The alignment of Small Language Models (SLMs) in the 70–500M parameter range using reinforcement learning is often considered unstable, though the underlying failure mechanisms have not been systematically investigated. In the State-of-the-Art (SOTA) research, fifteen (model, corpus) configurations were trained using Proximal Policy Optimization (PPO). The experiments included Pythia-70M, 160M, 410M and SmolLM2-135M, 360M on the TinyStories, CNN/DailyMail, and Wikitext-103 corpora. Three reproducible failure modes were identified in small-scale language models: silent LoRA parameter freezing in standard PEFT/TRL pipelines, numerical overflow in importance ratios when using bfloat16, and catastrophic policy collapse due to reward-model error. These issues were addressed using a merge-and-reinitialize adapter technique, float32 precision during PPO updates, and a three-layer safety mechanism comprising reward whitening, importance-ratio guarding, and weight rollback. In this paper, a capacity-headroom hypothesis is proposed, which states that PPO performance at the SLM scale depends on both a fluent supervised model (PPL<20) and a discriminative reward signal, rather than on the number of model parameters. The proposed system converged stably in all experiments and improved preference win rate over the SFT baseline in configurations with a fluent prior and an informative reward signal. Furthermore, it outperformed instruction-tuned baselines while requiring significantly less training data. All checkpoints, preference datasets, and training scripts are publicly released^§.

We propose CAST, a dual-stream architecture that utilizes channel-aware spatial transfer learning for isolated sign language recognition addressing the challenges of magnitude-only 60 GHz radar Range-Time Maps (RTM). The proposed framework combines three physics-aware architectures with pretrained vision backbones, which operate under radar-only constraints across clinical and alphabetical gestures. First, an explicit decibel-to-linear inversion is combined with a windowed fast Fourier transform that extracts Cadence Velocity Diagrams (CVD) while avoiding the harmonic artifacts that arise from the spectral analysis of log-compressed signals. Second, a cross-antenna spatial attention module applies attention to raw antenna channels before the convolution, preserving inter-receiver amplitude covariance. Third, an asymmetric cross-attention mechanism fuses representations from parallel ConvNeXt-Tiny (CVD) and EfficientNetV2-S (RTM) backbones. Extensive experiments reveal that the architecture achieves a Top-1 accuracy of 80.5

Foundation models (FMs), powered by self-supervised learning (SSL), have redefined the capabilities of artificial intelligence, demonstrating exceptional performance in domains like natural language processing and computer vision. These advances present a transformative opportunity for brain signal analysis. Unlike traditional supervised learning, which is limited by the scarcity of labeled neural data, SSL offers a promising solution by enabling models to learn meaningful representations from unlabeled data. This is particularly valuable in addressing the unique challenges of brain signals, including high noise levels, inter-subject variability, and low signal-to-noise ratios. This survey systematically reviews the emerging field of bridging brain signals with foundation models through the innovative application of SSL. It explores key SSL techniques, the development of brain-specific foundation models, their adaptation to downstream tasks, and the integration of brain signals with other modalities in multimodal SSL frameworks. The review also covers commonly used evaluation metrics and benchmark datasets that support comparative analysis. Finally, it highlights key challenges and outlines future research directions. This work aims to provide researchers with a structured understanding of this rapidly evolving field and a roadmap for developing generalizable brain foundation models powered by self-supervision.

Depression is a serious mental health illness that significantly affects an individual’s well-being and quality of life, making early detection crucial for adequate care and treatment. Detecting depression is often difficult, as it is based primarily on subjective evaluations during clinical interviews. Hence, the early diagnosis of depression, thanks to the content of social networks, has become a prominent research area. The extensive and diverse nature of user-generated information poses a significant challenge, limiting the accurate extraction of relevant temporal information and the effective fusion of data across multiple modalities. This paper introduces MMF-former, a multimodal depression detection network designed to retrieve depressive spatio-temporal high-level patterns from multimodal social media information. The transformer network with residual connections captures spatial features from videos, and a transformer encoder is exploited to design important temporal dynamics in audio. Moreover, the fusion architecture fused the extracted features through late and intermediate fusion strategies to find out the most relevant intermodal correlations among them. Finally, the proposed network is assessed on two large-scale depression detection datasets, and the results clearly reveal that it surpasses existing state-of-the-art approaches, improving the F1-Score by 13.92% for D-Vlog dataset and 7.74% for LMVD dataset. The code is made available publicly at https://github.com/rezwanh001/Large-Scale-Multimodal-Depression-Detection.

