Conference

We propose ReShift, the first aha-moment-driven reasoning-level backdoor framework for Vision–Language Models that redirects chain-of-thought trajectories while preserving surface-level coherence.

An empirical study of 547 real-world operational safety failures in LLM-based coding agents, revealing a taxonomy of 33 risk types and showing that over 65% of incidents arise during routine bug fixing and configuration tasks.

An empirical study of transforming legacy SRS documents into traceable synthetic pre-SRS artifacts for fine-grained evaluation of LLM-based requirements specification generation.

We propose PTA-GRPO, a two-stage framework that improves LLM reasoning by combining high-level planning guidance with guidance-aware reinforcement learning.

We propose a simulation-based framework called FairSense to detect and analyze long-term unfairness in ML-enabled systems.

We show that GitHub Copilot can generate code with the symptoms of SATD, both prompted and unprompted. Moreover, we demonstrate the tool’s ability to automatically repay SATD under different circumstances and qualitatively investigate the characteristics of successful and unsuccessful comments.

We proposed Fairify, an approach to make individual fairness verification tractable for the developers. The key idea is that many neurons in the NN always remain inactive when a smaller part of the input domain is considered. So, Fairify leverages white-box access to the models in production and then apply formal analysis based pruning.

We comprehensively study popular real-world ensembles: bagging, boosting, stacking and voting. We have developed a benchmark of 168 ensemble models collected from Kaggle on four popular fairness datasets. We use existing fairness metrics to understand the composition of fairness. Our results show that ensembles can be designed to be fairer without using mitigation techniques. We also identify the interplay between fairness composition and data characteristics to guide fair ensemble design.

Our approach includes two key innovations: a novel optimization function and a fairness-aware search space. By improving the default optimization function of AutoML and incorporating fairness objectives, we are able to mitigate bias with little to no loss of accuracy. Additionally, we propose a fairness-aware search space pruning method for AutoML to reduce computational cost and repair time.

We highlight the safety risks posed by feedback loops in ML-based systems, discuss why they are hard to detect with current software engineering methods, and propose research directions to better identify, monitor, and mitigate these effects.