Selected work

Representative Papers

Exact Distinguishability in Non-Markovian Decision Processes

Oct 01, 2026

This study addresses the challenge that the distinguishability assumptions underlying offline guarantees in non-Markovian decision processes are difficult to verify and prone to failure. Drawing upon finite automata theory and Bayesian inference, we prove the posterior odds invariance of observationally equivalent models, thereby providing the first precise characterization of indistinguishability. Based on this theoretical foundation, we propose PEC, a linear-time decision algorithm, and formally verify its core theorems using Lean 4. Experimental results demonstrate that conventional assumptions frequently fail in practice, whereas the PEC algorithm effectively restores distinguishability, ensuring that the theoretical guarantees for offline policy evaluation remain valid.

0 citationsRead paper

Drive, Pack, Fly: The Travelling Thief Problem with Drone

Aug 17, 2026

This study addresses the vehicle-drone collaborative collection and profit maximization problem under load-dependent travel times by proposing a Traveling Thief Problem with Drones (TTP-D) model that jointly optimizes item selection, route planning, and flight synchronization. A hybrid solution framework integrating mixed-integer programming, metaheuristics, and attention-based deep reinforcement learning is developed, featuring a learner-initialized hybrid solver to balance solution quality with computational efficiency. Experimental results demonstrate that this solver recovers baseline performance under low computational budgets and identifies the rental-to-profit ratio as a critical determinant of system profitability. These findings provide efficient decision support for complex collaborative scheduling in logistics applications where operational costs and payload dynamics significantly impact overall economic performance.

0 citationsRead paper

Beyond Independent Optimization: Compression, MoE Routing, and Quantization Interactions in Multimodal Edge Intelligence

Jul 23, 2026

This work addresses the limitations of existing edge-based multimodal intelligent systems, which typically optimize components such as compression, routing, and quantization in isolation, thereby overlooking their intricate couplings and failing to balance latency, energy consumption, and accuracy. The study systematically investigates the interaction mechanisms among visual/video token compression, KV caching, Mixture-of-Experts (MoE) routing, and low-bit quantization in large vision-language models, proposing a unified co-design framework. It introduces “temporal routing consistency” as a novel diagnostic metric for video MoE and advocates for routing-aware compression, cross-modal cache management, and hardware-aware design. The analysis reveals critical trade-offs—such as how compression perturbs feature distributions affecting routing decisions, how quantization interferes with expert assignment, and how caching strategies govern the retention of multimodal evidence—thereby establishing foundational design principles and open pathways for efficient edge multimodal systems.

0 citationsRead paper

Zero-Copy Semantic Contagion: An In-Memory Streaming Architecture for Evolving Attention Graphs

Jun 04, 2026

This study addresses the inability of traditional single-asset forecasting models to capture cross-firm event propagation, which results in delayed market responses. To overcome this limitation, the authors propose a heterogeneous Rust-Python streaming architecture that maps news in real time onto continuous-time heterogeneous graphs and models inter-firm influence through a dynamic attention mechanism. Key innovations include zero-copy streaming parsing, node-level continuous-time LSTMs, a multivariate Neural Hawkes process, bilinear latent projection, and adaptive neighborhood pruning. The system achieves an end-to-end latency of only 13 ms. Evaluated on the FNSPID corpus, it improves next-day return prediction accuracy at the 90th percentile by 1.70× over random guessing and by 3.36× over industry baselines. Ablation studies confirm that graph topology is the sole source of cross-firm predictive signals.

0 citationsRead paper

Attention-gated U-Net model for semantic segmentation of brain tumors and feature extraction for survival prognosis

Feb 14, 2026

Gliomas exhibit high heterogeneity in invasiveness, prognosis, and histology, necessitating precise segmentation and survival prediction methods. This work proposes an attention-gated recurrent residual U-Net with triplanar (2.5D) inputs—termed Attention-Gated R2U-Net—that uniquely integrates attention gates, residual connections, recurrent structures, and multiplanar input to enhance both segmentation accuracy and survival prediction while maintaining computational efficiency. Evaluated on the BraTS2021 validation set, the model achieves a Dice score of 0.900 for whole-tumor segmentation. Leveraging the extracted multiplanar features, an artificial neural network (ANN) is employed for survival regression, yielding a prediction accuracy of 45.71%, a mean squared error (MSE) of 108,318.128, and a Spearman correlation coefficient of 0.338.

