Pre-trained Neural Network Research Articles

Transfer learning for metamaterial design and simulation

Abstract We demonstrate transfer learning as a tool to improve the efficacy of training deep learning models based on residual neural networks (ResNets). Specifically, we examine its use for study of multi-scale electrically large metasurface arrays under open boundary conditions in electromagnetic metamaterials. Our aim is to assess the efficiency of transfer learning across a range of problem domains that vary in their resemblance to the original base problem for which the ResNet model was initially trained. We use a quasi-analytical discrete dipole approximation (DDA) method to simulate electrically large metasurface arrays to obtain ground truth data for training and testing of our deep neural network. Our approach can save significant time for examining novel metasurface designs by harnessing the power of transfer learning, as it effectively mitigates the pervasive data bottleneck issue commonly encountered in deep learning. We demonstrate that for the best case when the transfer task is sufficiently similar to the target task, a new task can be effectively trained using only a few data points yet still achieve a test mean absolute relative error of 3 % with a pre-trained neural network, realizing data reduction by a factor of 1000.

Abstract 2319: Deep learning-enabled dynamic infiltration and response to NK therapies in solid tumor organoids

Abstract Patient-derived organoid co-culture models have become a state of the art platform to evaluate the effectiveness of immunotherapeutic agents in preclinical studies. However, preclinical development as well as quantification of therapeutic potency and resistance require differential labeling of cells with fluorescent dyes or transgenic fluorescent proteins, which may dysregulate the effector cell function or produce undesired cytotoxicity. Deep learning-based image analysis allows the use of brightfield images to co-localize tumor organoids (TOs) with effector cells, and measure TO-specific responses to novel candidate immunotherapies in a label-free manner; this can accelerate their evaluation as therapeutic candidates in cancer patients and shed light on tumor-immune interaction mechanisms. Here we record multi-day time-lapse confocal microscopy images of TOs co-cultured with NK cells at increasing concentrations of target to effector cells. Two pre-trained U-Net convolutional neural networks are used to separately segment tumor and immune cells from the brightfield channel, and quantify immune cell infiltration over time. TO segmentation masks are registered with a vital dye (caspase 3/7) channel to selectively quantify TO cell death and correlate it with estimated immune cell infiltration across a large cohort of patient-derived TO models (including breast, colorectal, endometrial, gastric, head and neck, liver, lung and pancreatic cancer), at increasing concentrations of 6 different NK cell types (identified as A through F). We find that the estimated density of infiltrating immune cells is highly correlated with TO death, as quantified by fluorescence intensity of caspase over time (median Pearson’s r=0.89, P<0.001). Higher ratios of effector to target cells lead to higher activation of NK cells as measured by infiltrating cell density. Differential infiltration dynamics are observed across TOs and immune cell lines, as peak infiltration density is affected by co-culture time, ratio of target to effector cells, TO line, and NK cell type. These findings highlight that the paired segmentation models are able to measure varying degrees of infiltration across different types of immune cells and TO models. The approach described here is highly scalable and only requires capturing brightfield images of the assay, therefore eliminating the need to label cells and avoiding phototoxic effects or alteration of the immune cells activity. Time series measurements enable quantification of patterns of immune cell activation, including infiltration, migration and co-localization dynamics, which provide insights into the pharmacokinetics and mechanisms for specific immune therapies. Overall, this methodology enables high throughput screening of many therapeutic candidates across dozens to hundreds of unique TO-models, thereby facilitating targeted precision therapy. Citation Format: Luca Lonini, Sonal Khare, Timothy Don Lopez, Stanislaw Szydlo, Michael Streit, Olga A. Karginova, Mary K. Flaherty, Ameen Salhaudeen, Andre Kunert, Anna Oja, Kate Sasser, Martin C. Stumpe, Chi-Sing Ho. Deep learning-enabled dynamic infiltration and response to NK therapies in solid tumor organoids [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2024; Part 1 (Regular Abstracts); 2024 Apr 5-10; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2024;84(6_Suppl):Abstract nr 2319.

