Pre-trained Deep Neural Network Research Articles

Generating real-scene hologram through light field imaging and deep learning

Addressing the challenge of acquiring holograms from real-world scenes, this study introduces a novel approach leveraging light field cameras to capture light field data, which is subsequently transformed into authentic scene holograms. This methodology integrates light field imaging technology with a pre-trained deep neural network. To compensate for the limitations inherent in camera hardware, a super-resolution algorithm is employed. The conversion of light field information into RGB-D data facilitates its input into the deep neural network, enabling the inference of corresponding real-world scene holograms. Empirical evidence demonstrates that the system is capable of inferring high-resolution (1920 × 1080) real-world scene holograms within a timeframe of 5 s, utilizing hardware comprising an NVIDIA RTX 3060.

From Sleep Patterns to Heart Rhythms: Predicting Atrial Fibrillation from Overnight Polysomnograms.

Atrial fibrillation (AF) is often asymptomatic and thus under-observed. Given the high risks of stroke and heart failure among patients with AF, early prediction and effective management are crucial. Importantly, obstructive sleep apnea is highly prevalent among AF patients (60-90%); therefore, electrocardiogram (ECG) analysis from polysomnography (PSG), a standard diagnostic tool for subjects with suspected sleep apnea, presents a unique opportunity for the early prediction of AF. Our goal is to identify individuals at a high risk of developing AF in the future from a single-lead ECG recorded during standard PSGs. We analyzed 18,782 single-lead ECG recordings from 13,609 subjects at Massachusetts General Hospital, identifying AF presence using ICD-9/10 codes in medical records. Our dataset comprises 15,913 recordings without a medical record for AF and 2,056 recordings from patients who were first diagnosed with AF between 1 day to 15 years after the PSG recording. The PSG data were partitioned into training, validation, and test cohorts. In the first phase, a signal quality index (SQI) was calculated in 30-second windows and those with SQI < 0.95 were removed. From each remaining window, 150 hand-crafted features were extracted from time, frequency, time-frequency domains, and phase-space reconstructions of the ECG. A compilation of 12 statistical features summarized these window-specific features per recording, resulting in 1,800 features. We then updated a pre-trained deep neural network and data from the PhysioNet Challenge 2021 using transfer-learning to discriminate between recordings with and without AF using the same Challenge data. The model was applied to the PSG ECGs in 16-second windows to generate the probability of AF for each window. From the resultant probability sequence, 13 statistical features were extracted. Subsequently, we trained a shallow neural network to predict future AF using the extracted ECG and probability features. On the test set, our model demonstrated a sensitivity of 0.67, specificity of 0.81, and precision of 0.3 for predicting AF. Further, survival analysis for AF outcomes, using the log-rank test, revealed a hazard ratio of 8.36 (p-value of 1.93 × 10 -52 ). Our proposed ECG analysis method, utilizing overnight PSG data, shows promise in AF prediction despite a modest precision indicating the presence of false positive cases. This approach could potentially enable low-cost screening and proactive treatment for high-risk patients. Ongoing refinement, such as integrating additional physiological parameters could significantly reduce false positives, enhancing its clinical utility and accuracy.

A comprehensive exploration of approximate DNN models with a novel floating-point simulation framework

This paper introduces TorchAxf11https://github.com/knuscent/TorchAxf., a framework for fast simulation of diverse approximate deep neural network (DNN) models, including spiking neural networks (SNNs). The proposed framework utilizes various approximate adders and multipliers, supports industrial standard reduced precision floating-point formats, such as bfloat16, and accommodates user-customized precision representations. Leveraging GPU acceleration on the PyTorch framework, TorchAxf accelerates approximate DNN training and inference. In addition, it allows seamless integration of arbitrary approximate arithmetic algorithms with C/C++ behavioral models to emulate approximate DNN hardware accelerators.We utilize the proposed TorchAxf framework to assess twelve popular DNN models under approximate multiply-and-accumulate (MAC) operations. Through comprehensive experiments, we determine the suitable degree of floating-point arithmetic approximation for these DNN models without significant accuracy loss and offer the optimal reduced precision formats for each DNN model. Additionally, we demonstrate that approximate-aware re-training can rectify errors and enhance pre-trained DNN models under reduced precision formats. Furthermore, TorchAxf, operating on GPU, remarkably reduces simulation time for complex DNN models using approximate arithmetic by up to 131.38× compared to the baseline optimized CPU implementation. Finally, we compare the proposed framework with state-of-the-art frameworks to highlight its superiority.

