All-optical spiking neurosynaptic networks with self-learning capabilities.

Software implementations of brain-inspired computing underlie many important computational tasks, from image processing to speech recognition, artificial intelligence and deep learning applications. Yet, unlike real neural tissue, traditional computing architectures physically separate the core computing functions of memory and processing, making fast, efficient and low-energy computing difficult to achieve. To overcome such limitations, an attractive alternative is to design hardware that mimics neurons and synapses. Such hardware, when connected in networks or neuromorphic systems, processes information in a way more analogous to brains. Here we present an all-optical version of such a neurosynaptic system, capable of supervised and unsupervised learning. We exploit wavelength division multiplexing techniques to implement a scalable circuit architecture for photonic neural networks, successfully demonstrating pattern recognition directly in the optical domain. Such photonic neurosynaptic networks promise access to the high speed and high bandwidth inherent to optical systems, thus enabling the direct processing of optical telecommunication and visual data.

Deep learning and process understanding for data-driven Earth system science.

Machine learning approaches are increasingly used to extract patterns and insights from the ever-increasing stream of geospatial data, but current approaches may not be optimal when system behaviour is dominated by spatial or temporal context. Here, rather than amending classical machine learning, we argue that these contextual cues should be used as part of deep learning (an approach that is able to extract spatio-temporal features automatically) to gain further process understanding of Earth system science problems, improving the predictive ability of seasonal forecasting and modelling of long-range spatial connections across multiple timescales, for example. The next step will be a hybrid modelling approach, coupling physical process models with the versatility of data-driven machine learning.

Integrating heterogeneous knowledge graphs into drug-drug interaction extraction from the literature.

MOTIVATION: Most of the conventional deep neural network-based methods for drug-drug interaction (DDI) extraction consider only context information around drug mentions in the text. However, human experts use heterogeneous background knowledge about drugs to comprehend pharmaceutical papers and extract relationships between drugs. Therefore, we propose a novel method that simultaneously considers various heterogeneous information for DDI extraction from the literature. RESULTS: We first construct drug representations by conducting the link prediction task on a heterogeneous pharmaceutical knowledge graph (KG) dataset. We then effectively combine the text information of input sentences in the corpus and the information on drugs in the heterogeneous KG (HKG) dataset. Finally, we evaluate our DDI extraction method on the DDIExtraction-2013 shared task dataset. In the experiment, integrating heterogeneous drug information significantly improves the DDI extraction performance, and we achieved an F-score of 85.40%, which results in state-of-the-art performance. We evaluated our method on the DrugProt dataset and improved the performance significantly, achieving an F-score of 77.9%. Further analysis showed that each type of node in the HKG contributes to the performance improvement of DDI extraction, indicating the importance of considering multiple pieces of information. AVAILABILITY AND IMPLEMENTATION: Our code is available at https://github.com/tticoin/HKG-DDIE.git.

Human activity recognition using a single-photon direct time-of-flight sensor.

Single-Photon Avalanche Diode (SPAD) direct Time-of-Flight (dToF) sensors provide depth imaging over long distances, enabling the detection of objects even in the absence of contrast in colour or texture. However, distant objects are represented by just a few pixels and are subject to noise from solar interference, limiting the applicability of existing computer vision techniques for high-level scene interpretation. We present a new SPAD-based vision system for human activity recognition, based on convolutional and recurrent neural networks, which is trained entirely on synthetic data. In tests using real data from a 64×32 pixel SPAD, captured over a distance of 40 m, the scheme successfully overcomes the limited transverse resolution (in which human limbs are approximately one pixel across), achieving an average accuracy of 89% in distinguishing between seven different activities. The approach analyses continuous streams of video-rate depth data at a maximal rate of 66 FPS when executed on a GPU, making it well-suited for real-time applications such as surveillance or situational awareness in autonomous systems.

A Comparison of Five Algorithmic Methods and Machine Learning Pattern Recognition for Artifact Detection in Electronic Records of Five Different Vital Signs: A Retrospective Analysis.