Abstract 6150: MMP14 and MKLN1 as colorectal cancer susceptibility genes and a drug repurposing candidate from genome-wide association study in South Korea

Abstract Background: One of the major topics about colorectal cancer (CRC) is to find susceptible genes. Furthermore, functional interpretation of post-GWAS analysis gains attention and various approaches to find drug-repurposing candidates are used in cancer research. Our study aims to identify susceptible genes for CRC and provide drug-repurposing candidates. Methods: CRC patients in Seoul National University Hospital with prospective follow-up were recruited. From blood-derived DNA, genome-wide SNPs were genotyped using the Korea Biobank Array and 409,063 SNPs were extracted. SNP-based logistic regression model was fit for the event of CRC estimating beta values. Furthermore, post-GWAS analysis was conducted using regulatory function, multiple annotation and gene expression database. Drug-repurposing candidate was identified through the pre-trained deep neural network with susceptible genes. Results: We used 500 participants of CRC cases and 1,500 participants of healthy controls. For GWAS, statistically significant associations were mapped to MKLN1, MMP14, PTPN14 and MKL2 genes. For post-GWAS analysis, strong associations were found for MKLN1 (rs75170436, 7q32.3, Beta=-0.90, P-value=5.90×10-13) and MMP14 (rs3751489, 14q11.2, Beta=-1.91, P-value=2.31×10-12). From the drug-repurposing analysis, PPARδ/β agonist (GW0742; Binding score=3.79(MKLN1), 12.06(MMP14)) was identified for both genes with the highest score. Conclusions: We revealed MKLN1 and MMP14 genes as susceptible genes associated with CRC development. Additionally, we identified PPARδ/β agonist (GW0742) as a drug-repurposing candidate for the treatment of CRC. These findings can promote the understanding of CRC development and provide a novel therapeutic candidate for CRC patients in the future. Citation Format: Dabin Yun, Jung-Ho Yang, Sun-Seong Kweon, Jin-ah Sim, Minjung Kim, Ji Won Park, Seung Yong Jeong, Aesun Shin, Nan Song. MMP14 and MKLN1 as colorectal cancer susceptibility genes and a drug repurposing candidate from genome-wide association study in South Korea [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2024; Part 1 (Regular Abstracts); 2024 Apr 5-10; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2024;84(6_Suppl):Abstract nr 6150.

Abstract 5412: In vitro plasticity profiling of metastatic cancer cells with an imaging-based pipeline

Abstract Different types of cancer cells exhibit varying levels of invasiveness, migration behavior and ability to resume growth at distant sites. In order to classify cell lines for these fundamentally important traits of metastasis, we sought to generate an image-based in vitro assay that quantifies morphological and functional features of cancer cell plasticity. We developed an Imaging-Based Pipeline as an R Package for in vitro cell monitoring using the breast cancer cell lines MDA-MB-231 and MCF-7 that represent high-metastatic and low-metastatic cells in vivo, respectively, to identify relevant feature variables. Live cell imaging via the zenCELL owl system captured images for labeling and segmentation. Pre-trained neural network models from Cellpose, combined with user-calibrations, accurately segmented cells for tracking. The tracking pipeline calculated similarity coefficients, Jaccard index, between consecutive frames for cell trajectory analysis. Geometric and morphological features, speed, and distance measurements were derived for assessing structural motility and proliferation-associated changes. The analysis of two cell line models revealed significant variations in morphological changes, with MDA-MB-231 showing more rapid changes between round and elongated shape compared to MCF-7 cells. Moreover, the analysis of cell motility, consistent with the morphological changes, revealed that TGF-beta1 induced migration in MDA-MB-231 cells much stronger than in MCF-7 cells. Likewise, differences in proliferation induction upon addition of growth factor was monitored, showing that MCF-7 cells are more responsive to the growth stimulation than MDA-MB-231 cells. Consequently, the morphological plasticity of cancer cells, closely linked to the epithelial-to-mesenchymal transition (EMT) state, emerges as a defining factor for the in vitro plasticity of these cells. Quantified proliferation rate and migration speed of the cells was then used to establish an in vitro "plasticity index" that captures a cell line’s ability to switch between migration and proliferation. This index will undergo validation using cell lines with well-established metastatic propensities. In conclusion, we developed an Imaging-Based Pipeline that can effectively explore cancer cell plasticity in vitro and monitor the impact of genetic or pharmacologic perturbations on this critical characteristic of aggressive cancer cells. Citation Format: Saba Sameri, Durdam Das, Stefanie Michaelis, Joachim Wegener, Hedayatollah Hosseini, Martin Hoffmann. In vitro plasticity profiling of metastatic cancer cells with an imaging-based pipeline [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2024; Part 1 (Regular Abstracts); 2024 Apr 5-10; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2024;84(6_Suppl):Abstract nr 5412.

