb Composite Structures 954 and Property Prediction by Machine Learning. Molecules 2024, 29, 2736, https://doi.org/10.3390/molecules29122736. 955 104. Jalal, M.; Moradi-Dastjerdi, R.; Bidram, M. Big data in nanocomposites: ONN approach and mesh-free method for functionally 956 graded carbon nanotube-reinforced composites. J Comput Des Eng 2019, 6, 209-223, https://doi.org/10.1016/j.jcde.2018.05.003. 957 https://doi.org/10.1021/acs.chemmater.9b03025https://doi.org/10.1126/science.aay3062https://doi.org/10.1038/s41563-020-00833-zhttps://doi.org/10.1016/j.apsusc.2008.01.107https://doi.org/10.1038/s41378-022-00350-whttps://doi.org/10.1039/d0cp02950ehttps://doi.org/10.1039/d0na00634chttps://doi.org/10.1557/s43577-022-00413-3https://doi.org/10.1088/2632-2153/ad1a4ehttps://doi.org/10.1016/j.carbon.2020.06.086https://doi.org/10.1021/acsnano.0c06809https://doi.org/10.1021/acsami.2c22634https://doi.org/10.1021/acs.jcim.4c00021https://doi.org/10.1038/s41378-023-00507-1https://doi.org/10.1038/s41598-021-03674-1https://doi.org/10.1016/j.bios.2021.113085https://doi.org/10.1149/2162-8777/acfb38https://doi.org/10.1021/acsnano.1c08271https://doi.org/10.1515/ntrev-2021-0093https://doi.org/10.3390/molecules29122736https://doi.org/10.1016/j.jcde.2018.05.003Nanomaterials 2024, 14, x FOR PEER REVIEW 22 of 22   105. Huang, J.S.; Liew, J.X.; Liew, K.M. Data-driven machine learning approach for exploring and assessing mechanical properties 958 of carbon nanotube-reinforced cement composites. Compos Struct 2021, 267, 113917, 959 https://doi.org/10.1016/j.compstruct.2021.113917. 960 106. Bagherzadeh, F.; Shafighfard, T. Ensemble Machine Learning approach for evaluating the material characterization of carbon 961 nanotube-reinforced cementitious composites. Case Stud Constr Mat 2022, 17, e01537, https://doi.org/10.1016/j.cscm.2022.e01537. 962 107. Okasha, N.M.; Mirrashid, M.; Naderpour, H.; Ciftcioglu, A.O.; Meddage, D.P.P.; Ezami, N. Machine learning approach to 963 predict the mechanical properties of cementitious materials containing carbon nanotubes. Developments in the Built Environment 964 2024, 19, 100494, https://doi.org/10.1016/j.dibe.2024.100494. 965 108. Le, T.-T. Prediction of tensile strength of polymer carbon nanotube composites using practical machine learning method. J. 966 Compos Mater. 2021, 55, 787-811, https://doi.org/10.1177/0021998320953540. 967 109. Liu, B.K.; Vu-Bac, N.; Zhuang, X.Y.; Fu, X.L.; Rabczuk, T. Stochastic integrated machine learning based multiscale approach for 968 the prediction of the thermal conductivity in carbon nanotube reinforced polymeric composites. Compos. Sci. Technol. 2022, 224, 969 109425, https://doi.org/10.1016/j.compscitech.2022.109425. 970 110. Liu, B.K.; Vu-Bac, N.; Zhuang, X.Y.; Fu, X.L.; Rabczuk, T. Stochastic full-range multiscale modeling of thermal conductivity of 971 Polymeric carbon nanotubes composites: A machine learning approach. Compos Struct 2022, 289, 115393, 972 https://doi.org/10.1016/j.compstruct.2022.115393. 973 111. Champa-Bujaico, E.; Díez-Pascual, A.M.; Garcia-Diaz, P.; Sessini, V.; Mosquera, M.E.G. Machine learning algorithms to optimize 974 the properties of bio-based poly (butylene succinate-co- butylene adipate) nanocomposites with carbon nanotubes. Ind Crop 975 Prod 2024, 219, 119018, https://doi.org/10.1016/j.indcrop.2024.119018. 