Selected work

Representative Papers

A budget-dependent crossover between coverage- and response-based training-set selection for machine-learned interatomic potentials

Sep 05, 2026

Selecting compact training sets for machine-learned interatomic potentials requires deciding whether to preserve structural diversity or target configurations on which models disagree. The better choice can depend on how much data is retained, making a comparison at one training-set size insufficient. Here we link selection criteria to prediction accuracy through a budget-resolved comparison of retrained MACE models on GAP-20 Carbon and pooled revised MD17. Structural coverage is compared with a response-guided selector that targets disagreement between a coverage-trained model and a full-data reference. This retrospective response witness tests the value of model disagreement for compressing an already labelled pool. At 5\%, coverage gives smaller absolute deviations from the full-data error than random sampling across four force endpoints in both datasets. The witness has larger deviations than coverage at 1\% and 5\%, but the ordering reverses at 20\%. At 20\%, witness-selected models also lower direct held-out force errors by 0.46--5.89\% relative to coverage, with all eight paired training-seed intervals favouring the witness. Six errors fall below the full-data reference. Mean force-error reductions are 0.164--0.167~meV~$\text{\AA}^{-1}$, with larger gains for tail and masked endpoints. Complementary analyses show that learned similarity preserves the coverage ranking, while selecting by frozen-model error gives higher error than embedding coverage. These findings establish retained-data budget as a deciding variable in atomistic training-set selection and provide a direct test of when response-guided compression improves on structural coverage.

0 citationsRead paper

Real-time virtual circuits for plasma shape control via neural network emulators: integration and testing in the MAST-U PCS

Aug 26, 2026

本文通过在MAST-U PCS中集成基于神经网络的虚拟电路，实现等离子体形状的实时控制，并强调了验证流程以确保控制框架的可靠性。

0 citationsRead paper

eIRWR: Enhanced Iterative Random Walk with Restart for Scalable Root Cause Analysis in Microservices

Aug 08, 2026

This work addresses the challenge of root cause localization in microservice architectures, where cascading anomalies are often obscured by downstream noise and weak source signals. To this end, the authors propose eIRWR, a method that leverages the service dependency graph within a personalized PageRank framework. By incorporating a transition matrix enhanced with power-law restart focusing, self-loops, and backward edges, and optimizing belief propagation through outer-loop iterations, eIRWR enables precise anomaly evidence propagation and accurate accumulation of root-source probabilities. Evaluated on large-scale Alibaba datasets (12K–25K nodes), eIRWR achieves an MRR of 0.75 under moderate observability—2.8× higher than the best baseline—and 0.94 under high observability, with per-inference latency below 25 ms, demonstrating strong potential for online deployment.

0 citationsRead paper

Improving the Energy Efficiency of High Throughput Computing: A Measurement-Based Case Study

Aug 06, 2026

This study addresses the challenge of optimizing server energy efficiency in high-throughput computing environments, where performance and energy consumption are often at odds. Leveraging real-world operational data and targeted experiments, the work systematically investigates how server configurations influence power consumption, performance, and carbon emissions, uncovering key barriers to implementing effective energy-saving measures in practice. Through empirical power monitoring, workload modeling, and carbon footprint assessment, the authors identify critical factors governing energy efficiency and propose a practical configuration strategy that simultaneously ensures performance guarantees and advances low-carbon objectives. Evaluated under representative high-throughput workloads, the proposed approach achieves substantial reductions in both energy use and carbon emissions.

0 citationsRead paper

Scalar-pathway fidelity improves physical accuracy in short-range equivariant interatomic potentials

Jun 14, 2026

This work addresses the limited accuracy of short-range equivariant interatomic potential models in representing energy landscapes, which stems from insufficient aggregation capacity and spectral resolution in scalar (ℓ=0) channels. While preserving the equivariant tensor backbone, the authors introduce two lightweight, symmetry-preserving modules—Physics-Aware Neighborhood (PAN) pooling and Physics-Guided Spectral (PGS) mixer—that operate exclusively on scalar channels. These modules incorporate coordinate-sensitive modulation and enhanced radial spectral bases, establishing scalar-path fidelity as a critical design dimension for the first time. Integrated into architectures such as MACE, Allegro, and NequIP, the approach reduces prediction errors for forces, energies, and stresses by 22–27%, 19–22%, and 27–28%, respectively, across Ag, Si, LiF, and MD17/rMD17 datasets, with only ~5% additional inference overhead.