0 citationsRead paper

Recent publications

Latest Papers

Exact Distinguishability in Non-Markovian Decision Processes

Oct 01, 2026

This study addresses the challenge that the distinguishability assumptions underlying offline guarantees in non-Markovian decision processes are difficult to verify and prone to failure. Drawing upon finite automata theory and Bayesian inference, we prove the posterior odds invariance of observationally equivalent models, thereby providing the first precise characterization of indistinguishability. Based on this theoretical foundation, we propose PEC, a linear-time decision algorithm, and formally verify its core theorems using Lean 4. Experimental results demonstrate that conventional assumptions frequently fail in practice, whereas the PEC algorithm effectively restores distinguishability, ensuring that the theoretical guarantees for offline policy evaluation remain valid.

0 citationsRead paper

Drive, Pack, Fly: The Travelling Thief Problem with Drone

Aug 17, 2026

This study addresses the vehicle-drone collaborative collection and profit maximization problem under load-dependent travel times by proposing a Traveling Thief Problem with Drones (TTP-D) model that jointly optimizes item selection, route planning, and flight synchronization. A hybrid solution framework integrating mixed-integer programming, metaheuristics, and attention-based deep reinforcement learning is developed, featuring a learner-initialized hybrid solver to balance solution quality with computational efficiency. Experimental results demonstrate that this solver recovers baseline performance under low computational budgets and identifies the rental-to-profit ratio as a critical determinant of system profitability. These findings provide efficient decision support for complex collaborative scheduling in logistics applications where operational costs and payload dynamics significantly impact overall economic performance.

0 citationsRead paper

Beyond Independent Optimization: Compression, MoE Routing, and Quantization Interactions in Multimodal Edge Intelligence

Jul 23, 2026

This work addresses the limitations of existing edge-based multimodal intelligent systems, which typically optimize components such as compression, routing, and quantization in isolation, thereby overlooking their intricate couplings and failing to balance latency, energy consumption, and accuracy. The study systematically investigates the interaction mechanisms among visual/video token compression, KV caching, Mixture-of-Experts (MoE) routing, and low-bit quantization in large vision-language models, proposing a unified co-design framework. It introduces “temporal routing consistency” as a novel diagnostic metric for video MoE and advocates for routing-aware compression, cross-modal cache management, and hardware-aware design. The analysis reveals critical trade-offs—such as how compression perturbs feature distributions affecting routing decisions, how quantization interferes with expert assignment, and how caching strategies govern the retention of multimodal evidence—thereby establishing foundational design principles and open pathways for efficient edge multimodal systems.

0 citationsRead paper

Zero-Copy Semantic Contagion: An In-Memory Streaming Architecture for Evolving Attention Graphs

Jun 04, 2026

This study addresses the inability of traditional single-asset forecasting models to capture cross-firm event propagation, which results in delayed market responses. To overcome this limitation, the authors propose a heterogeneous Rust-Python streaming architecture that maps news in real time onto continuous-time heterogeneous graphs and models inter-firm influence through a dynamic attention mechanism. Key innovations include zero-copy streaming parsing, node-level continuous-time LSTMs, a multivariate Neural Hawkes process, bilinear latent projection, and adaptive neighborhood pruning. The system achieves an end-to-end latency of only 13 ms. Evaluated on the FNSPID corpus, it improves next-day return prediction accuracy at the 90th percentile by 1.70× over random guessing and by 3.36× over industry baselines. Ablation studies confirm that graph topology is the sole source of cross-firm predictive signals.

0 citationsRead paper

Attention-gated U-Net model for semantic segmentation of brain tumors and feature extraction for survival prognosis

Feb 14, 2026

Gliomas exhibit high heterogeneity in invasiveness, prognosis, and histology, necessitating precise segmentation and survival prediction methods. This work proposes an attention-gated recurrent residual U-Net with triplanar (2.5D) inputs—termed Attention-Gated R2U-Net—that uniquely integrates attention gates, residual connections, recurrent structures, and multiplanar input to enhance both segmentation accuracy and survival prediction while maintaining computational efficiency. Evaluated on the BraTS2021 validation set, the model achieves a Dice score of 0.900 for whole-tumor segmentation. Leveraging the extracted multiplanar features, an artificial neural network (ANN) is employed for survival regression, yielding a prediction accuracy of 45.71%, a mean squared error (MSE) of 108,318.128, and a Spearman correlation coefficient of 0.338.

0 citationsRead paper