976 112. Zarei, A.; Farazin, A. Machine Learning Insights into the Influence of Carbon Nanotube Dimensions on Nanocomposite 977 Properties: A Comprehensive Exploration. Journal of Computational Applied Mechanics 2024, 55, 462-472, 978 https://doi.org/10.22059/jcamech.2024.376321.1086. 979 113. Chen, G.H.; Dodson, B.; Johnson, F.; Hancu, I.; Fiveland, E.; Zhang, W.; Galligan, C.; Puleo, C.; Davis, R.C.; Ashe, J.; Vanfleet, 980 R.R. Tissue-susceptibility matched carbon nanotube electrodes for magnetic resonance imaging. J. Magn. Reson. 2018, 295, 72- 981 79, https://doi.org/10.1016/j.jmr.2018.08.003. 982 114. Ouassil, N.; Pinals, R.L.; Del Bonis-O'Donnell, J.T.; Wang, J.W.; Landry, M.P. Supervised learning model predicts protein 983 adsorption to carbon nanotubes. Sci Adv 2022, 8, eabm0898, https://doi.org/10.1126/sciadv.abm0898. 984 115. Lee, D.; Lee, J.; Kim, W.; Suh, Y.; Park, J.; Kim, S.; Kim, Y.; Kwon, S.; Jeong, S. Systematic Selection of High-Affinity ssDNA 985 Sequences to Carbon Nanotubes. Adv Sci 2024, 11, 2308915, https://doi.org/10.1002/advs.202308915. 986 116. Lin, Z.W.; Yang, Y.; Jagota, A.; Zheng, M. Machine Learning-Guided Systematic Search of DNA Sequences for Sorting Carbon 987 Nanotubes. ACS Nano 2022, 16, 4705-4713, https://doi.org/10.1021/acsnano.1c11448. 988 117. Kotzabasaki, M.; Sotiropoulos, I.; Charitidis, C.; Sarimveis, H. Machine learning methods for multi-walled carbon nanotubes 989 (MWCNT) genotoxicity prediction. Nanoscale Adv 2021, 3, 3167-3176, https://doi.org/10.1039/d0na00600a. 990 118. Gong, X.; Renegar, N.; Levi, R.; Strano, M.S. Machine learning for the discovery of molecular recognition based on single-walled 991 carbon nanotube corona-phases. npj Comput Mater 2022, 8, 135, https://doi.org/10.1038/s41524-022-00795-7. 992  993  994 Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual 995 author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any 996 injury to people or property resulting from any ideas, methods, instructions or products referred to in the content. 997  998 https://doi.org/10.1016/j.compstruct.2021.113917https://doi.org/10.1016/j.cscm.2022.e01537https://doi.org/10.1016/j.dibe.2024.100494https://doi.org/10.1177/0021998320953540https://doi.org/10.1016/j.compscitech.2022.109425https://doi.org/10.1016/j.compstruct.2022.115393https://doi.org/10.1016/j.indcrop.2024.119018https://doi.org/10.22059/jcamech.2024.376321.1086https://doi.org/10.1016/j.jmr.2018.08.003https://doi.org/10.1126/sciadv.abm0898https://doi.org/10.1002/advs.202308915https://doi.org/10.1021/acsnano.1c11448https://doi.org/10.1039/d0na00600ahttps://doi.org/10.1038/s41524-022-00795-7 1. Introduction 1.1. Machine Learning 1.2. Carbon Nanotubes 2. Machine Learning-Assisted CNT Synthesis 2.1. Data-Driven Optimization of CNT Synthesis 2.2. Autonomous CNT Synthesis 3. Machine Learning-Assisted CNT Characterization 3.1. Machine Learning for Rapid Identification of Raman Spectroscopy 3.2. Machine Learning for Automated Chirality Determination from TEM Images 4. Machine Learning-Assisted CNT Applications 4.1. CNT Applications in Electronics and Sensors 4.2. CNT Applications in Composite Materials 4.3. CNT Applications in the Biomedical Field 5. Conclusion and Perspective References