Contrastive pre-training and 3D convolution neural network for RNA and small molecule binding affinity prediction.

The diverse structures and functions inherent in RNAs present a wealth of potential drug targets. Some small molecules are anticipated to serve as leading compounds, providing guidance for the development of novel RNA-targeted therapeutics. Consequently, the determination of RNA-small molecule binding affinity is a critical undertaking in the landscape of RNA-targeted drug discovery and development. Nevertheless, to date, only one computational method for RNA-small molecule binding affinity prediction has been proposed. The prediction of RNA-small molecule binding affinity remains a significant challenge. The development of a computational model is deemed essential to effectively extract relevant features and predict RNA-small molecule binding affinity accurately. In this study, we introduced RLaffinity, a novel deep learning model designed for the prediction of RNA-small molecule binding affinity based on 3D structures. RLaffinity integrated information from RNA pockets and small molecules, utilizing a 3D convolutional neural network (3D-CNN) coupled with a contrastive learning-based self-supervised pre-training model. To the best of our knowledge, RLaffinity was the first deep learning based method for the prediction of RNA-small molecule binding affinity. Our experimental results exhibited RLaffinity's superior performance compared to baseline methods, revealed by all metrics. The efficacy of RLaffinity underscores the capability of 3D-CNN to accurately extract both global pocket information and local neighbor nucleotide information within RNAs. Notably, the integration of a self-supervised pre-training model significantly enhanced predictive performance. Ultimately, RLaffinity was also proved as a potential tool for RNA-targeted drugs virtual screening. https://github.com/SaisaiSun/RLaffinity.

Extraction of Forest Road Information from CubeSat Imagery Using Convolutional Neural Networks

Forest roads provide access to remote wooded areas, serving as key transportation routes and contributing to human impact on the local environment. However, large animals, such as bears (Ursus sp.), moose (Alces alces), and caribou (Rangifer tarandus caribou), are affected by their presence. Many publicly available road layers are outdated or inaccurate, making the assessment of landscape objectives difficult. To address these gaps in road location data, we employ CubeSat Imagery from the Planet constellation to predict the occurrence of road probabilities using a SegNet Convolutional Neural Network. Our research examines the potential of a pre-trained neural network (VGG-16 trained on ImageNet) transferred to the remote sensing domain. The classification is refined through post-processing, which considers spatial misalignment and road width variability. On a withheld test subset, we achieve an overall accuracy of 99.1%, a precision of 76.1%, and a recall of 91.2% (F1-Score: 83.0%) after considering these effects. We investigate the performance with respect to canopy coverage using a spectral greenness index, topography (slope and aspect), and land cover metrics. Results found that predictions are best in flat areas, with low to medium canopy coverage, and in the forest (coniferous and deciduous) land cover classes. The results are vectorized into a drivable road network, allowing for vector-based routing and coverage analyses. Our approach digitized 14,359 km of roads in a 23,500 km2 area in British Columbia, Canada. Compared to a governmental dataset, our method missed 10,869 km but detected an additional 5774 km of roads connected to the network. Finally, we use the detected road locations to investigate road age by accessing an archive of Landsat data, allowing spatiotemporal modelling of road access to remote areas. This provides important information on the development of the road network over time and the calculation of impacts, such as cumulative effects on wildlife.

Enhancing Arabic Sign Language Interpretation: Leveraging Convolutional Neural Networks and Transfer Learning

In a world essentializing communication for human connection, the deaf community encounters distinct barriers. Sign language, their main communication method is rich in hand gestures but not widely understood outside their community, necessitating interpreters. The existing solutions for sign language recognition depend on extensive datasets for model training, risking overfitting with complex models. The scarcity of details on dataset sizes and model specifics in studies complicates the scalability and verification of these technologies. Furthermore, the omission of precise accuracy metrics in some research leaves the effectiveness of gesture recognition by these models in question. The key phases of this study are Data collection, Data preprocessing, Feature extraction using CNN and finally transfer learning-based classification. The purpose of utilizing CNN and transfer learning is to tap into pre-trained neural networks for optimizing performance on new, related tasks by reusing learned patterns, thus accelerating development and improving accuracy. Data preprocessing further involves resizing of images, normalization, standardization, color space conversion, augmentation and noise reduction. This phase is capable enough to prune the image dataset by improving the efficiency of the classifier. In the subsequent phase, feature extraction has been performed that includes the convolution layer, feature mapping, pooling layer and dropout layer to obtain refined features from the images. These refined features are used for classification using ResNet. Three different datasets are utilized for the assessment of proposed model. The ASL-DS-I Dataset includes a total of 5832 images of hand gestures whereas, ASL-DS-II contains 54,049 images and ASL-DS-III dataset includes 7857 images adopted from specified web links. The obtained results have been evaluated by using standard metrics including ROC curve, Precision, Recall and F-measure. Meticulous experimental analysis and comparison with three standard baseline methods demonstrated that the proposed model gives an impressive recognition accuracy of 96.25%, 95.85% and 97.02% on ASL-DS-I, ASL-DS-II and ASL-DS-III, respectively.