Optimizing Breast Cancer Detection: Deep Transfer Learning Empowered by SVM Classifiers

Objectives: The research aims to enhance breast cancer detection accuracy and effectiveness using deep transfer learning and pre-trained neural networks. It analyses breast ultrasound images and identifies important characteristics using pre-trained networks. The goal is to create a more efficient and accurate automated system for breast cancer detection. Methods: The study uses breast ultrasound cancer image data from the Kaggle Data Repository to extract informative features, identify cancer-related characteristics, and classify them into benign, malignant, and normal tissue. Pre-trained Deep Neural Networks (DNNs) extract these features and feed them into a 10-fold cross-validation SVM classifier. The SVM is evaluated using various kernel functions to identify the best kernel for separating data points. This methodology aims to achieve accurate classification of breast cancer in ultrasound images. Findings: The study confirms the effectiveness of deep transfer learning for breast cancer detection in ultrasound images, with Inception V3 outperforming VGG-16 and VGG-19 in extracting relevant features. The combination of Inception V3 and the SVM classifier with a polynomial kernel achieved the highest classification accuracy, indicating its ability to model complex relationships. The study demonstrated an AUC of 0.944 and a classification accuracy of 87.44% using the Inception V3 + SVM polynomial. Novelty: This research demonstrates the potential of deep transfer learning and SVM classifiers for accurate breast cancer detection in ultrasound images. It integrates Inception V3, VGG-16, and VGG-19 for breast cancer detection, demonstrating improved classification accuracy. The combination of Inception V3 and SVM (polynomial) achieved a significant AUC (0.944) and classification accuracy (87.44%), outperforming other models tested. This research underscores the potential of these technologies for accurate breast cancer detection in ultrasound images. Keywords: Breast Cancer, Deep Learning, Feature Extraction, Inception-v3, SVM, Transfer Learning

Enhancing cervical cancer detection and robust classification through a fusion of deep learning models

Cervical cancer, the second most prevalent cancer affecting women, arises from abnormal cell growth in the cervix, a crucial anatomical structure within the uterus. The significance of early detection cannot be overstated, prompting the use of various screening methods such as Pap smears, colposcopy, and Human Papillomavirus (HPV) testing to identify potential risks and initiate timely intervention. These screening procedures encompass visual inspections, Pap smears, colposcopies, biopsies, and HPV-DNA testing, each demanding the specialized knowledge and skills of experienced physicians and pathologists due to the inherently subjective nature of cancer diagnosis. In response to the imperative for efficient and intelligent screening, this article introduces a groundbreaking methodology that leverages pre-trained deep neural network models, including Alexnet, Resnet-101, Resnet-152, and InceptionV3, for feature extraction. The fine-tuning of these models is accompanied by the integration of diverse machine learning algorithms, with ResNet152 showcasing exceptional performance, achieving an impressive accuracy rate of 98.08%. It is noteworthy that the SIPaKMeD dataset, publicly accessible and utilized in this study, contributes to the transparency and reproducibility of our findings. The proposed hybrid methodology combines aspects of DL and ML for cervical cancer classification. Most intricate and complicated features from images can be extracted through DL. Further various ML algorithms can be implemented on extracted features. This innovative approach not only holds promise for significantly improving cervical cancer detection but also underscores the transformative potential of intelligent automation within the realm of medical diagnostics, paving the way for more accurate and timely interventions.

Manifesto of Deep Learning Architecture for Aspect Level Sentiment Analysis to extract customer criticism

Sentiment analysis, a critical task in natural language processing, aims to automatically identify and classify the sentiment expressed in textual data. Aspect-level sentiment analysis focuses on determining sentiment at a more granular level, targeting specific aspects or features within a piece of text. In this paper, we explore various techniques for sentiment analysis, including traditional machine learning approaches and state-of-the-art deep learning models. Additionally, deep learning techniques has been utilized to identifying and extracting specific aspects from text, addressing aspect-level ambiguity, and capturing nuanced sentiments for each aspect. These datasets are valuable for conducting aspect-level sentiment analysis. In this article, we explore a language model based on pre-trained deep neural networks. This model can analyze sequences of text to classify sentiments as positive, negative, or neutral without explicit human labeling. To evaluate these models, data from Twitter's US airlines sentiment database was utilized. Experiments on this dataset reveal that the BERT, RoBERTA and DistilBERT model outperforms than the ML based model in accuracy and is more efficient in terms of training time. Notably, our findings showcase significant advancements over previous state-of-the-art methods that rely on supervised feature learning, bridging existing gaps in sentiment analysis methodologies. Our findings shed light on the advancements and challenges in sentiment analysis, offering insights for future research directions and practical applications in areas such as customer feedback analysis, social media monitoring, and opinion mining.

