Ahmed Fares

In this paper, a novel deep reinforcement learning approach is presented for adaptive traffic signal control, using a three-dimensional convolutional neural network (Conv3D) to process spatiotemporal traffic-state representations that leverage a compact lane-slot occupancy matrix. The proposed state encodes only binary vehicle positions across spatial slots and temporal frames. Simultaneously, aggregated features such as waiting time or speed are completely excluded. This representation forms a three-dimensional tensor where time is modeled as a distinct temporal dimension rather than channels, enabling the Conv3D encoder to jointly extract spatio-temporal features across spatial and temporal dimensions. The state representation is more memory-efficient than image-based discrete traffic state encoding methods that use square image matrices, while eliminating the need for handcrafted feature engineering. The proposed encoder learns patterns including vehicle flow, queue propagation, and traffic wave dynamics directly from raw position data. Integrated into a deep Q-network (DQN) architecture, the system is evaluated on a real-world traffic scenario from the RESCO benchmark. The proposed Conv3D-based DQN agent outperforms fixed-time control and both features-based baselines (DQN and PPO) across multiple performance metrics, achieving better results in time loss, duration, waiting time, and queue metrics.

Robotic-assisted rehabilitation is a promising method for improving motor function in individuals with upper limb impairments. However, generation of personalized and adaptive assistance patterns remains a challenge. Our study introduces a Learn-from-Therapist Demonstration (LfTD) framework, which employs Dynamic Movement Primitives (DMP) to train a robot arm to learn from therapist skills. Initially, therapist movements were captured via visual tracking, and the DMP accurately learned and replicated these motions via a robotic arm to assist patients. These movements were then effectively generalized to new goals while maintaining the original motion patterns. Meanwhile, a Model Reference Adaptive Controller (MRAC) has been utilized to refine the robot's adaptive performance while assuring demonstration tracking. We assessed the LfTD's efficacy with a simulated two-link robot, which showed excellent learning, adaptation, and ability to perform complex rehabilitation tasks with precise trajectory tracking. Further tests evaluated the MRAC's robustness against introduced human deviations, demonstrating its resilience and adaptability. These findings highlight LfTD’s potential to improve upper limb robotic-assisted rehabilitation through precise, adaptable motion replication, setting the stage for clinical trials with actual robots.

Compared to conventional image content analysis tasks, visual emotion analysis is perceived as a complex, abstract, and potentially culturally dependent endeavor. The accuracy of automatic image-based emotion recognition remains a challenge, and the most significant obstacles are the affective gap and scarcity of data, particularly labeled data. To address these challenge, this paper proposes a saliency-guided masked image modeling approach. Specifically, the proposed framework employs multi-modal large model to generate more emotional images, thereby reducing the impact of data scarcity on model performance. Subsequently, neuroimaging and behavioral studies have demonstrated that human visual attention is attracted by the emotional relevance of a stimulus. In light of this, our model employs a saliency-guided masking strategy to identify emotion-related regions for masking sampling to fit the affective gap. In contrast to the conventional approach of using the original pixel values for the reconstruction target, our model eliminates high-frequency components from the pixels, thus enhancing the generalizability of the model. The use of this unsupervised representation learning approach enables the model to exhibit outstanding recognition performance in downstream emotion recognition tasks on three standard emotion datasets. Furthermore, ablation experiments, robustness test, and visualization experiments corroborate the effectiveness of the proposed method.