Copyright and sui generis protection of trained neural networks

The article provides a legal analysis of the possibility of copyright and sui generis protection of pre-trained (pre-taught) neural networks in their objective expression, taking into account new legislation, in particular, Law No. 2811-IX. The components of the trained neural network were analyzed in the context of their possible legal qualification as objects of copyright and sui generis in order to determine effective ways of protecting such objects both individually and as the trained neural network in general.

SENKRON SIKIŞTIRMA DÖNÜŞÜMÜ VE DERİN ÖĞRENME KULLANILARAK FOTOPLETİSMOGRAFİ TABANLI KAN BASINCI KESTİRİMİ

Cardiovascular diseases are one of the deadliest health problems. Hypertension is the most common reason for cardiovascular diseases. Keeping the blood pressure (BP) level under control is the only way to protect against the deadly results of hypertension. Therefore, monitoring BP regularly makes it possible to detect dangerous conditions in patients with hypertension. With the rapid developments in computers and sensor technologies, it is becoming possible to monitor BP levels continuously by using photoplethysmogram (PPG) signals. This work presents a non-invasive BP prediction method using one channel PPG signal. We employed the Synchrosqueezing Transform to obtain Time-Frequency (TF) images of the PPG signals. The TF images were used to feed a pre-trained deep neural network. We estimated the BP levels inside the 5-second intervals. Our method estimates BP levels with a mean error (ME) of 0.2148 mmHg and -0.0370 mmHg in the systolic and diastolic blood pressure (SBP and DBP) respectively. The ME values of our method are in the applicable levels. The standard deviation (SD) of our method is 5.0642 mmHg for DBP and 10.9904 mmHg for SBP. The upper limit specified by the AAMI is 8 mmHg. Also, our method is coherent with grades A and B according to the BHS standard.

Improving chemical reaction yield prediction using pre-trained graph neural networks

Graph neural networks (GNNs) have proven to be effective in the prediction of chemical reaction yields. However, their performance tends to deteriorate when they are trained using an insufficient training dataset in terms of quantity or diversity. A promising solution to alleviate this issue is to pre-train a GNN on a large-scale molecular database. In this study, we investigate the effectiveness of GNN pre-training in chemical reaction yield prediction. We present a novel GNN pre-training method for performance improvement.Given a molecular database consisting of a large number of molecules, we calculate molecular descriptors for each molecule and reduce the dimensionality of these descriptors by applying principal component analysis. We define a pre-text task by assigning a vector of principal component scores as the pseudo-label to each molecule in the database. A GNN is then pre-trained to perform the pre-text task of predicting the pseudo-label for the input molecule. For chemical reaction yield prediction, a prediction model is initialized using the pre-trained GNN and then fine-tuned with the training dataset containing chemical reactions and their yields. We demonstrate the effectiveness of the proposed method through experimental evaluation on benchmark datasets.

Data Augmentation Techniques for Accurate Action Classification in Stroke Patients with Hemiparesis.

Stroke survivors with hemiparesis require extensive home-based rehabilitation. Deep learning-based classifiers can detect actions and provide feedback based on patient data; however, this is difficult owing to data sparsity and heterogeneity. In this study, we investigate data augmentation and model training strategies to address this problem. Three transformations are tested with varying data volumes to analyze the changes in the classification performance of individual data. Moreover, the impact of transfer learning relative to a pre-trained one-dimensional convolutional neural network (Conv1D) and training with an advanced InceptionTime model are estimated with data augmentation. In Conv1D, the joint training data of non-disabled (ND) participants and double rotationally augmented data of stroke patients is observed to outperform the baseline in terms of F1-score (60.9% vs. 47.3%). Transfer learning pre-trained with ND data exhibits 60.3% accuracy, whereas joint training with InceptionTime exhibits 67.2% accuracy under the same conditions. Our results indicate that rotational augmentation is more effective for individual data with initially lower performance and subset data with smaller numbers of participants than other techniques, suggesting that joint training on rotationally augmented ND and stroke data enhances classification performance, particularly in cases with sparse data and lower initial performance.