Automatic emotion recognition has attracted significant interest in healthcare, thanks to remarkable developments made recently in smart and innovative technologies. A real-time emotion recognition system allows for continuous monitoring, comprehension, and enhancement of the physical entity's capacities, along with continuing advice for enhancing quality of life and well-being in the context of personalized healthcare. Multimodal emotion recognition presents a significant challenge in terms of efficiently using the diverse modalities present in the data. In this article, we introduce a Multi-Stage Fusion Network (MSF-Net) for emotion recognition capable of extracting multimodal information and achieving significant performances. We propose utilizing the transformer-based structure to extract deep features from facial expressions. We exploited two visual descriptors, local binary pattern and Oriented FAST and Rotated BRIEF, to retrieve the computer vision- based features from the facial videos. A feature-level fusion network integrates the extraction of features from these modules, directing the output into the triplet attention technique. This module employs a three-branch architecture to compute attention weights to capture cross-dimensional interactions efficiently. The temporal dependencies in physiological signals are modeled by a Bi-directional Gated Recurrent Unit (Bi-GRU) in forward and backward directions at each time step. Lastly, the output feature representations from the triplet attention module and the extracted high-level patterns from Bi-GRU are fused and fed into the classification module to recognize emotion. The extensive experimental evaluations revealed that the proposed MSF-Net outperformed the state-of-the-art approaches on two popular datasets, BioVid Emo DB and MGEED. Finally, we tested the proposed MSF-Net in the Internet of Things environment to facilitate real-world scalable smart healthcare application.

Continuous Sign Language Recognition (CSLR) faces multiple challenges, including significant inter-signer variability and poor generalization to novel sentence structures. Traditional solutions frequently fail to handle these issues efficiently. For overcoming these constraints, we propose a dual-architecture framework. For the Signer-Independent (SI) challenge, we propose a Signer-Invariant Conformer that combines convolutions with multi-head self-attention to learn robust, signer-agnostic representations from pose-based skeletal keypoints. For the Unseen-Sentences (US) task, we designed a Multi-Scale Fusion Transformer with a novel dual-path temporal encoder that captures both fine-grained posture dynamics, enabling the model's ability to comprehend novel grammatical compositions. Experiments on the challenging Isharah-1000 dataset establish a new standard for both CSLR benchmarks. The proposed conformer architecture achieves a Word Error Rate (WER) of 13.07% on the SI challenge, a reduction of 13.53% from the state-of-the-art. On the US task, the transformer model scores a WER of 47.78%, surpassing previous work. In the SignEval 2025 CSLR challenge, our team placed 2nd in the US task and 4th in the SI task, demonstrating the performance of these models. The findings validate our key hypothesis: that developing task-specific networks designed for the particular challenges of CSLR leads to considerable performance improvements and establishes a new baseline for further research. The source code is available at: https://github.com/rezwanh001/MSLR-Pose86K-CSLR-Isharah.

Accurate recognition of sign language in healthcare communication poses a significant challenge, requiring frameworks that can accurately interpret complex multimodal gestures. To deal with this, we propose FusionEnsembleNet, a novel attention-based ensemble of spatiotemporal networks that dynamically fuses visual and motion data to enhance recognition accuracy. The proposed approach processes RGB video and range Doppler map radar modalities synchronously through four different spatiotemporal networks. For each network, features from both modalities are continuously fused using an attention-based fusion module before being fed into an ensemble of classifiers. Finally, the outputs of these four different fused channels are combined in an ensemble classification head, thereby enhancing the model's robustness. Experiments demonstrate that FusionEnsembleNet outperforms state-of-the-art approaches with a test accuracy of 99.44% on the large-scale MultiMeDaLIS dataset for Italian Sign Language. Our findings indicate that an ensemble of diverse spatiotemporal networks, unified by attention-based fusion, yields a robust and accurate framework for complex, multimodal isolated gesture recognition tasks. The source code is available at: https://github.com/rezwanh001/Multimodal-Isolated-Italian-Sign-Language-Recognition.