A deep learning approach for Direct Immunofluorescence pattern recognition of Autoimmune Bullous Diseases.

Artificial intelligence (AI) is reshaping healthcare, using machine and deep learning to enhance disease management. Dermatology has seen improved diagnostics, particularly in skin cancer detection, through the integration of AI. However, the potential of AI in automating immunofluorescence imaging for autoimmune bullous skin diseases remains untapped. While direct immunofluorescence (DIF) supports diagnosis, its manual interpretation can hinder efficiency. The use of deep learning to automatically classify DIF patterns, including the Intercellular Pattern (ICP) and the Linear Pattern (LP), holds promise for improving the diagnosis of autoimmune bullous skin diseases. The objectives of this study are to develop AI algorithms for automated classification of autoimmune bullous skin disease DIF patterns, such as ICP and LP. This aims to enhance diagnostic accuracy, streamline disease management, and improve patient outcomes through deep learning-driven immunofluorescence interpretation. We collected immunofluorescence images from skin biopsies of patients suspected of AIBD between January 2022 and January 2024. Skin tissue was obtained via 5-mm punch biopsy, prepared for direct immunofluorescence. Experienced dermatologists classified the images into three classes: ICP, LP, and negative. To evaluate our deep learning approach, we divided the images into training (436) and test sets (93). We employed transfer learning with pre-trained deep neural networks and conducted 5-fold cross-validation to assess model performance. Our dataset's class imbalance was addressed using weighted loss and data augmentation strategies. The models were trained for 50 epochs using Pytorch, achieving an image size of 224x224 for both CNNs and the Swin Transformer. Our study compared six CNNs and the Swin transformer for AIBDs image classification, with the Swin transformer achieving the highest average validation accuracy of 98.5%. On a separate test set, the best model attained an accuracy of 94.6%, demonstrating 95.3% sensitivity and 97.5% specificity across AIBDs classes. Visualization with Grad-CAM highlighted the model's reliance on characteristic patterns for accurate classification. The study highlighted CNN's accuracy in identifying DIF features. This approach aids automated analysis and reporting, offering reproducibility, speed, data handling, and cost-efficiency. Integrating deep learning in skin immunofluorescence promises precise diagnostics and streamlined reporting in this branch of dermatology.

Analysis of a Deep Learning Model for 12-Lead ECG Classification Reveals Learned Features Similar to Diagnostic Criteria.

Despite their remarkable performance, deep neural networks remain unadopted in clinical practice, which is considered to be partially due to their lack of explainability. In this work, we apply explainable attribution methods to a pre-trained deep neural network for abnormality classification in 12-lead electrocardiography to open this "black box" and understand the relationship between model prediction and learned features. We classify data from two public databases (CPSC 2018, PTB-XL) and the attribution methods assign a "relevance score" to each sample of the classified signals. This allows analyzing what the network learned during training, for which we propose quantitative methods: average relevance scores over a) classes, b) leads, and c) average beats. The analyses of relevance scores for atrial fibrillation and left bundle branch block compared to healthy controls show that their mean values a) increase with higher classification probability and correspond to false classifications when around zero, and b) correspond to clinical recommendations regarding which lead to consider. Furthermore, c) visible P-waves and concordant T-waves result in clearly negative relevance scores in atrial fibrillation and left bundle branch block classification, respectively. Results are similar across both databases despite differences in study population and hardware. In summary, our analysis suggests that the DNN learned features similar to cardiology textbook knowledge.

Automated Tuberculosis Classification with Chest X-Rays Using Deep Neural Networks -Case Study: Nigerian Public Health

Tuberculosis, a contagious lung ailment, stands as a prominent global mortality factor. Its significant impact on public health in Nigeria necessitates comprehensive intervention strategies. Detecting, preventing, and treating this disease remains imperative. Chest X-ray (CXR) images hold a pivotal role among diagnostic tools. Recent strides in deep learning have notably improved medical image analysis. In this research, we harnessed publicly available and proprietary CXR image datasets to construct robust models. Leveraging pre-trained deep neural networks, we aimed to enhance tuberculosis detection. Impressively, our experimentation yielded remarkable outcomes. Notably, f1-scores of 98% and 86% were attained on the respective public and private datasets. These results underscore the potency of deep neural networks in effectively identifying tuberculosis from CXR images. The study emphasizes the promise of this technology in combating the disease's spread and impact.