Transfer Learning with EfficientNetV2S for Automatic Face Shape Classification

The classification of human face shapes, a pivotal aspect of one’s appearance, plays a crucial role in diverse fields like beauty, cosmetics, healthcare, and security. In this paper, we present a multi-step methodology for face shape classification, harnessing the potential of transfer learning and a pretrained EfficientNetV2S neural network. Our approach comprises key phases, including preprocessing, augmentation, training, and testing, ensuring a comprehensive and reliable solution. The preprocessing step involves precise face detection, cropping, and image scaling, laying a solid foundation for accurate feature extraction. Our methodology utilizes a publicly available dataset of female celebrities, comprising five face shape classes: heart, oblong, oval, round, and square. By augmenting this dataset during training, we magnify its diversity, enabling better generalization and enhancing the model’s robustness. With the EfficientNetV2S neural network, we employ transfer learning, leveraging pretrained weights to optimize accuracy, training speed, and parameter size. The result is a highly efficient and effective model, which outperforms state-of-the-art approaches on the same dataset, boasting an outstanding overall accuracy of 96.32%. Our findings demonstrate the efficiency of our approach, proving its potential in the field of face shape classification. The success of our methodology holds promise for various applications, offering valuable insights into beauty analysis, cosmetic recommendations, and personalized healthcare.

Edge impulse-based pretrained neural network for diagnosing COVID-19

COVID-19 has wreaked havoc on a global scale, primarily owing to its extraordinary contagiousness, thereby straining local healthcare systems to their limits. While the Reverse Transcription Polymerase Chain Reaction (RT-PCR) test is known for its specificity, it suffers from time-consuming procedures and a notable false negative rate. Consequently, there is an immediate imperative for a swift and precise diagnostic approach. This paper introduces a novel concept, employing artificial intelligence, to address these challenges effectively. Specifically, this study employed a transfer learning model provided by the Edge Impulse platform and a dataset containing chest X-ray images of children. This study pre-processed these images and trained and tested them several times using different image sizes and network architectures. The experimental results show that the model achieves very high accuracy (>99%) with 160*160 image size and version 1.0 or 0.35 of the network architecture. These results clearly support the hypothesis that migration learning can play an important role in the fast and accurate diagnosis of COVID-19 with appropriate image size and network architecture. This research can be used as a way to rapidly train locally adapted AI models to achieve rapid assisted diagnosis of this type of acute infectious disease.

NEMA NU 2-2018 evaluation and image quality optimization of a new generation digital 32-cm axial field-of-view Omni Legend PET-CT using a genetic evolutionary algorithm

A performance evaluation was conducted on the new General Electric (GE) digital Omni Legend PET-CT system with 32 cm extended field of view. The first commercially available clinical digital bismuth germanate system. The system does not use time of flight (ToF). Testing was performed in accordance with the NEMA NU2–2018 standard. A comparison was made between two other commercial GE scanners with extended fields of view; the Discovery MI − 6 ring (ToF enabled) and the Discovery IQ (non-ToF). A genetic evolutionary algorithm was developed to optimize image reconstruction parameters from image quality assessments. The Omni demonstrated average spatial resolutions at 1 cm radial offset as 3.9 mm FWHM. The total system sensitivity at the center was 44.36 cps/kBq. The peak NECR was measured as 501 kcps at 17.8 kBq ml−1 with a 35.48% scatter fraction. The maximum count-rate error below NECR peak was 5.5%. Using standard iterative reconstructions, sphere contrast recovery coefficients were from 52.7 ± 3.2% (10 mm) to 92.5 ± 2.4% (37 mm). The PET-CT co-registration accuracy was 2.4 mm. In place of ToF, the Omni employs software corrections through a pre-trained neural network (PDL) (trained on non-ToF to ToF) that takes Bayesian penalized likelihood reconstruction (Q.Clear) images as input. The optimum parameters for image reconstruction, determined using the genetic algorithm were a Q.Clear parameter, β, of 350 and a ‘medium’ PDL setting. Using standard iterative reconstructions, the Omni initially showed increased background variability compared to the Discovery MI. With optimized PDL reconstruction parameters selected using the genetic algorithm the performance of the Omni surpassed that of the Discovery MI on all NEMA tests. The genetic algorithm’s demonstrated ability to enhance image quality in PET-CT imaging underscores the importance of algorithm driven optimization and underscores the requirement to validate its use in the clinical setting.