BACKGROUND: Research on electronic health record physiologic data is common, invariably including artifacts. Traditionally, these artifacts have been handled using simple filter techniques. The authors hypothesized that different artifact detection algorithms, including machine learning, may be necessary to provide optimal performance for various vital signs and clinical contexts. METHODS: In a retrospective single-center study, intraoperative operating room and intensive care unit (ICU) electronic health record datasets including heart rate, oxygen saturation, blood pressure, temperature, and capnometry were included. All records were screened for artifacts by at least two human experts. Classical artifact detection methods (cutoff, multiples of SD [z-value], interquartile range, and local outlier factor) and a supervised learning model implementing long short-term memory neural networks were tested for each vital sign against the human expert reference dataset. For each artifact detection algorithm, sensitivity and specificity were calculated. RESULTS: A total of 106 (53 operating room and 53 ICU) patients were randomly selected, resulting in 392,808 data points. Human experts annotated 5,167 (1.3%) data points as artifacts. The artifact detection algorithms demonstrated large variations in performance. The specificity was above 90% for all detection methods and all vital signs. The neural network showed significantly higher sensitivities than the classic methods for heart rate (ICU, 33.6%; 95% CI, 33.1 to 44.6), systolic invasive blood pressure (in both the operating room [62.2%; 95% CI, 57.5 to 71.9] and the ICU [60.7%; 95% CI, 57.3 to 71.8]), and temperature in the operating room (76.1%; 95% CI, 63.6 to 89.7). The CI for specificity overlapped for all methods. Generally, sensitivity was low, with only the z-value for oxygen saturation in the operating room reaching 88.9%. All other sensitivities were less than 80%. CONCLUSIONS: No single artifact detection method consistently performed well across different vital signs and clinical settings. Neural networks may be a promising artifact detection method for specific vital signs.

Scaling up molecular pattern recognition with DNA-based winner-take-all neural networks.

From bacteria following simple chemical gradients1 to the brain distinguishing complex odour information2, the ability to recognize molecular patterns is essential for biological organisms. This type of information-processing function has been implemented using DNA-based neural networks3, but has been limited to the recognition of a set of no more than four patterns, each composed of four distinct DNA molecules. Winner-take-all computation4 has been suggested5,6 as a potential strategy for enhancing the capability of DNA-based neural networks. Compared to the linear-threshold circuits7 and Hopfield networks8 used previously3, winner-take-all circuits are computationally more powerful4, allow simpler molecular implementation and are not constrained by the number of patterns and their complexity, so both a large number of simple patterns and a small number of complex patterns can be recognized. Here we report a systematic implementation of winner-take-all neural networks based on DNA-strand-displacement9,10 reactions. We use a previously developed seesaw DNA gate motif3,11,12, extended to include a simple and robust component that facilitates the cooperative hybridization13 that is involved in the process of selecting a 'winner'. We show that with this extended seesaw motif DNA-based neural networks can classify patterns into up to nine categories. Each of these patterns consists of 20 distinct DNA molecules chosen from the set of 100 that represents the 100 bits in 10 × 10 patterns, with the 20 DNA molecules selected tracing one of the handwritten digits '1' to '9'. The network successfully classified test patterns with up to 30 of the 100 bits flipped relative to the digit patterns 'remembered' during training, suggesting that molecular circuits can robustly accomplish the sophisticated task of classifying highly complex and noisy information on the basis of similarity to a memory.

An Improved k-Nearest Neighbor Algorithm for Recognition and Classification of Thyroid Nodules.

OBJECTIVES: To complete the task of automatic recognition and classification of thyroid nodules and solve the problem of high classification error rates when the samples are imbalanced. METHODS: An improved k-nearest neighbor (KNN) algorithm is proposed and a method for automatic thyroid nodule classification based on the improved KNN algorithm is established. In the improved KNN algorithm, we consider not only the number of class labels for various classes of data in KNNs, but also the corresponding weights. And we use the Minkowski distance measure instead of the Euclidean distance measure. RESULTS: A total of 508 ultrasound images of thyroid nodules, including 415 benign nodules and 93 malignant nodules, were used in the paper. Experimental results show the improved KNN has 0.872549 accuracy, 0.867347 precision, 1 recall, and 0.928962 F1-score. At the same time, we also considered the influence of different distance weights, the value of k, different distance measures on the classification results. CONCLUSIONS: A comparison result shows that our method has a better performance than the traditional KNN and other classical machine learning methods.

