Selected work

Representative Papers

Fairis: Fairness-Aware Aggregation with Provable Influence Containment against Fairness Poisoning Attacks in Collaborative Machine Learning

Aug 06, 2026

This work addresses fairness poisoning attacks in collaborative machine learning, where malicious clients degrade group fairness while preserving model accuracy to evade accuracy-based defenses. To counter this threat, the authors propose Fairis, a server-side dynamic reweighting mechanism that adjusts client update weights based on their local fairness metrics—normalized using Equal Opportunity Difference—and integrates norm clipping with a safety parameter η for robust control. Fairis provides the first provable bound on the impact of fairness poisoning attacks and offers three theoretical guarantees: monotonic weight decay, demographic participation, and non-manipulability. It also resists collusion by a minority of adversarial clients. Experiments on the Taiwanese Credit dataset demonstrate that Fairis reduces the influence of stealthy attackers by 41%–54% while consistently assigning positive weights to honest clients, substantially outperforming existing approaches.

0 citationsRead paper

LLM-CEG: Extending the Classification Error Gauge Framework for Privacy Auditing of Large Language Models

Apr 26, 2026

This work addresses the absence of a systematic privacy auditing framework for large language models (LLMs), which hinders the quantification of trade-offs between privacy and utility. The study extends the classification error gap (x-CEG) framework to the LLM setting, introducing LLM-CEG alongside an accompanying engineering pipeline, LLM-SIED. This approach integrates differentially private stochastic gradient descent (DP-SGD) fine-tuning, membership inference attacks to assess privacy leakage, and perplexity to evaluate utility, thereby enabling auditable and regulation-aligned privacy-compliant deployment. Experiments on DistilGPT-2 with synthetic clinical personally identifiable information (PII) demonstrate that the proposed method reduces attack success rates by 71.5% while improving out-of-distribution utility by 47–50%, confirming the implicit regularization effect of differential privacy under specific fine-tuning conditions.

0 citationsRead paper

GPU Programming for AI Workflow Development on AWS SageMaker: An Instructional Approach

Sep 17, 2025

Current AI agent development curricula inadequately address GPU architecture and programming competencies, limiting students’ ability to optimize compute-intensive AI workflows. Method: We designed and delivered a specialized GPU architecture and programming course for undergraduate and master’s students, centered on building a Retrieval-Augmented Generation (RAG) system. The curriculum integrates CUDA programming, HPC performance analysis, and hands-on deployment on AWS SageMaker, forming a cohesive pedagogical pipeline: architectural understanding → parallel implementation → AI workflow deployment → performance optimization. Contribution/Results: This work pioneers the deep integration of low-level GPU programming with AI agent development in an academic setting. Leveraging cloud infrastructure enables scalable, low-cost experiential learning. Evaluation demonstrates significant improvements in students’ parallel computing proficiency, GPU-accelerated implementation skills, and capacity to solve large-scale problems. The course effectively exposes common performance bottlenecks, while validating AWS SageMaker as an economical and feasible platform for GPU-centric education—offering a reusable pedagogical framework for computational skill development in STEM.

0 citationsRead paper

VaCDA: Variational Contrastive Alignment-based Scalable Human Activity Recognition

May 08, 2025

Wearable sensor data exhibit strong domain heterogeneity across users, devices, and wearing positions, while annotation remains costly. To address these challenges, this paper proposes a multi-source domain adaptation framework for human activity recognition (HAR). Our method uniquely integrates variational autoencoders (VAEs) with contrastive learning to construct a shared low-dimensional latent space, enabling unsupervised intra-class compactness alignment and inter-class separability enhancement simultaneously—thereby mitigating cross-domain distribution shifts. Crucially, the approach operates without target-domain labels and supports joint knowledge transfer from multiple source domains. Extensive experiments on benchmark public datasets demonstrate that our method significantly outperforms state-of-the-art baselines in both cross-position and cross-device scenarios, achieving an average accuracy improvement of over 5.2%.

