Learning from waste: Machine Learning for health risk prediction and computer vision-based sorting in Ghana
研究通过机器学习预测健康风险，并使用计算机视觉进行废物分类，以解决加纳因不当废物处理导致的公共卫生问题。
0 citationsRead paper
Selected work
Representative Papers
Clinical Assistant for Remote Engagement Link (CARE-link): A Web-Based Electronic Health Records Software for Managing Diabetes
This work proposes an open-source, web-based clinical decision support platform to address fragmented outpatient data, inefficient clinician–patient communication, and high follow-up burdens in gestational diabetes management. The platform introduces a novel dual-endpoint architecture that leverages large language models (LLMs) to intelligently aggregate and summarize patients’ extramural health data, providing clinicians with context-aware decision support. Personalized lifestyle guidance and treatment explanations are delivered directly to patients via WhatsApp. Designed with a modular architecture, the system integrates electronic health records and messaging interfaces to significantly enhance clinical oversight and patient adherence, strengthen continuity of care, and reduce the need for in-person follow-ups. Its adaptable framework also holds promise for extension to other chronic disease management contexts.
0 citationsRead paper
Entropy Gate: Entropy Quenching for Near-Lossless Token Compression in LLM Pipelines
This work addresses the inefficiency in large language model (LLM) inference caused by excessive computation and communication overhead from low-information tokens. The authors propose Entropy Gate, a novel framework that introduces thermodynamic entropy quenching into LLM token compression. By integrating statistical, structural, and positional features into a multi-factor information energy metric, the method employs adaptive temperature scheduling and Boltzmann-based survival probabilities to dynamically prune low-energy tokens, augmented with semantic fidelity gating and context deduplication. Theoretically, selecting tokens in descending order of information energy maximizes semantic retention and approaches the information-theoretic compression limit. Experiments demonstrate 40–60% compression rates across five prompt types while maintaining semantic similarity (SE > 0.80); with energy-squared amplification and external memory, agent tasks achieve total compression of 88–96%, supporting stateless, model-agnostic deployment.
0 citationsRead paper
Optimal Quantum Differential Privacy via Fisher Information Spectral Analysis
This work addresses the challenge of balancing privacy preservation and data utility in quantum computing by proposing a geometry-aware differential privacy framework grounded in the spectral structure of quantum Fisher information (QFI). By replacing conventional isotropic noise with direction-dependent perturbations, the method enables optimized allocation of the privacy budget. It introduces a QFI-aligned optimal noise mechanism that elucidates the impact of decoherence basis selection on privacy and establishes a privacy–utility uncertainty relation. Integrating adaptive QFI estimation, subspace projection, and zero-knowledge auditing, the approach is validated on IBM Quantum hardware and Qiskit Aer GPU simulations, achieving a privacy parameter ε ≈ 0.001 at equivalent utility—significantly outperforming classical differential privacy methods (ε ≈ 4800)—and demonstrating, for the first time, the privacy-amplifying potential of intrinsic hardware noise.
0 citationsRead paper
The Streaming Reservoir Convergence Theorem: A Prospect-Theoretic Framework for Multi-Provider Adaptive Streaming
This work addresses the high latency and playback interruptions in existing adaptive bitrate streaming systems caused by linear probing, passive switching, and cold standby mechanisms. The authors propose a reservoir model based on concurrent multi-source probing and prefetching of backup streams, formulated as a concurrent reservoir-filling problem. By simultaneously probing all providers and maintaining k pre-validated backup streams, the system achieves sub-second, seamless failover. Theoretical contributions include a harmonic lower bound for reservoir safety, an acceleration ratio for concurrent probing, monotonic quality convergence under lazy loading, and a jitter-resistant switching rule grounded in prospect theory (α=β=0.88, λ=2.25). Experiments across 12 HLS providers demonstrate that with k=3, the mean time to exhaustion improves by 9.15×, and concurrent probing yields a 4.27× speedup under a 40% failure rate, comprehensively validating the theoretical findings.