Microscopic assessment of histopathology images is vital for accurate cancer diagnosis and treatment. Whole Slide Image (WSI) classification and captioning have become crucial tasks in computer-aided pathology. However, microscopic WSIs face challenges such as redundant patches and unknown patch positions due to subjective pathologist captures. Moreover, generating automatic pathology captions remains a significant challenge. To address these challenges, a novel GNN-ViTCap framework is introduced for classification and caption generation from histopathological microscopic images. A visual feature extractor is used to extract feature embeddings. The redundant patches are then removed by dynamically clustering images using deep embedded clustering and extracting representative images through a scalar dot attention mechanism. The graph is formed by constructing edges from the similarity matrix, connecting each node to its nearest neighbors. Therefore, a graph neural network is utilized to extract and represent contextual information from both local and global areas. The aggregated image embeddings are then projected into the language model's input space using a linear layer and combined with input caption tokens to fine-tune the large language models for caption generation. Our proposed method is validated using the BreakHis and PatchGastric microscopic datasets. The GNN-ViTCap method achieves an F-1-Score of 0.934 and AUC of 0.963 for classification, along with BLEU@4 = 0.811 and METEOR = 0.569 for captioning. Experimental analysis demonstrates that the GNN-ViTCap architecture outperforms state-of-the-art (SOTA) approaches, providing a reliable and efficient approach for patient diagnosis using microscopy images.

Depression is a major mental health condition that severely impacts the emotional and physical well-being of individuals. The simple nature of data collection from social media platforms has attracted significant interest in properly utilizing this information for mental health research. A Multimodal Depression Detection Network (MDD-Net), utilizing acoustic and visual data obtained from social media networks, is proposed in this work where mutual transformers are exploited to efficiently extract and fuse multimodal features for efficient depression detection. The MDD-Net consists of four core modules: an acoustic feature extraction module for retrieving relevant acoustic attributes, a visual feature extraction module for extracting significant high-level patterns, a mutual transformer for computing the correlations among the generated features and fusing these features from multiple modalities, and a detection layer for detecting depression using the fused feature representations. The extensive experiments are performed using the multimodal D-Vlog dataset, and the findings reveal that the developed multimodal depression detection network surpasses the state-of-the-art by up to 17.37% for F1-Score, demonstrating the greater performance of the proposed system. The source code is accessible at https://github.com/rezwanh001/Multimodal-Depression-Detection.

Captioning Whole Slide Images (WSIs) for pathological analysis is an essential but not extensively explored aspect of computer-aided pathological diagnosis. Challenges arise from insufficient datasets and the effectiveness of model training. Generating automatic caption reports for various gastric adenocarcinoma images is another challenge. In this paper, we introduce a hybrid method referred to as TransUAAE-CapGen to generate histopathological captions from WSI patches. The TransUAAE-CapGen architecture consists of a hybrid UNet-based Advereasrial Autoencoder (AAE) for feature extraction and a transformer for caption generation. The hybrid UNet-based AAE extracted complex tissue properties from histopathological patches, transforming them into low-dimensional embeddings. The embeddings are then fed into the transformer to generate concise captions. Our proposed method is validated using the PatchGastricADC22 dataset. The TransUAAE-CapGen model provides the best estimated accuracy of BLEU-4 = 86.8%, METEOR = 59.6%, a ROUGE = 89.3%, and CIDEr = 7.72%. Experimental analysis indicates that the TransUAAE-CapGen architecture outperforms the traditional LSTM-based model for the caption generation task. Our findings reveal that the proposed architecture can effectively generate accurate and precise reports for medical image analysis.

Deep learning techniques have drawn considerable interest in emotion recognition due to recent technological developments in healthcare analytics. Automatic patient emotion recognition can assist healthcare analytics by providing feedback to the stakeholders of competent healthcare about the conditions of the patients and their satisfaction levels. In this paper, we propose a novel model-level fusion technique based on deep learning for enhanced emotion recognition from multimodal signals to monitor patients in connected healthcare. The representative visual features from the video signals are extracted through the Depthwise Separable Convolution Neural Network, and the optimized temporal attributes are derived from the multiple physiological data utilizing Bi-directional Long Short-Term Memory. A soft attention method fused the high multimodal features obtained from the two data modalities to retrieve the most significant features by focusing on emotionally salient parts of the features. We exploited two face detection methods, Histogram of Oriented Gradients and Convolutional Neural Network-based face detector (ResNet-34), to observe the effects of facial features on emotion recognition. Lastly, extensive experimental evaluations have been conducted using the widely used Bio Vid Emo DB multimodal dataset to verify the performance of the proposed architecture. Experimental results show that the developed fusion architecture improved the accuracy of emotion recognition from multimodal signals and outperformed the performance of both state-of-the-art techniques and baseline methods.

