arXiv subjects

Sitong Zhang

Publications and source records attributed to Sitong Zhang.

3 recordsLinked to original sources

arXiv2026

Conduit: An Experience Data Plane for Distributed Reinforcement Learning

Distributed reinforcement learning (RL) scales training by parallelizing actors and learners around an Experience Buffer. As RL workloads grow, however, the buffer becomes more than a replay queue: it is the storage substrate of a large-capacity, latency-critical experience path that every iteration traverses to move, transform, sample, and batch experiences before learner updates can begin. Existing RL systems embed this path inside framework control flow or expose it as a request-driven buffer service, leaving experience placement fixed and experience-path work difficult to schedule independently as a runtime-level optimization target. We present Conduit, a framework-agnostic runtime that exposes RL experience management as an explicit systems optimization problem. At its core is the Experience Data Plane (EDP), a runtime abstraction that separates RL experience-handling semantics from framework-specific execution logic by exposing experience ingestion, experience placement, and experience delivery as explicit control points. Built on EDP, Conduit introduces capacity-constrained, bandwidth-aware placement, which distributes experience state across CPU/GPU memory tiers and nodes under heterogeneous interconnect and device-memory constraints, and latency-aware scheduling, which controls when experience-path handling runs to reduce exposed experience-path latency while preserving RL semantics. Integrated with RLlib without changing its framework execution logic, Conduit reduces exposed experience-path latency by up to 97% and end-to-end iteration latency by up to 38%, scales to 1,024 GPUs, and preserves convergence.

cs.DC↗

arXiv2025

An Interpretable Two-Stage Feature Decomposition Method for Deep Learning-based SAR ATR

Synthetic aperture radar automatic target recognition (SAR ATR) has seen significant performance improvements with deep learning. However, the black-box nature of deep SAR ATR introduces low confidence and high risks in decision-critical SAR applications, hindering practical deployment. To address this issue, deep SAR ATR should provide an interpretable reasoning basis $r_b$ and logic $λ_w$, forming the reasoning logic $\sum_{i} {{r_b^i} \times {λ_w^i}} =pred$ behind the decisions. Therefore, this paper proposes a physics-based two-stage feature decomposition method for interpretable deep SAR ATR, which transforms uninterpretable deep features into attribute scattering center components (ASCC) with clear physical meanings. First, ASCCs are obtained through a clustering algorithm. To extract independent physical components from deep features, we propose a two-stage decomposition method. In the first stage, a feature decoupling and discrimination module separates deep features into approximate ASCCs with global discriminability. In the second stage, a multilayer orthogonal non-negative matrix tri-factorization (MLO-NMTF) further decomposes the ASCCs into independent components with distinct physical meanings. The MLO-NMTF elegantly aligns with the clustering algorithms to obtain ASCCs. Finally, this method ensures both an interpretable reasoning process and accurate recognition results. Extensive experiments on four benchmark datasets confirm its effectiveness, showcasing the method's interpretability, robust recognition performance, and strong generalization capability.

eess.IV↗

arXiv2023

An adaptive preconditioning scheme for the self-consistent field iteration and generalized stacking-fault energy calculations

The generalized stacking-fault energy (GSFE) is the fundamental but key parameter for the plastic deformation of materials. We perform first-principles calculations by full-potential linearized augmented planewave (FLAPW) method to evaluate the GSFE based on the single-shift and triple-shift supercell models. Different degrees of defects are introduced in the two models, thereby affecting the convergence of the self-consistent field (SCF) iterations. We present an adaptive preconditioning scheme which can identify the long-wavelength divergence behavior of the Jacobian during the SCF iteration and automatically switch on the Kerker preconditioning to accelerate the convergence. We implement this algorithm in Elk-7.2.42 package and calculate the GSFE curves for Al, Cu, and Si (111) plane <-1-12> direction. We found that the single-shift and triple-shift supercell models have equivalent calculation accuracy and are within the experimental data uncertainty. For computational efficiency, the triple-shift supercell model is preferable due to its better convergence, exhibiting lower degree of defect compared to the single-shift supercell model.

cond-mat.mtrl-sci↗