0 citationsRead paper
Recent publications
Latest Papers
Learning from waste: Machine Learning for health risk prediction and computer vision-based sorting in Ghana
研究通过机器学习预测健康风险，并使用计算机视觉进行废物分类，以解决加纳因不当废物处理导致的公共卫生问题。
0 citationsRead paper
Clinical Assistant for Remote Engagement Link (CARE-link): A Web-Based Electronic Health Records Software for Managing Diabetes
This work proposes an open-source, web-based clinical decision support platform to address fragmented outpatient data, inefficient clinician–patient communication, and high follow-up burdens in gestational diabetes management. The platform introduces a novel dual-endpoint architecture that leverages large language models (LLMs) to intelligently aggregate and summarize patients’ extramural health data, providing clinicians with context-aware decision support. Personalized lifestyle guidance and treatment explanations are delivered directly to patients via WhatsApp. Designed with a modular architecture, the system integrates electronic health records and messaging interfaces to significantly enhance clinical oversight and patient adherence, strengthen continuity of care, and reduce the need for in-person follow-ups. Its adaptable framework also holds promise for extension to other chronic disease management contexts.
0 citationsRead paper
Entropy Gate: Entropy Quenching for Near-Lossless Token Compression in LLM Pipelines
This work addresses the inefficiency in large language model (LLM) inference caused by excessive computation and communication overhead from low-information tokens. The authors propose Entropy Gate, a novel framework that introduces thermodynamic entropy quenching into LLM token compression. By integrating statistical, structural, and positional features into a multi-factor information energy metric, the method employs adaptive temperature scheduling and Boltzmann-based survival probabilities to dynamically prune low-energy tokens, augmented with semantic fidelity gating and context deduplication. Theoretically, selecting tokens in descending order of information energy maximizes semantic retention and approaches the information-theoretic compression limit. Experiments demonstrate 40–60% compression rates across five prompt types while maintaining semantic similarity (SE > 0.80); with energy-squared amplification and external memory, agent tasks achieve total compression of 88–96%, supporting stateless, model-agnostic deployment.
0 citationsRead paper
Optimal Quantum Differential Privacy via Fisher Information Spectral Analysis
This work addresses the challenge of balancing privacy preservation and data utility in quantum computing by proposing a geometry-aware differential privacy framework grounded in the spectral structure of quantum Fisher information (QFI). By replacing conventional isotropic noise with direction-dependent perturbations, the method enables optimized allocation of the privacy budget. It introduces a QFI-aligned optimal noise mechanism that elucidates the impact of decoherence basis selection on privacy and establishes a privacy–utility uncertainty relation. Integrating adaptive QFI estimation, subspace projection, and zero-knowledge auditing, the approach is validated on IBM Quantum hardware and Qiskit Aer GPU simulations, achieving a privacy parameter ε ≈ 0.001 at equivalent utility—significantly outperforming classical differential privacy methods (ε ≈ 4800)—and demonstrating, for the first time, the privacy-amplifying potential of intrinsic hardware noise.
0 citationsRead paper
The Streaming Reservoir Convergence Theorem: A Prospect-Theoretic Framework for Multi-Provider Adaptive Streaming
This work addresses the high latency and playback interruptions in existing adaptive bitrate streaming systems caused by linear probing, passive switching, and cold standby mechanisms. The authors propose a reservoir model based on concurrent multi-source probing and prefetching of backup streams, formulated as a concurrent reservoir-filling problem. By simultaneously probing all providers and maintaining k pre-validated backup streams, the system achieves sub-second, seamless failover. Theoretical contributions include a harmonic lower bound for reservoir safety, an acceleration ratio for concurrent probing, monotonic quality convergence under lazy loading, and a jitter-resistant switching rule grounded in prospect theory (α=β=0.88, λ=2.25). Experiments across 12 HLS providers demonstrate that with k=3, the mean time to exhaustion improves by 9.15×, and concurrent probing yields a 4.27× speedup under a 40% failure rate, comprehensively validating the theoretical findings.
0 citationsRead paper