Understanding how the human brain processes and interprets multimedia content represents a frontier challenge in neuroscience and artificial intelligence. This study introduces a novel approach to decode semantic information from electroencephalogram (EEG) signals recorded during visual stimulus perception. We present DCT-ViT, a spatial–temporal transformer architecture that pioneers automated semantic recognition from brain activity patterns, advancing beyond conventional brain state classification to interpret higher level cognitive understanding. Our methodology addresses three fundamental innovations: First, we develop a topology-preserving 2D electrode mapping that, combined with temporal indexing, generates 3D spatial–temporal representations capturing both anatomical relationships and dynamic neural correlations. Second, we integrate discrete cosine transform (DCT) embeddings with standard patch and positional embeddings in the transformer architecture, enabling frequency-domain analysis that quantifies activation variability across spectral bands and enhances attention mechanisms. Third, we introduce the Semantics-EEG dataset comprising ten semantic categories extracted from visual stimuli, providing a benchmark for brain-perceived semantic recognition research. The proposed DCT-ViT model achieves 72.28% recognition accuracy on Semantics-EEG, substantially outperforming LSTM-based and attention-augmented recurrent baselines. Ablation studies demonstrate that DCT embeddings contribute meaningfully to model performance, validating their effectiveness in capturing frequency-specific neural signatures. Interpretability analyses reveal neurobiologically plausible attention patterns, with visual semantics activating occipital–parietal regions and abstract concepts engaging frontal–temporal networks, consistent with established cognitive neuroscience models. To address systematic misclassification between perceptually similar categories, we develop a hierarchical classification framework with boundary refinement mechanisms. This approach substantially reduces confusion between overlapping semantic categories, elevating overall accuracy to 76.15%. Robustness evaluations demonstrate superior noise resilience, effective cross-subject generalization, and few-shot transfer capabilities to novel categories. This work establishes the technical foundation for brain–computer interfaces capable of decoding semantic understanding, with implications for assistive technologies, cognitive assessment, and human–AI interaction. Both the Semantics-EEG dataset and DCT-ViT implementation are publicly released to facilitate reproducibility and advance research in neural semantic decoding.

Despite recent advances in AI, multimodal culturally diverse datasets are still limited, particularly for regions in the Middle East and Africa. In this paper, we introduce EgMM-Corpus, a multimodal dataset dedicated to Egyptian culture. By designing and running a new data collection pipeline, we collected over 3,000 images, covering 313 concepts across landmarks, food, and folklore. Each entry in the dataset is manually validated for cultural authenticity and multimodal coherence. EgMM-Corpus aims to provide a reliable resource for evaluating and training vision-language models in an Egyptian cultural context. We further evaluate the zero-shot performance of Contrastive Language-Image Pre-training CLIP on EgMM-Corpus, on which it achieves 21.2% Top-1 accuracy and 36.4% Top-5 accuracy in classification. These results underscore the existing cultural bias in large-scale vision-language models and demonstrate the importance of EgMM-Corpus as a benchmark for developing culturally aware models.

This study presents a novel deep learning framework for classifying visual images based on brain responses recorded through electroencephalogram (EEG) signals. The primary challenge in EEG-based visual pattern recognition lies in the inherent spatiotemporal variability of neural signals across different individuals and recording sessions, which severely limits the generalization capabilities of classification models. Our work specifically addresses the task of identifying which image category a person is viewing based solely on their recorded brain activity. The proposed methodology incorporates three primary components: first, a brain hemisphere asymmetry-based dimensional reduction approach to extract discriminative lateralization features while addressing high-dimensional data constraints; second, an advanced channel selection algorithm utilizing Fisher score methodology to identify electrodes with optimal spatial representativeness across participants; and third, a Dynamic Temporal Warping (DTW) alignment technique to synchronize temporal signal variations with respect to selected reference channels. Comprehensive experimental validation on a visual image classification task using a publicly available EEG-based visual classification dataset, ImageNet-EEG, demonstrates that the proposed disambiguation framework substantially improves classification accuracy while simultaneously enhancing model convergence characteristics. The integrated approach not only outperforms individual component implementations but also accelerates the learning process, thereby reducing training data requirements for EEG-based applications. These findings suggest that systematic spatiotemporal disambiguation represents a promising direction for developing robust and generalizable EEG classification systems across diverse neurological and brain–computer interface applications.

Digital healthcare infrastructure is crucial for global medical service delivery. Egypt faces EHR adoption barriers: only 314 hospitals had such systems as of Oct 2024. This limits data management and decision-making. This project introduces an EHR system for Egypt's Universal Health Insurance and healthcare ecosystem. It simplifies data management by centralizing medical histories with a scalable micro-services architecture and polyglot persistence for real-time access and provider communication. Clinical workflows are enhanced via patient examination and history tracking. The system uses the Llama3-OpenBioLLM-70B model to generate summaries of medical histories, provide chatbot features, and generate AI-based medical reports, enabling efficient searches during consultations. A Vision Transformer (ViT) aids in pneumonia classification. Evaluations show the AI excels in capturing details (high recall) but needs improvement in concise narratives. With optimization (retrieval-augmented generation, local data fine-tuning, interoperability protocols), this AI-driven EHR could enhance diagnostic support, decision-making, and healthcare delivery in Egypt.