MeshID: Few-Shot Finger Gesture Based User Identification Using Orthogonal Signal Interference.

Radio frequency (RF) technology has been applied to enable advanced behavioral sensing in human-computer interaction. Due to its device-free sensing capability and wide availability on Internet of Things devices. Enabling finger gesture-based identification with high accuracy can be challenging due to low RF signal resolution and user heterogeneity. In this paper, we propose MeshID, a novel RF-based user identification scheme that enables identification through finger gestures with high accuracy. MeshID significantly improves the sensing sensitivity on RF signal interference, and hence is able to extract subtle individual biometrics through velocity distribution profiling (VDP) features from less-distinct finger motions such as drawing digits in the air. We design an efficient few-shot model retraining framework based on first component reverse module, achieving high model robustness and performance in a complex environment. We conduct comprehensive real-world experiments and the results show that MeshID achieves a user identification accuracy of 95.17% on average in three indoor environments. The results indicate that MeshID outperforms the state-of-the-art in identification performance with less cost.

Nonlinear Regularization Decoding Method for Speech Recognition.

Existing end-to-end speech recognition methods typically employ hybrid decoders based on CTC and Transformer. However, the issue of error accumulation in these hybrid decoders hinders further improvements in accuracy. Additionally, most existing models are built upon Transformer architecture, which tends to be complex and unfriendly to small datasets. Hence, we propose a Nonlinear Regularization Decoding Method for Speech Recognition. Firstly, we introduce the nonlinear Transformer decoder, breaking away from traditional left-to-right or right-to-left decoding orders and enabling associations between any characters, mitigating the limitations of Transformer architectures on small datasets. Secondly, we propose a novel regularization attention module to optimize the attention score matrix, reducing the impact of early errors on later outputs. Finally, we introduce the tiny model to address the challenge of overly large model parameters. The experimental results indicate that our model demonstrates good performance. Compared to the baseline, our model achieves recognition improvements of 0.12%, 0.54%, 0.51%, and 1.2% on the Aishell1, Primewords, Free ST Chinese Corpus, and Common Voice 16.1 datasets of Uyghur, respectively.

Exploring the Impact of the NULL Class on In-the-Wild Human Activity Recognition.

Monitoring activities of daily living (ADLs) plays an important role in measuring and responding to a person's ability to manage their basic physical needs. Effective recognition systems for monitoring ADLs must successfully recognize naturalistic activities that also realistically occur at infrequent intervals. However, existing systems primarily focus on either recognizing more separable, controlled activity types or are trained on balanced datasets where activities occur more frequently. In our work, we investigate the challenges associated with applying machine learning to an imbalanced dataset collected from a fully in-the-wild environment. This analysis shows that the combination of preprocessing techniques to increase recall and postprocessing techniques to increase precision can result in more desirable models for tasks such as ADL monitoring. In a user-independent evaluation using in-the-wild data, these techniques resulted in a model that achieved an event-based F1-score of over 0.9 for brushing teeth, combing hair, walking, and washing hands. This work tackles fundamental challenges in machine learning that will need to be addressed in order for these systems to be deployed and reliably work in the real world.

Using Graphs to Perform Effective Sensor-Based Human Activity Recognition in Smart Homes.