0 citationsRead paper

Recent publications

Latest Papers

A budget-dependent crossover between coverage- and response-based training-set selection for machine-learned interatomic potentials

Sep 05, 2026

Selecting compact training sets for machine-learned interatomic potentials requires deciding whether to preserve structural diversity or target configurations on which models disagree. The better choice can depend on how much data is retained, making a comparison at one training-set size insufficient. Here we link selection criteria to prediction accuracy through a budget-resolved comparison of retrained MACE models on GAP-20 Carbon and pooled revised MD17. Structural coverage is compared with a response-guided selector that targets disagreement between a coverage-trained model and a full-data reference. This retrospective response witness tests the value of model disagreement for compressing an already labelled pool. At 5\%, coverage gives smaller absolute deviations from the full-data error than random sampling across four force endpoints in both datasets. The witness has larger deviations than coverage at 1\% and 5\%, but the ordering reverses at 20\%. At 20\%, witness-selected models also lower direct held-out force errors by 0.46--5.89\% relative to coverage, with all eight paired training-seed intervals favouring the witness. Six errors fall below the full-data reference. Mean force-error reductions are 0.164--0.167~meV~$\text{\AA}^{-1}$, with larger gains for tail and masked endpoints. Complementary analyses show that learned similarity preserves the coverage ranking, while selecting by frozen-model error gives higher error than embedding coverage. These findings establish retained-data budget as a deciding variable in atomistic training-set selection and provide a direct test of when response-guided compression improves on structural coverage.

0 citationsRead paper

Real-time virtual circuits for plasma shape control via neural network emulators: integration and testing in the MAST-U PCS

Aug 26, 2026

本文通过在MAST-U PCS中集成基于神经网络的虚拟电路，实现等离子体形状的实时控制，并强调了验证流程以确保控制框架的可靠性。

0 citationsRead paper

eIRWR: Enhanced Iterative Random Walk with Restart for Scalable Root Cause Analysis in Microservices

Aug 08, 2026

This work addresses the challenge of root cause localization in microservice architectures, where cascading anomalies are often obscured by downstream noise and weak source signals. To this end, the authors propose eIRWR, a method that leverages the service dependency graph within a personalized PageRank framework. By incorporating a transition matrix enhanced with power-law restart focusing, self-loops, and backward edges, and optimizing belief propagation through outer-loop iterations, eIRWR enables precise anomaly evidence propagation and accurate accumulation of root-source probabilities. Evaluated on large-scale Alibaba datasets (12K–25K nodes), eIRWR achieves an MRR of 0.75 under moderate observability—2.8× higher than the best baseline—and 0.94 under high observability, with per-inference latency below 25 ms, demonstrating strong potential for online deployment.

0 citationsRead paper

Improving the Energy Efficiency of High Throughput Computing: A Measurement-Based Case Study

Aug 06, 2026

This study addresses the challenge of optimizing server energy efficiency in high-throughput computing environments, where performance and energy consumption are often at odds. Leveraging real-world operational data and targeted experiments, the work systematically investigates how server configurations influence power consumption, performance, and carbon emissions, uncovering key barriers to implementing effective energy-saving measures in practice. Through empirical power monitoring, workload modeling, and carbon footprint assessment, the authors identify critical factors governing energy efficiency and propose a practical configuration strategy that simultaneously ensures performance guarantees and advances low-carbon objectives. Evaluated under representative high-throughput workloads, the proposed approach achieves substantial reductions in both energy use and carbon emissions.

0 citationsRead paper

Scalar-pathway fidelity improves physical accuracy in short-range equivariant interatomic potentials

Jun 14, 2026

This work addresses the limited accuracy of short-range equivariant interatomic potential models in representing energy landscapes, which stems from insufficient aggregation capacity and spectral resolution in scalar (ℓ=0) channels. While preserving the equivariant tensor backbone, the authors introduce two lightweight, symmetry-preserving modules—Physics-Aware Neighborhood (PAN) pooling and Physics-Guided Spectral (PGS) mixer—that operate exclusively on scalar channels. These modules incorporate coordinate-sensitive modulation and enhanced radial spectral bases, establishing scalar-path fidelity as a critical design dimension for the first time. Integrated into architectures such as MACE, Allegro, and NequIP, the approach reduces prediction errors for forces, energies, and stresses by 22–27%, 19–22%, and 27–28%, respectively, across Ag, Si, LiF, and MD17/rMD17 datasets, with only ~5% additional inference overhead.

0 citationsRead paper