Harnessing Edge Information for Improved Robustness in Vision Transformers

Deep Neural Networks (DNNs) have demonstrated remarkable accuracy in vision classification tasks. However, they exhibit vulnerability to additional noises known as adversarial attacks. Previous studies hypothesize that this vulnerability might stem from the fact that high-accuracy DNNs heavily rely on irrelevant and non-robust features, such as textures and the background. In this work, we reveal that edge information extracted from images can provide relevant and robust features related to shapes and the foreground. These features assist pretrained DNNs in achieving improved adversarial robustness without compromising their accuracy on clean images. A lightweight and plug-and-play EdgeNet is proposed, which can be seamlessly integrated into existing pretrained DNNs, including Vision Transformers, a recent family of state-of-the-art models for vision classification. Our EdgeNet can process edges derived from either clean nature images or noisy adversarial images, yielding robust features which can be injected into the intermediate layers of the frozen backbone DNNs. The cost of obtaining such edges using conventional edge detection algorithms (e.g., Canny edge detector) is marginal, and the cost of training the EdgeNet is equivalent to that of fine-tuning the backbone network with techniques such as Adapter.

Causality-Inspired Invariant Representation Learning for Text-Based Person Retrieval

Text-based Person Retrieval (TPR) aims to retrieve relevant images of specific pedestrians based on the given textual query. The mainstream approaches primarily leverage pretrained deep neural networks to learn the mapping of visual and textual modalities into a common latent space for cross-modality matching. Despite their remarkable achievements, existing efforts mainly focus on learning the statistical cross-modality correlation found in training data, other than the intrinsic causal correlation. As a result, they often struggle to retrieve accurately in the face of environmental changes such as illumination, pose, and occlusion, or when encountering images with similar attributes. In this regard, we pioneer the observation of TPR from a causal view. Specifically, we assume that each image is composed of a mixture of causal factors (which are semantically consistent with text descriptions) and non-causal factors (retrieval-irrelevant, e.g., background), and only the former can lead to reliable retrieval judgments. Our goal is to extract text-critical robust visual representation (i.e., causal factors) and establish domain invariant cross-modality correlations for accurate and reliable retrieval. However, causal/non-causal factors are unobserved, so we emphasize that ideal causal factors that can simulate causal scenes should satisfy two basic principles:1） Independence: being independent of non-causal factors, and 2）Sufficiency: being causally sufficient for TPR across different environments. Building on that, we propose an Invariant Representation Learning method for TPR (IRLT), that enforces the visual representations to satisfy the two aforementioned critical properties. Extensive experiments on three datasets clearly demonstrate the advantages of IRLT over leading baselines in terms of accuracy and generalization.

FlexiBO: A Decoupled Cost-Aware Multi-objective Optimization Approach for Deep Neural Networks (Abstract Reprint)

The design of machine learning systems often requires trading off different objectives, for example, prediction error and energy consumption for deep neural networks (DNNs). Typically, no single design performs well in all objectives; therefore, finding Pareto-optimal designs is of interest. The search for Pareto-optimal designs involves evaluating designs in an iterative process, and the measurements are used to evaluate an acquisition function that guides the search process. However, measuring different objectives incurs different costs. For example, the cost of measuring the prediction error of DNNs is orders of magnitude higher than that of measuring the energy consumption of a pre-trained DNN as it requires re-training the DNN. Current state-of-the-art methods do not consider this difference in objective evaluation cost, potentially incurring expensive evaluations of objective functions in the optimization process. In this paper, we develop a novel decoupled and cost-aware multi-objective optimization algorithm, which we call Flexible Multi-Objective Bayesian Optimization (FlexiBO) to address this issue. For evaluating each design, FlexiBO selects the objective with higher relative gain by weighting the improvement of the hypervolume of the Pareto region with the measurement cost of each objective. This strategy, therefore, balances the expense of collecting new information with the knowledge gained through objective evaluations, preventing FlexiBO from performing expensive measurements for little to no gain. We evaluate FlexiBO on seven state-of-the-art DNNs for image recognition, natural language processing (NLP), and speech-to-text translation. Our results indicate that, given the same total experimental budget, FlexiBO discovers designs with 4.8% to 12.4% lower hypervolume error than the best method in state-of-the-art multi-objective optimization.