There has been a resurgence of applications focused on human activity recognition (HAR) in smart homes, especially in the field of ambient intelligence and assisted-living technologies. However, such applications present numerous significant challenges to any automated analysis system operating in the real world, such as variability, sparsity, and noise in sensor measurements. Although state-of-the-art HAR systems have made considerable strides in addressing some of these challenges, they suffer from a practical limitation: they require successful pre-segmentation of continuous sensor data streams prior to automated recognition, i.e., they assume that an oracle is present during deployment, and that it is capable of identifying time windows of interest across discrete sensor events. To overcome this limitation, we propose a novel graph-guided neural network approach that performs activity recognition by learning explicit co-firing relationships between sensors. We accomplish this by learning a more expressive graph structure representing the sensor network in a smart home in a data-driven manner. Our approach maps discrete input sensor measurements to a feature space through the application of attention mechanisms and hierarchical pooling of node embeddings. We demonstrate the effectiveness of our proposed approach by conducting several experiments on CASAS datasets, showing that the resulting graph-guided neural network outperforms the state-of-the-art method for HAR in smart homes across multiple datasets and by large margins. These results are promising because they push HAR for smart homes closer to real-world applications.

Pedestrian Pose Recognition Based on Frequency-Modulated Continuous-Wave Radar with Meta-Learning.

With the continuous advancement of autonomous driving and monitoring technologies, there is increasing attention on non-intrusive target monitoring and recognition. This paper proposes an ArcFace SE-attention model-agnostic meta-learning approach (AS-MAML) by integrating attention mechanisms into residual networks for pedestrian gait recognition using frequency-modulated continuous-wave (FMCW) millimeter-wave radar through meta-learning. We enhance the feature extraction capability of the base network using channel attention mechanisms and integrate the additive angular margin loss function (ArcFace loss) into the inner loop of MAML to constrain inner loop optimization and improve radar discrimination. Then, this network is used to classify small-sample micro-Doppler images obtained from millimeter-wave radar as the data source for pose recognition. Experimental tests were conducted on pose estimation and image classification tasks. The results demonstrate significant detection and recognition performance, with an accuracy of 94.5%, accompanied by a 95% confidence interval. Additionally, on the open-source dataset DIAT-µRadHAR, which is specially processed to increase classification difficulty, the network achieves a classification accuracy of 85.9%.

ST-TGR: Spatio-Temporal Representation Learning for Skeleton-Based Teaching Gesture Recognition.

Teaching gesture recognition is a technique used to recognize the hand movements of teachers in classroom teaching scenarios. This technology is widely used in education, including for classroom teaching evaluation, enhancing online teaching, and assisting special education. However, current research on gesture recognition in teaching mainly focuses on detecting the static gestures of individual students and analyzing their classroom behavior. To analyze the teacher's gestures and mitigate the difficulty of single-target dynamic gesture recognition in multi-person teaching scenarios, this paper proposes skeleton-based teaching gesture recognition (ST-TGR), which learns through spatio-temporal representation. This method mainly uses the human pose estimation technique RTMPose to extract the coordinates of the keypoints of the teacher's skeleton and then inputs the recognized sequence of the teacher's skeleton into the MoGRU action recognition network for classifying gesture actions. The MoGRU action recognition module mainly learns the spatio-temporal representation of target actions by stacking a multi-scale bidirectional gated recurrent unit (BiGRU) and using improved attention mechanism modules. To validate the generalization of the action recognition network model, we conducted comparative experiments on datasets including NTU RGB+D 60, UT-Kinect Action3D, SBU Kinect Interaction, and Florence 3D. The results indicate that, compared with most existing baseline models, the model proposed in this article exhibits better performance in recognition accuracy and speed.

Dynamic Japanese Sign Language Recognition Throw Hand Pose Estimation Using Effective Feature Extraction and Classification Approach.

Japanese Sign Language (JSL) is vital for communication in Japan's deaf and hard-of-hearing community. But probably because of the large number of patterns, 46 types, there is a mixture of static and dynamic, and the dynamic ones have been excluded in most studies. Few researchers have been working to develop a dynamic JSL alphabet, and their performance accuracy is unsatisfactory. We proposed a dynamic JSL recognition system using effective feature extraction and feature selection approaches to overcome the challenges. In the procedure, we follow the hand pose estimation, effective feature extraction, and machine learning techniques. We collected a video dataset capturing JSL gestures through standard RGB cameras and employed MediaPipe for hand pose estimation. Four types of features were proposed. The significance of these features is that the same feature generation method can be used regardless of the number of frames or whether the features are dynamic or static. We employed a Random forest (RF) based feature selection approach to select the potential feature. Finally, we fed the reduced features into the kernels-based Support Vector Machine (SVM) algorithm classification. Evaluations conducted on our proprietary newly created dynamic Japanese sign language alphabet dataset and LSA64 dynamic dataset yielded recognition accuracies of 97.20% and 98.40%, respectively. This innovative approach not only addresses the complexities of JSL but also holds the potential to bridge communication gaps, offering effective communication for the deaf and hard-of-hearing, and has broader implications for sign language recognition systems globally.