Human Activity Recognition (HAR) is the process of automatic recognition of Activities of Daily Living (ADL) from human motion data captured in various data modalities by wearable and ambient sensors. Advances in Deep Learning, especially Convolutional Neural Networks (CNNs) and sequential models have revolutionalized HAR from video data sources. Although these models are incredibly accurate and utilize both spatial and temporal information, they require huge computation and memory resources — making them unsuitable for edge or wearable applications. Tiny Machine Learning (TinyML) aims to optimize these models in terms of compute and memory requirements – aiming to make them suitable for always-on resource constrained devices – leading to a reduction in communication latency and network traffic for HAR frameworks. In this paper, we propose a two-stream multi-resolution fusion architecture for HAR from video data modality. The context stream takes a resized image as input and the fovea stream takes the cropped center portion of the resized image as input, reducing the overall dimensionality. We tested two quantization methods: Post-Training Quantization (PTQ) and Quantization Aware Training (QAT) to optimize these models for deployment in edge devices and tested the performance in two challenging video datasets: KTH and UCF11. We performed ablation studies to validate the two-stream model performance. We deployed the proposed architecture in commercial resource constrained devices and monitored their performance in terms of inference latency and power consumption. The results indicate that the proposed architecture clearly outperforms other relevant single-stream models tested in this work in terms of accuracy, precision, recall, and F1-Score while also reducing the overall model size.

Human Activity Recognition (HAR) has become a crucial element for smart healthcare applications due to the fast adoption of wearable sensors and mobile technologies. Most of the existing human activity recognition frameworks deal with a single modality of data that degrades the reliability and recognition accuracy of the system for heterogeneous data sources. In this article, we propose a multi-level feature fusion technique for multimodal human activity recognition using multi-head Convolutional Neural Network (CNN) with Convolution Block Attention Module (CBAM) to process the visual data and Convolutional Long Short Term Memory (ConvLSTM) for dealing with the time-sensitive multi-source sensor information. The architecture is developed to be able to analyze and retrieve channel and spatial dimension features through the use of three branches of CNN along with CBAM for visual information. The ConvLSTM network is designed to capture temporal features from the multiple sensors' time-series data for efficient activity recognition. An open-access multimodal HAR dataset named UP-Fall detection dataset is utilized in experiments and evaluations to measure the performance of the developed fusion architecture. Finally, we deployed an Internet of Things (IoT) system to test the proposed fusion network in real-world smart healthcare application scenarios. The findings from the experimental results reveal that the developed multimodal HAR framework surpasses the existing state-of-the-art methods in terms of multiple performance metrics.

Nowadays, the Internet has spread to practically every country around the world and is having unprecedented effects on people's lives. The Internet of Things (IoT) is getting more popular and has a high level of interest in both practitioners and academicians in the age of wireless communication due to its diverse applications. The IoT is a technology that enables everyday things to become savvier, everyday computation towards becoming intellectual, and everyday communication to become a little more insightful. In this paper, the most common and popular IoT device capabilities, architectures, and protocols are demonstrated in brief to provide a clear overview of the IoT technology to the researchers in this area. The common IoT device capabilities including hardware (Raspberry Pi, Arduino, and ESP8266) and software (operating systems, and built-in tools) platforms are described in detail. The widely used architectures that have been recently evolved and used are the three-layer architecture, SOA-based architecture, and middleware-based architecture. The popular protocols for IoT are demonstrated which include CoAP, MQTT, XMPP, AMQP, DDS, LoWPAN, BLE, and Zigbee that are frequently utilized to develop smart IoT applications. Additionally, this research provides an in-depth overview of the potential healthcare applications based on IoT technologies in the context of addressing various healthcare concerns. Finally, this paper summarizes state-of-the-art knowledge, highlights open issues and shortcomings, and provides recommendations for further studies which would be quite beneficial to anyone with a desire to work in this field and make breakthroughs to get expertise in this area.

Human Activity Recognition (HAR) plays a significant role in the everyday life of people because of its ability to learn extensive high-level information about human activity from wearable or stationary devices. A substantial amount of research has been conducted on HAR and numerous approaches based on deep learning have been exploited by the research community to classify human activities. The main goal of this review is to summarize recent works based on a wide range of deep neural networks architecture, namely convolutional neural networks (CNNs) for human activity recognition. The reviewed systems are clustered into four categories depending on the use of input devices like multimodal sensing devices, smartphones, radar, and vision devices. This review describes the performances, strengths, weaknesses, and the used hyperparameters of CNN architectures for each reviewed system with an overview of available public data sources. In addition, a discussion of the current challenges to CNN-based HAR systems is presented. Finally, this review is concluded with some potential future directions that would be of great assistance for the researchers who would like to contribute to this field. We conclude that CNN-based approaches are suitable for effective and accurate human activity recognition system applications despite challenges including availability of data regarding composite or group activities, high computational resource requirements, data privacy concerns, and edge computing limitations. For widespread adaptation, future research should be focused on more efficient edge computing techniques, datasets incorporating contextual information with activities, more explainable methodologies, and more robust systems.