Editorial

Dear Readers,  It gives me great pleasure to announce the second regular issue of 2024. I would like to thank all the authors for their sound research and the editorial board for the extremely valuable reviews and suggestions for improvement. These contributions together with the generous support of the consortium members enable us to run our journal and maintain its quality.  I would still like to expand our editorial board: If you are a tenured associate professor or above with a good publication record, please apply to join our editorial board. We are also interested in high-quality proposals for special issues on new topics and emerging trends. And finally, we are still looking for some financial support for 2024 to cover all our expenses. We would be very grateful if your library or institution can support us. We would then be happy to add it to our consortium list.  In this regular issue, I am very pleased to present 6 accepted papers by 19 authors from 9 countries: Canada, China, Croatia, India, Kazakhstan, México, Sri Lanka, Ukraine, and Vietnam.  Petra Grd, Igor Tomičić, and Ena Barčić from Croatia address in their article a multi-step methodology for face shape classification that is based on the potential of transfer learning and a pretrained EfficientNetV2S neural network.  Lizbeth Alejandra Hernández-González, Ulises Juárez-Martínez, Jezreel Mejía, and Alberto Aguilar-Laserre from México focus their research on applying the naturalistic programming paradigm within a software development process using a naturalistic software development method.  In a joint research, Shanshan Jia from China, Gaukhar A. Kamalova from the Republic of Kazakhstan, and Dmytro Mykhalevskiy from Ukraine report on a mobile handover technique aligning with the neighbour discovery paradigm in 6LoWPAN.  Ajay Kumar from India is investigating a mechanism to assess machine learning approaches for software effort estimation (SEE) modeling in the context of accuracy measures, specifically exploring machine learning techniques for SEE modeling as a multi-criteria decision making (MCDM) problem.  Anne Perera and Amitha Caldera from Sri Lanka conduct a comprehensive review on sentiment analysis in the context of mix of languages, phonetic typing and lexical borrowing in web communication.  And last but not least, in a collaboration between researchers from Vietnam and Canada, Tien Quang Dam, Nghia Thinh Nguyen, Trung Viet Le, Tran Duc Le, Sylvestre Uwizeyemungu, and Thang Le-Dinh look into malware detection methods, specifically leveraging machine learning to encode critical information from portable executable (PE) headers into visual representations of ransomware samples.  Enjoy Reading!  Cordially,  Christian Gütl, Managing Editor-in-Chief Graz University of Technology, Graz, Austria

Diagnosis of Alzheimer's Disease and Mild Cognitive Impairment Using Convolutional Neural Networks.

Alzheimer's disease and mild cognitive impairment are common diseases in the elderly, affecting more than 50 million people worldwide in 2020. Early diagnosis is crucial for managing these diseases, but their complexity poses a challenge. Convolutional neural networks have shown promise in accurate diagnosis. The main objective of this research is to diagnose Alzheimer's disease and mild cognitive impairment in healthy individuals using convolutional neural networks. This study utilized three different convolutional neural network models, two of which were pre-trained models, namely AlexNet and DenseNet, while the third model was a CNN1D-LSTM neural network. Among the neural network models used, the AlexNet demonstrated the highest accuracy, exceeding 98%, in diagnosing mild cognitive impairment and Alzheimer's disease in healthy individuals. Furthermore, the accuracy of the DenseNet and CNN1D-LSTM models is 88% and 91.89%, respectively. The research highlights the potential of convolutional neural networks in diagnosing mild cognitive impairment and Alzheimer's disease. The use of pre-trained neural networks and the integration of various patient data contribute to achieving accurate results. The high accuracy achieved by the AlexNet neural network underscores its effectiveness in disease classification. These findings pave the way for future research and improvements in the field of diagnosing these diseases using convolutional neural networks, ultimately aiding in early detection and effective management of mild cognitive impairment and Alzheimer's disease.