A Framework for Real-Time Gestural Recognition and Augmented Reality for Industrial Applications.

In recent decades, technological advancements have transformed the industry, highlighting the efficiency of automation and safety. The integration of augmented reality (AR) and gesture recognition has emerged as an innovative approach to create interactive environments for industrial equipment. Gesture recognition enhances AR applications by allowing intuitive interactions. This study presents a web-based architecture for the integration of AR and gesture recognition, designed to interact with industrial equipment. Emphasizing hardware-agnostic compatibility, the proposed structure offers an intuitive interaction with equipment control systems through natural gestures. Experimental validation, conducted using Google Glass, demonstrated the practical viability and potential of this approach in industrial operations. The development focused on optimizing the system's software and implementing techniques such as normalization, clamping, conversion, and filtering to achieve accurate and reliable gesture recognition under different usage conditions. The proposed approach promotes safer and more efficient industrial operations, contributing to research in AR and gesture recognition. Future work will include improving the gesture recognition accuracy, exploring alternative gestures, and expanding the platform integration to improve the user experience.

Human Action Recognition and Note Recognition: A Deep Learning Approach Using STA-GCN.

Human action recognition (HAR) is growing in machine learning with a wide range of applications. One challenging aspect of HAR is recognizing human actions while playing music, further complicated by the need to recognize the musical notes being played. This paper proposes a deep learning-based method for simultaneous HAR and musical note recognition in music performances. We conducted experiments on Morin khuur performances, a traditional Mongolian instrument. The proposed method consists of two stages. First, we created a new dataset of Morin khuur performances. We used motion capture systems and depth sensors to collect data that includes hand keypoints, instrument segmentation information, and detailed movement information. We then analyzed RGB images, depth images, and motion data to determine which type of data provides the most valuable features for recognizing actions and notes in music performances. The second stage utilizes a Spatial Temporal Attention Graph Convolutional Network (STA-GCN) to recognize musical notes as continuous gestures. The STA-GCN model is designed to learn the relationships between hand keypoints and instrument segmentation information, which are crucial for accurate recognition. Evaluation on our dataset demonstrates that our model outperforms the traditional ST-GCN model, achieving an accuracy of 81.4%.

Effects of Training and Calibration Data on Surface Electromyogram-Based Recognition for Upper Limb Amputees.

Surface electromyogram (sEMG)-based gesture recognition has emerged as a promising avenue for developing intelligent prostheses for upper limb amputees. However, the temporal variations in sEMG have rendered recognition models less efficient than anticipated. By using cross-session calibration and increasing the amount of training data, it is possible to reduce these variations. The impact of varying the amount of calibration and training data on gesture recognition performance for amputees is still unknown. To assess these effects, we present four datasets for the evaluation of calibration data and examine the impact of the amount of training data on benchmark performance. Two amputees who had undergone amputations years prior were recruited, and seven sessions of data were collected for analysis from each of them. Ninapro DB6, a publicly available database containing data from ten healthy subjects across ten sessions, was also included in this study. The experimental results show that the calibration data improved the average accuracy by 3.03%, 6.16%, and 9.73% for the two subjects and Ninapro DB6, respectively, compared to the baseline results. Moreover, it was discovered that increasing the number of training sessions was more effective in improving accuracy than increasing the number of trials. Three potential strategies are proposed in light of these findings to enhance cross-session models further. We consider these findings to be of the utmost importance for the commercialization of intelligent prostheses, as they demonstrate the criticality of gathering calibration and cross-session training data, while also offering effective strategies to maximize the utilization of the entire dataset.