AbstractThe confrontation of COVID-19 pandemic has become one of the promising challenges of the world healthcare. Accurate and fast diagnosis of COVID-19 cases is essential for correct medical treatment to control this pandemic. Compared with the reverse-transcription polymerase chain reaction (RT-PCR) method, chest radiography imaging techniques are shown to be more effective to detect coronavirus. For the limitation of available medical images, transfer learning is better suited to classify patterns in medical images. This paper presents a combined architecture of convolutional neural network (CNN) and recurrent neural network (RNN) to diagnose COVID-19 from chest X-rays. The deep transfer techniques used in this experiment are VGG19, DenseNet121, InceptionV3, and Inception-ResNetV2. CNN is used to extract complex features from samples and classified them using RNN. The VGG19-RNN architecture achieved the best performance among all the networks in terms of accuracy in our experiments. Finally, Gradient-weighted Class Activation Mapping (Grad-CAM) was used to visualize class-specific regions of images that are responsible to make decision. The system achieved promising results compared to other existing systems and might be validated in the future when more samples would be available. The experiment demonstrated a good alternative method to diagnose COVID-19 for medical staff.

Human Activity Recognition (HAR) has emerged as a potential research topic for smart healthcare owing to the fast growth of wearable and smart devices in recent years. The significant applications of HAR in ambient assisted living environments include monitoring the daily activities of elderly and cognitively impaired individuals to assist them by observing their health status. In this research, we present a deep learning-based fusion approach for multimodal HAR that fuses the different modalities of data to obtain robust outcomes. Here, Convolutional Neural Networks (CNNs) retrieve the high-level attributes from the image data, and the Convolutional Long Short Term Memory (ConvLSTM) is utilized to capture significant patterns from the multi-sensory data. Finally, the extracted features from the modalities are fused through self-attention mechanisms that enhance the relevant activity data and inhibit the superfluous and possibly confusing information by measuring their compatibility. Lastly, extensive tests have been performed to measure the efficiency and robustness of the developed fusion approach using the UP-Fall detection dataset. It is evident from the experimental findings that the proposed fusion technique outperforms the existing state-of-the-art and achieves relatively better performance.

Advances in deep learning, especially Convolutional Neural Networks (CNNs) have revolutionized intelligent frame-works such as Human Activity Recognition (HAR) systems by effectively and efficiently inferring human activity from various modalities of data. However, the training and inference of CNNs are often resource-intensive. Recent research developments are focused on bringing the effectiveness of CNNs in resource con-strained edge devices through Tiny Machine Learning (TinyML). However, this is extremely hard to achieve due to the limitations in memory, compute power, and energy of resource constrained edge devices. This paper provides a benchmark to understand these trade-offs among variations of CNN network architectures, different training methodologies, and different modalities of data in the context of HAR, TinyML, and edge devices. We tested and reported the performance of CNN and Depthwise Separable CNN (DSCNN) models as well as two training methodologies: Quantization Aware Training (QAT) and Post-training Quantization (PTQ) on five commonly used benchmark datasets containing image and time-series data: UP-Fall, Fall Detection Dataset (FDD), PAMAP2, UCI-HAR, and WISDM. We also deployed and tested the performance of the model-based standalone applications on multiple commonly available resource constrained edge devices in terms of inference time and power consumption. The experimental results demonstrate the effectiveness and feasibility of Tiny ML for HAR in edge devices.

This paper aims to introduce a robust framework for forecasting demand, including data preprocessing, data transformation and standardization, feature selection, cross-validation, and regression ensemble framework. Bagging (random forest regression (RFR)), boosting (gradient boosting regression (GBR) and extreme gradient boosting regression (XGBR)), and stacking (STACK) are employed as ensemble models. Different machine learning (ML) approaches, including support vector regression (SVR), extreme learning machine (ELM), and multilayer perceptron neural network (MLP), are adopted as reference models. In order to maximize the determination coefficient ( R 2 ) value and reduce the root mean square error (RMSE), hyperparameters are set using the grid search method. Using a steel industry dataset, all tests are carried out under identical experimental conditions. In this context, STACK1 (ELM + GBR + XGBR-SVR) and STACK2 (ELM + GBR + XGBR-LASSO) models provided better performance than other models. The highest accuracies of R2 of 0.97 and 0.97 are obtained using STACK1 and STACK2, respectively. Moreover, the rank according to performances is STACK1, STACK2, XGBR, GBR, RFR, MLP, ELM, and SVR. As it improves the performance of models and reduces the risk of decision-making, the ensemble method can be used to forecast the demand in a steel industry one month ahead.