Publications & Selected Projects

A complete record of my peer-reviewed conference and workshop publications, along with selected research engineering projects. All publications are authored under the name Ahmed Hassan/Ahmed Hasan.

📚 Publications

Neural Residual Maps: Proposal-Level False Positive Suppression for Fixed-Camera Object Detection

• First Author
• Accepted at the British Machine Vision Conference (BMVC 2026)

Engineered Neural Residual Maps (NRM), a proposal-level background-rescoring formulation mitigating persistent static distractors without detector fine-tuning, improving P-AUC by up to +9.50 across YOLO11m and RT-DETR-L backbones.

TEMPO at SignEval 2026: Signer-Independent Temporal Modeling and Vocabulary-Constrained Post-Processing for Arabic CSLR

• First Author
• Published at IEEE/CVF Computer Vision and Pattern Recognition (CVPR) Workshops 2026
• Paper Link

Secured 1st place globally in the SignEval 2026 Continuous Sign Language Recognition challenge, establishing a 5.68% Word Error Rate (WER) via signer-invariant temporal representations and vocabulary-constrained beam search.

DAgent: A Multi-Agent System for Device-Aware Assistance

• First Author
• Published in the 35th International Conference on Computer Theory and Applications (ICCTA 2025)

• IEEE Xplore

  GitHub

Architected a modular multi-agent orchestration framework (Tracer, RAG, Code Synthesizer, OS Controller) optimizing localized task-execution fidelity and achieving an operational correctness score of 4.87/5.0.

Media Bias in The Guardian: Sentiment and Semantics in Coverage of the Israeli-Palestinian Conflict

• Co-First Author
• Published in the 9th International Conference on Computer-Human Interaction Research and Applications (CHIRA 2025)
• DOI Link

Engineered a computational NLP framework utilizing contextual representations to quantify longitudinal framing biases, demonstrating statistically significant divergence including a 13.5% elevation in threat-oriented lexical polarity.

ALEX-GYM-1: A Hybrid 3D Pose & Vision Model for Automated Exercise Evaluation

• First Author
• Published in the 22nd International Conference on Informatics in Control, Automation and Robotics (ICINCO 2025)

• DOI Link

  GitHub

Formulated a dual-stream multimodal architecture fusing 3D CNN spatiotemporal visual representations with coordinate pose kinematics, achieving a 30.0% reduction in Hamming Loss and a 79.5% accuracy gain over unimodal pose baselines.

TypePlus: A Deep Learning Architecture for Keystroke Authentication

• Co-First Author
• Published in the 2nd International Conference on Intelligent Systems, Blockchain, and Communication Technologies (ISBCom 2025)

• Springer Link

  GitHub

Designed a lightweight, non-transformer continuous keystroke biometrics architecture with temporal weighted-attention pooling, establishing a 2.86% Equal Error Rate (EER) on the Aalto dataset with sub-millisecond inference latency.

PPE-Det: Lightweight Object Detection Models for Edge-Based Safety Monitoring

• Co-First Author
• Published in the 2nd International Conference on Intelligent Systems, Blockchain, and Communication Technologies (ISBCom 2025)

• Springer Link

  GitHub

Curated a domain-specific 5,000-sample industrial dataset and benchmarked sub-3M parameter architectures (YOLOv9t, YOLOv11n) on Raspberry Pi 5, identifying optimal Pareto bounds between mAP and edge memory bandwidth.

DeepCat: Deep Learning Approach to Understand Cat Body Language

• Co-First Author
• Published in the 11th International Japan-Africa Conference on Electronics, Communications, and Computations (JAC-ECC 2023)

• IEEE Xplore

  GitHub

Built an on-device hierarchical affective state classification pipeline over 10,000+ visual samples, mapping structural keypoints to affective taxonomies with 97.0% ocular localization precision and 85.0% caudal posture accuracy.