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.

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.

Transfer learning model for estimating site amplification factors from limited microtremor H/V spectral ratios

Summary Site amplification factors (SAFs) of seismic ground motions are essential in evaluating and estimating seismic hazards. In our previous study, the authors proposed a simple and cost-effective method to estimate a SAF based on a deep neural network (DNN) model and microtremor horizontal-to-vertical spectral ratio (MHVR). Since the previous DNN model was based on the observed SAFs and MHVRs within a limited district in Japan, the applicability of the previous model to non-source regions with different site conditions was limited. This study explored the application of a transfer learning (TL) technique to adapt an existing (pre-trained) DNN model to new regions and a different database. The SAFs obtained through generalized spectral inversion technique (GIT) at the seismic observation stations (K-NET and KiK-net) in Japan were collated as the ground truth for site effects. MHVRs recorded at the stations in several districts of Japan were collected to construct a dataset for the development of the TL model. Subsequently, a TL model was constructed, leveraging the neural network layers and their weights from the pre-trained model while incorporating additional neural network layers to enhance the performance. During the training process, a total data set of 112 sites was divided into training set, validation set, and external test set by 1:1:5. Utilizing a cross-validation strategy, the residuals between pSAFs (pseudo-SAFs) estimated by the TL model and the observed SAFs were analyzed for the external test set containing 80 sites. The results showed that the TL model outperformed the pre-trained DNN model. The cross-validation results demonstrated that almost consistent prediction results were obtained regardless of any combination of 16 sites selected as the training set. Furthermore, by contrasting the influence of varying training set sizes on the performance of the TL model and comparing the TL model to a DNN model with an extended training set, the effectiveness of constructing the model with the limited number of data (16 sites) was ascertained. Finally, the effectiveness and limitations of the TL model were evaluated using MHVRs with peak frequencies falling outside the training set's range.

Children develop adult-like visual sensitivity to image memorability by the age of 4.

Adults have been shown to consistently remember and forget certain images despite large individual differences, suggesting a population-wide sensitivity to an image's intrinsic memorability-a measure of how successfully an image is remembered. While a decade of research has focused on image memorability among adults, the developmental trajectory of these consistencies in memory is understudied. Here, we investigate by what age children gain adult-like sensitivity to the image memorability effect. We utilized data from Saragosa-Harris et al. (2021), where 137 children aged between 3 and 5 years old encoded animal-scene image pairs and then after a 5-min, 24-hr, or 1-week delay performed a cued recognition task for each scene target given its animal cue. We tested adults' memory of the same scene images using ResMem (Needell & Bainbridge, 2022), a pretrained deep neural network that predicts adult image memorability scores, and using an online behavioral continuous recognition task (N = 116). Results showed that ResMem predictions, as a proxy of adults' memory, predicted scene memory of children by the age of 4 and were the most predictive of children's memory across ages after a long, 1-week delay. Children at age 3 show nonadult-like consistent memory patterns, implying that the nonadult-like memory patterns were not due to poor memory performance. Instead, 3-year-olds may have consistently used certain visual memory strategies that become less optimal as they age. Our results suggest that adult-like sensitivity to image memorability emerges by the age of 4 through experience. (PsycInfo Database Record (c) 2024 APA, all rights reserved).

Lab to Multiscene Generalization for Non-Line-of-Sight Identification With Small-Scale Datasets

Ultra-wideband (UWB) wireless indoor positioning systems rely on time of flight (TOF) to estimate distances but can be biased and miscalculated due to non-line-of-sight (NLOS) transmission channels in complex environments. Therefore, to remove errors, several machine learning techniques have been proposed for identifying NLOS signals from Channel Impulse Responses (CIRs). However, as CIR signals could be heavily influenced by various environments, current NLOS classifiers are not universal to provide satisfactory accuracy for new scenarios and require detailed measurements on a large number of CIRs for training. Hence, we propose a generalization method based on data augmentation via noise injection and transfer learning to allow the deep neural network (DNN) trained under a lab condition to be applied to various and even harsh practical scenarios with the need to measure massive training data minimized. This paper presents the first demonstration that it is effective to utilize a lab-based pre-trained DNN for real-world transfer and white Gaussian noise data augmentation for ML-based NLOS identification on UWB CIRs to address the problem when it is not feasible to measure sufficient training data. Our testing results show that in two scenarios, corridor and parking lot, with only 50 CIR signals as the training set, the accuracy of the NLOS identification model after applying the proposed method is increased from 84.4% to 98.8% and from 81.1% to 97.1%, respectively.

