Selected work

Representative Papers

Managing Iterative Hybrid Quantum-Classical Optimization as a First-Class Scientific Workflow

Sep 24, 2026

This study addresses the lack of orchestration, recovery, and portability in driver scripts for hybrid quantum-classical optimization by modeling iterative decomposition-solving-aggregation loops as scientific workflows. A dedicated orchestration layer is constructed to uniformly manage task generation, data provenance, and fault recovery. The proposed workflow model incorporates termination predicates, subproblem-level recovery, and QPU-to-classical-backend failover, while integrating ADMM, hierarchical partitioning, speculative re-execution, and quantum-HPC middleware. Experiments quantify the orchestration overhead across different decomposition patterns, validate system robustness under fault injection, and perform cross-device latency analysis.

0 citationsRead paper

A Block Decomposed QUBO Workflow for Chromosome-Y Phylogeny Reconstruction

Sep 24, 2026

This study addresses the computational challenges of topology selection and rooting in reconstructing the human Y-chromosome phylogeny by proposing a novel framework that integrates quantum-inspired optimization. Methodologically, phylogenetic decision-making is formulated as a Quadratic Unconstrained Binary Optimization (QUBO) problem, where the Alternating Direction Method of Multipliers (ADMM) enables modular decomposition to overcome qubit budget constraints, and a Divide-and-Conquer QUBO Optimization (DCQO) solver facilitates efficient optimization. The results demonstrate that, when applied to VCF data parsing, this approach generates annotated rooted trees alongside diagnostic visualizations. It significantly outperforms conventional greedy heuristic algorithms, thereby validating its scalability and application potential for large-scale phylogenetic analyses in population genomics.

0 citationsRead paper

Hiding Latencies in Network-Based Image Loading for Deep Learning

Mar 28, 2025

To address data loading bottlenecks and local storage overload caused by high network/storage latency in deep learning image training, this paper proposes a unified data management framework built upon scalable NoSQL databases (e.g., Cassandra). The method introduces three key innovations: (1) an adaptive out-of-order incremental prefetching mechanism that effectively masks I/O latency across intercontinental high-latency networks; (2) the first low-overhead integration of scalable NoSQL backends with native PyTorch/TensorFlow data loaders; and (3) RDMA- and HTTP/3–enabled transport optimizations. Experimental evaluation demonstrates that the framework achieves 92% of local SSD throughput, reduces metadata query latency by three orders of magnitude, and decreases per-node storage pressure by 76%. These results validate its effectiveness in decoupling compute scalability from local storage constraints while maintaining high training efficiency.

0 citationsRead paper

Recent publications

Latest Papers

Managing Iterative Hybrid Quantum-Classical Optimization as a First-Class Scientific Workflow

Sep 24, 2026

This study addresses the lack of orchestration, recovery, and portability in driver scripts for hybrid quantum-classical optimization by modeling iterative decomposition-solving-aggregation loops as scientific workflows. A dedicated orchestration layer is constructed to uniformly manage task generation, data provenance, and fault recovery. The proposed workflow model incorporates termination predicates, subproblem-level recovery, and QPU-to-classical-backend failover, while integrating ADMM, hierarchical partitioning, speculative re-execution, and quantum-HPC middleware. Experiments quantify the orchestration overhead across different decomposition patterns, validate system robustness under fault injection, and perform cross-device latency analysis.

0 citationsRead paper

A Block Decomposed QUBO Workflow for Chromosome-Y Phylogeny Reconstruction

Sep 24, 2026

This study addresses the computational challenges of topology selection and rooting in reconstructing the human Y-chromosome phylogeny by proposing a novel framework that integrates quantum-inspired optimization. Methodologically, phylogenetic decision-making is formulated as a Quadratic Unconstrained Binary Optimization (QUBO) problem, where the Alternating Direction Method of Multipliers (ADMM) enables modular decomposition to overcome qubit budget constraints, and a Divide-and-Conquer QUBO Optimization (DCQO) solver facilitates efficient optimization. The results demonstrate that, when applied to VCF data parsing, this approach generates annotated rooted trees alongside diagnostic visualizations. It significantly outperforms conventional greedy heuristic algorithms, thereby validating its scalability and application potential for large-scale phylogenetic analyses in population genomics.

0 citationsRead paper

Hiding Latencies in Network-Based Image Loading for Deep Learning

Mar 28, 2025

To address data loading bottlenecks and local storage overload caused by high network/storage latency in deep learning image training, this paper proposes a unified data management framework built upon scalable NoSQL databases (e.g., Cassandra). The method introduces three key innovations: (1) an adaptive out-of-order incremental prefetching mechanism that effectively masks I/O latency across intercontinental high-latency networks; (2) the first low-overhead integration of scalable NoSQL backends with native PyTorch/TensorFlow data loaders; and (3) RDMA- and HTTP/3–enabled transport optimizations. Experimental evaluation demonstrates that the framework achieves 92% of local SSD throughput, reduces metadata query latency by three orders of magnitude, and decreases per-node storage pressure by 76%. These results validate its effectiveness in decoupling compute scalability from local storage constraints while maintaining high training efficiency.

0 citationsRead paper