Beyond Repository Boundaries: Cross-Repository Graph Retrieval for Code Generation
为解决代码生成中依赖环境兼容性问题，提出CrossCoder框架，通过跨仓库知识图谱整合外部库，并采用多跳检索策略增强上下文理解。
0 citationsRead paper
Selected work
Representative Papers
Variance Driven Exploration: A Provable and Efficient Methodology for Pure Exploration in Highly Stochastic Environments
本文提出了一种名为VarDE的方法，通过最小化最终决策的不确定性来解决高度随机环境中的纯探索问题，并在多个核心问题上展示了其优越性。
0 citationsRead paper
SCENARIODIFF: A Scenario-level Guidance Framework for Multimodal Time Series Forecasting--Extended Version
为了解决多模态时间序列预测中的外部事件影响问题，提出SCENARIODIFF框架，通过分层上下文推理和锚点指导生成更准确的预测。
0 citationsRead paper
When Denoising Hurts: Rethinking the Terminal Step of Diffusion Time Series Forecasters -- Extended Version
This study addresses the statistical drift and performance degradation caused by persistent low-noise denoising in diffusion-based time series forecasting by elucidating the detrimental mechanisms of over-denoising. We propose a label-free global stopping criterion and a Bernoulli time-step sampler focusing on high-noise regions to jointly optimize training sampling distributions and inference termination points. Experiments across eight real-world datasets demonstrate that this approach effectively circumvents the over-denoising trap, significantly improving prediction accuracy while accelerating inference. The proposed method achieves superior overall performance compared to existing mainstream techniques, establishing a new paradigm for the efficient application of diffusion models in time series forecasting.
0 citationsRead paper
EnclaveScale: Hardware-Assisted Edge-DP for Secure Data Centre Power Telemetry
This work addresses the limitations of existing encrypted telemetry schemes, which struggle to support high-frequency (10 Hz) power data streams and lack robust source authentication, rendering them vulnerable to spoofing by malicious hosts. To overcome these challenges, the authors propose a distributed hardware-assisted telemetry architecture that integrates DCAP remote attestation, event-level differential privacy, and SPDM-based authentication to establish a secure first-mile layer. The design further incorporates Byzantine fault tolerance and GPU enclave-based global verification to enable traceable, extraction-attack-resistant, high-resolution AI modeling of power transients. Experimental results demonstrate that the system achieves 0% success rate against post-extraction attacks across 32 GCP Confidential VMs, with a per-enclave throughput of 131,406 samples/second and an authentication overhead of merely 0.23 microseconds per sample. On H100/A100/L4 platforms, it attains a dynamic scheduling error of 1.3 MW, significantly outperforming centralized differential privacy baselines.
0 citationsRead paper
Recent publications
Latest Papers
Beyond Repository Boundaries: Cross-Repository Graph Retrieval for Code Generation
为解决代码生成中依赖环境兼容性问题，提出CrossCoder框架，通过跨仓库知识图谱整合外部库，并采用多跳检索策略增强上下文理解。
0 citationsRead paper
Variance Driven Exploration: A Provable and Efficient Methodology for Pure Exploration in Highly Stochastic Environments
本文提出了一种名为VarDE的方法，通过最小化最终决策的不确定性来解决高度随机环境中的纯探索问题，并在多个核心问题上展示了其优越性。
0 citationsRead paper
SCENARIODIFF: A Scenario-level Guidance Framework for Multimodal Time Series Forecasting--Extended Version
为了解决多模态时间序列预测中的外部事件影响问题，提出SCENARIODIFF框架，通过分层上下文推理和锚点指导生成更准确的预测。
0 citationsRead paper
When Denoising Hurts: Rethinking the Terminal Step of Diffusion Time Series Forecasters -- Extended Version
This study addresses the statistical drift and performance degradation caused by persistent low-noise denoising in diffusion-based time series forecasting by elucidating the detrimental mechanisms of over-denoising. We propose a label-free global stopping criterion and a Bernoulli time-step sampler focusing on high-noise regions to jointly optimize training sampling distributions and inference termination points. Experiments across eight real-world datasets demonstrate that this approach effectively circumvents the over-denoising trap, significantly improving prediction accuracy while accelerating inference. The proposed method achieves superior overall performance compared to existing mainstream techniques, establishing a new paradigm for the efficient application of diffusion models in time series forecasting.
0 citationsRead paper
EnclaveScale: Hardware-Assisted Edge-DP for Secure Data Centre Power Telemetry
This work addresses the limitations of existing encrypted telemetry schemes, which struggle to support high-frequency (10 Hz) power data streams and lack robust source authentication, rendering them vulnerable to spoofing by malicious hosts. To overcome these challenges, the authors propose a distributed hardware-assisted telemetry architecture that integrates DCAP remote attestation, event-level differential privacy, and SPDM-based authentication to establish a secure first-mile layer. The design further incorporates Byzantine fault tolerance and GPU enclave-based global verification to enable traceable, extraction-attack-resistant, high-resolution AI modeling of power transients. Experimental results demonstrate that the system achieves 0% success rate against post-extraction attacks across 32 GCP Confidential VMs, with a per-enclave throughput of 131,406 samples/second and an authentication overhead of merely 0.23 microseconds per sample. On H100/A100/L4 platforms, it attains a dynamic scheduling error of 1.3 MW, significantly outperforming centralized differential privacy baselines.
0 citationsRead paper