A Deep Learning Method for the Health State Prediction of Lithium-Ion Batteries Based on LUT-Memory and Quantization

The precise determination of the state of health (SOH) of lithium-ion batteries is critical in the domain of battery management systems. The proposed model in this research paper emulates any deep learning or machine learning model by utilizing a Look Up Table (LUT) memory to store all activation inputs and their corresponding outputs. The operation that follows the completion of training is referred to as the LUT memory preparation procedure. This method’s lookup process supplants the inference process entirely and simply. This is achieved by discretizing the input data and features before binarizing them. The term for the aforementioned operation is the LUT inference method. This procedure was evaluated in this study using two distinct neural network architectures: a bidirectional long short-term memory (LSTM) architecture and a standard fully connected neural network (FCNN). It is anticipated that considerably greater efficiency and velocity will be achieved during the inference procedure when the pre-trained deep neural network architecture is inferred directly. The principal aim of this research is to construct a lookup table that effectively establishes correlations between the SOH of lithium-ion batteries and ensures a degree of imprecision that is tolerable. According to the results obtained from the NASA PCoE lithium-ion battery dataset, the proposed methodology exhibits a performance that is largely comparable to that of the initial machine learning models. Utilizing the error assessment metrics RMSE, MAE, and (MAPE), the accuracy of the SOH prediction has been quantitatively evaluated. The indicators mentioned above demonstrate a significant degree of accuracy when predicting SOH.

Detecting anomalous proteins using deep representations.

Many advances in biomedicine can be attributed to identifying unusual proteins and genes. Many of these proteins' unique properties were discovered by manual inspection, which is becoming infeasible at the scale of modern protein datasets. Here, we propose to tackle this challenge using anomaly detection methods that automatically identify unexpected properties. We adopt a state-of-the-art anomaly detection paradigm from computer vision, to highlight unusual proteins. We generate meaningful representations without labeled inputs, using pretrained deep neural network models. We apply these protein language models (pLM) to detect anomalies in function, phylogenetic families, and segmentation tasks. We compute protein anomaly scores to highlight human prion-like proteins, distinguish viral proteins from their host proteome, and mark non-classical ion/metal binding proteins and enzymes. Other tasks concern segmentation of protein sequences into folded and unstructured regions. We provide candidates for rare functionality (e.g. prion proteins). Additionally, we show the anomaly score is useful in 3D folding-related segmentation. Our novel method shows improved performance over strong baselines and has objectively high performance across a variety of tasks. We conclude that the combination of pLM and anomaly detection techniques is a valid method for discovering a range of global and local protein characteristics.

Object classification and visualization with edge artificial intelligence for a customized camera trap platform

The camera traps have revolutionized the image and video capture in ecology and are often used to monitor and record animal presence. With miniaturization of low power electronic devices, better battery technologies, and software advancements, it has become possible to use the edge devices, such as Raspberry Pi as camera traps that can not only capture images and videos, but can also enable sophisticated image processing, and off-site communications. These developments can help to provide near real-time insights and reduce the manual processing of images. The on-board image classification and visualization is facilitated by the advancements in the Deep Neural Networks (DNN), transfer learning approaches, and software libraries.This paper provides an investigation of image classification with transfer learning approaches using pre-trained DNN models, and visualizations with Explainable Artificial Intelligence (XAI) techniques on Raspberry Pi Zero (RPi-Z) edge device. The MobileNetV2 model was used for image classification on the Florida-Part1 dataset obtaining the results for precision, recall, and F1-score as 0.95, 0.96, and 0.95 respectively. We also compared the model performance of MobileNetV2, EfficientNetV2B0, and MobileViT models for classification on the Extinction dataset with the best results for precision, recall, and F1-score as 0.97, 0.96, and 0.96 respectively, obtained with the EfficientNetV2B0 model. Two XAI techniques, Gradient-weighted-Class Activation Mapping (Grad-CAM) and Occlusion Sensitivity were used for visualization through heatmaps, to highlight the relative importance of the image areas contributing to the DNN model's prediction, that can also help to understand the model's performance and bias. The results provide practical use case scenarios for utilizing the transfer learning approaches, model optimization and deployment to edge devices, and model visualizations in ecological research.