Electroencephalogram-Based Facial Gesture Recognition Using Self-Organizing Map.

Brain-computer interfaces (BCIs) allow information to be transmitted directly from the human brain to a computer, enhancing the ability of human brain activity to interact with the environment. In particular, BCI-based control systems are highly desirable because they can control equipment used by people with disabilities, such as wheelchairs and prosthetic legs. BCIs make use of electroencephalograms (EEGs) to decode the human brain's status. This paper presents an EEG-based facial gesture recognition method based on a self-organizing map (SOM). The proposed facial gesture recognition uses α, ß, and θ power bands of the EEG signals as the features of the gesture. The SOM-Hebb classifier is utilized to classify the feature vectors. We utilized the proposed method to develop an online facial gesture recognition system. The facial gestures were defined by combining facial movements that are easy to detect in EEG signals. The recognition accuracy of the system was examined through experiments. The recognition accuracy of the system ranged from 76.90% to 97.57% depending on the number of gestures recognized. The lowest accuracy (76.90%) occurred when recognizing seven gestures, though this is still quite accurate when compared to other EEG-based recognition systems. The implemented online recognition system was developed using MATLAB, and the system took 5.7 s to complete the recognition flow.

Hand Gesture Recognition Based on High-Density Myoelectricity in Forearm Flexors in Humans.

Electromyography-based gesture recognition has become a challenging problem in the decoding of fine hand movements. Recent research has focused on improving the accuracy of gesture recognition by increasing the complexity of network models. However, training a complex model necessitates a significant amount of data, thereby escalating both user burden and computational costs. Moreover, owing to the considerable variability of surface electromyography (sEMG) signals across different users, conventional machine learning approaches reliant on a single feature fail to meet the demand for precise gesture recognition tailored to individual users. Therefore, to solve the problems of large computational cost and poor cross-user pattern recognition performance, we propose a feature selection method that combines mutual information, principal component analysis and the Pearson correlation coefficient (MPP). This method can filter out the optimal subset of features that match a specific user while combining with an SVM classifier to accurately and efficiently recognize the user's gesture movements. To validate the effectiveness of the above method, we designed an experiment including five gesture actions. The experimental results show that compared to the classification accuracy obtained using a single feature, we achieved an improvement of about 5% with the optimally selected feature as the input to any of the classifiers. This study provides an effective guarantee for user-specific fine hand movement decoding based on sEMG signals.

Dual Stream Long Short-Term Memory Feature Fusion Classifier for Surface Electromyography Gesture Recognition.

Gesture recognition using electromyography (EMG) signals has prevailed recently in the field of human-computer interactions for controlling intelligent prosthetics. Currently, machine learning and deep learning are the two most commonly employed methods for classifying hand gestures. Despite traditional machine learning methods already achieving impressive performance, it is still a huge amount of work to carry out feature extraction manually. The existing deep learning methods utilize complex neural network architectures to achieve higher accuracy, which will suffer from overfitting, insufficient adaptability, and low recognition accuracy. To improve the existing phenomenon, a novel lightweight model named dual stream LSTM feature fusion classifier is proposed based on the concatenation of five time-domain features of EMG signals and raw data, which are both processed with one-dimensional convolutional neural networks and LSTM layers to carry out the classification. The proposed method can effectively capture global features of EMG signals using a simple architecture, which means less computational cost. An experiment is conducted on a public DB1 dataset with 52 gestures, and each of the 27 subjects repeats every gesture 10 times. The accuracy rate achieved by the model is 89.66%, which is comparable to that achieved by more complex deep learning neural networks, and the inference time for each gesture is 87.6 ms, which can also be implied in a real-time control system. The proposed model is validated using a subject-wise experiment on 10 out of the 40 subjects in the DB2 dataset, achieving a mean accuracy of 91.74%. This is illustrated by its ability to fuse time-domain features and raw data to extract more effective information from the sEMG signal and select an appropriate, efficient, lightweight network to enhance the recognition results.