0 citationsRead paper

Recent publications

Latest Papers

Fairis: Fairness-Aware Aggregation with Provable Influence Containment against Fairness Poisoning Attacks in Collaborative Machine Learning

Aug 06, 2026

This work addresses fairness poisoning attacks in collaborative machine learning, where malicious clients degrade group fairness while preserving model accuracy to evade accuracy-based defenses. To counter this threat, the authors propose Fairis, a server-side dynamic reweighting mechanism that adjusts client update weights based on their local fairness metrics—normalized using Equal Opportunity Difference—and integrates norm clipping with a safety parameter η for robust control. Fairis provides the first provable bound on the impact of fairness poisoning attacks and offers three theoretical guarantees: monotonic weight decay, demographic participation, and non-manipulability. It also resists collusion by a minority of adversarial clients. Experiments on the Taiwanese Credit dataset demonstrate that Fairis reduces the influence of stealthy attackers by 41%–54% while consistently assigning positive weights to honest clients, substantially outperforming existing approaches.

0 citationsRead paper

LLM-CEG: Extending the Classification Error Gauge Framework for Privacy Auditing of Large Language Models

Apr 26, 2026

This work addresses the absence of a systematic privacy auditing framework for large language models (LLMs), which hinders the quantification of trade-offs between privacy and utility. The study extends the classification error gap (x-CEG) framework to the LLM setting, introducing LLM-CEG alongside an accompanying engineering pipeline, LLM-SIED. This approach integrates differentially private stochastic gradient descent (DP-SGD) fine-tuning, membership inference attacks to assess privacy leakage, and perplexity to evaluate utility, thereby enabling auditable and regulation-aligned privacy-compliant deployment. Experiments on DistilGPT-2 with synthetic clinical personally identifiable information (PII) demonstrate that the proposed method reduces attack success rates by 71.5% while improving out-of-distribution utility by 47–50%, confirming the implicit regularization effect of differential privacy under specific fine-tuning conditions.

0 citationsRead paper

GPU Programming for AI Workflow Development on AWS SageMaker: An Instructional Approach

Sep 17, 2025

Current AI agent development curricula inadequately address GPU architecture and programming competencies, limiting students’ ability to optimize compute-intensive AI workflows. Method: We designed and delivered a specialized GPU architecture and programming course for undergraduate and master’s students, centered on building a Retrieval-Augmented Generation (RAG) system. The curriculum integrates CUDA programming, HPC performance analysis, and hands-on deployment on AWS SageMaker, forming a cohesive pedagogical pipeline: architectural understanding → parallel implementation → AI workflow deployment → performance optimization. Contribution/Results: This work pioneers the deep integration of low-level GPU programming with AI agent development in an academic setting. Leveraging cloud infrastructure enables scalable, low-cost experiential learning. Evaluation demonstrates significant improvements in students’ parallel computing proficiency, GPU-accelerated implementation skills, and capacity to solve large-scale problems. The course effectively exposes common performance bottlenecks, while validating AWS SageMaker as an economical and feasible platform for GPU-centric education—offering a reusable pedagogical framework for computational skill development in STEM.

0 citationsRead paper

VaCDA: Variational Contrastive Alignment-based Scalable Human Activity Recognition

May 08, 2025

Wearable sensor data exhibit strong domain heterogeneity across users, devices, and wearing positions, while annotation remains costly. To address these challenges, this paper proposes a multi-source domain adaptation framework for human activity recognition (HAR). Our method uniquely integrates variational autoencoders (VAEs) with contrastive learning to construct a shared low-dimensional latent space, enabling unsupervised intra-class compactness alignment and inter-class separability enhancement simultaneously—thereby mitigating cross-domain distribution shifts. Crucially, the approach operates without target-domain labels and supports joint knowledge transfer from multiple source domains. Extensive experiments on benchmark public datasets demonstrate that our method significantly outperforms state-of-the-art baselines in both cross-position and cross-device scenarios, achieving an average accuracy improvement of over 5.2%.

0 citationsRead paper