Deep Learning-Based Vehicle Type and Color Classification to Support Safe Autonomous Driving

This technology can prevent accidents involving large vehicles, such as trucks or buses, by selecting an optimal driving lane for safe autonomous driving. This paper proposes a method for detecting forward-driving vehicles within road images obtained from a vehicle’s DashCam. The proposed method also classifies the types and colors of the detected vehicles. The proposed method uses a YOLO deep learning network for vehicle detection based on a pre-trained ResNet-50 convolutional neural network. Additionally, a Resnet-50 CNN-based object classifier, using transfer learning, was used to classify vehicle types and colors. Vehicle types were classified into four categories based on size whereas vehicle colors were classified into eight categories. During autonomous driving, vehicle types are used to determine driving lanes, whereas vehicle colors are used to distinguish the road infrastructure, such as lanes, vehicles, roads, backgrounds, and buildings. The datasets used for learning consisted of road images acquired in various driving environments. The proposed method achieved a vehicle detection accuracy of 91.5%, vehicle type classification accuracy of 93.9%, and vehicle color classification accuracy of 94.2%. It accurately detected vehicles and classified their types and colors. These can be applied to autonomous and safe driving support systems to enhance the safety of autonomous vehicles.

Lightweight and high-precision materials property prediction using pre-trained Graph Neural Networks and its application to small dataset

Abstract Large data sets are essential for building deep learning models. However, generating large datasets with higher theoretical levels and larger computational models remains difficult due to the high cost of first-principles calculation. Here, we propose a lightweight and highly accurate machine learning approach using pre-trained Graph Neural Networks (GNNs) for industrially important but difficult to scale models. The proposed method was applied to a small dataset of graphene surface systems containing surface defects, and achieved comparable accuracy with six orders of magnitude faster learning than when the GNN was trained from scratch.

Transfer learning graph representations of molecules for pKa, 13C-NMR, and solubility

We explore transfer learning models from a pre-trained graph convolutional neural network representation of molecules, obtained from SchNet, to predict 13C-NMR, pKa, and log S solubility. SchNet learns a graph representation of a molecule by associating each atom with an “embedding vector” and interacts the atom-embeddings with each other by leveraging graph convolutional filters on their interatomic distances. We pre-trained SchNet on molecular energy and demonstrate that the pre-trained atomistic embeddings can then be used as a transferable representation for a wide array of properties. On the one hand, for atomic properties such as micro-pK1 and 13C-NMR, we investigate two models, one linear and one neural net, that input pre-trained atom-embeddings of a particular atom (e.g. carbon) and predict a local property (e.g., 13C-NMR). On the other hand, for molecular properties such as solubility, a size-extensive graph model is built using the embeddings of all atoms in the molecule as input. For all cases, qualitatively correct predictions are made with relatively little training data (<1000 training points), showcasing the ease with which pre-trained embeddings pick up on important chemical patterns. The proposed models successfully capture well-understood trends of pK1 and solubility. This study advances our understanding of current neural net graph representations and their capacity for transfer learning applications in chemistry.

The impact of climate change news on the US stock market

PurposeThis paper aims to investigate the impact of climate change news on the dynamics of US stock indices.Design/methodology/approachThe empirical basis of the study was 3,209 news articles. Sentiment analysis was performed by a pre-trained bidirectional FinBERT neural network. Thematic modeling is based on the neural network, BERTopic.FindingsThe results show that news sentiment can influence the dynamics of stock indices. In addition, five main news topics (finance and politics natural disasters and consequences industrial sector and Innovations activism and culture coronavirus pandemic) were identified, which showed a significant impact on the financial market.Originality/valueFirst, we extend the theoretical concepts. This study applies signaling theory and overreaction theory to the US stock market in the context of climate change. Second, in addition to the news sentiment, the impact of major news topics on US stock market returns is examined. Third, we examine the impact of sentimental and thematic news variables on US stock market indicators of economic sectors. Previous works reveal the impact of climate change news on specific sectors of the economy. This paper includes stock indices of the economic sectors most related to the topic of climate change. Fourth, the research methodology consists of modern algorithms. An advanced textual analysis method for sentiment classification is applied: a pre-trained bidirectional FinBERT neural network. Modern thematic modeling is carried out using a model based on the neural network, BERTopic. The most extensive topics are “finance and politics of climate change” and “natural disasters and consequences.”