arXiv subjects

Md Rayhanul Masud

Publications and source records attributed to Md Rayhanul Masud.

3 recordsLinked to original sources

arXiv2026

TAME: Token Attribution and Masking for Emergent misalignment

Fine-tuning an aligned language model on narrow, flawed data can induce harmful behavior far outside the training domain, known as emergent misalignment (EM). Prior work has localized EM in model weights, activations, and training documents, but it remains unclear which training tokens carry the relevant fine-tuning signal. We introduce TAME (Token Attribution and Masking for Emergent Misalignment), a three-stage framework: token attribution scores how strongly the fine-tuning update raises each response token's likelihood, using forward passes through a released LoRA adapter; signal characterization finds patterns among high-attribution tokens; and causal validation tests them by attribution-guided loss masking. On released EM organisms and a 6,849-example medical-advice split, attribution is concentrated (the top 5% of tokens hold 32% of the mass) and, in Llama, depleted for medical vocabulary but enriched for a register of unwarranted certainty, even after controlling for token rarity. Masking high-attribution tokens during fresh fine-tuning cuts EM by 23x in Llama and 36x in Qwen, with the perplexity cost concentrated on the targeted register rather than on medical content; an equal random mask leaves EM unchanged. In Llama, the attribution pattern suggests that EM-relevant signal lies more in how confidently flawed content is expressed than in its domain vocabulary; the causal masking effect itself holds across both model families.

cs.LG↗

arXiv2026

Reward Engineering for Software Tasks: A Survey of Reinforcement Learning Approaches

Reinforcement learning is increasingly used for code-centric software engineering tasks, including code generation, understanding, repair, testing, and optimization, especially with the rise of large language models and autonomous agents. A core challenge in these settings is reward design. Unlike standard RL domains with clear scalar objectives, software tasks involve competing goals such as correctness, security, efficiency, and readability, which are difficult to capture with a single reward. As a result, RL-for-SE systems rely on heterogeneous signals, including compilation results, unit tests, coverage metrics, retrieval scores, and learned preferences. Yet this work remains scattered across tasks and communities. This survey provides the first systematic review of reward engineering for RL in software tasks. We organize prior work by reward source, granularity, and aggregation. We then distill the findings into practical guidance, including a decision matrix, a design guide, and a reward engineering reporting standard for future RL-for-SE systems.

cs.SE↗

arXiv2024

Unveiling A Hidden Risk: Exposing Educational but Malicious Repositories in GitHub

Are malicious repositories hiding under the educational label in GitHub? Recent studies have identified collections of GitHub repositories hosting malware source code with notable collaboration among the developers. Thus, analyzing GitHub repositories deserves inevitable attention due to its open-source nature providing easy access to malicious software code and artifacts. Here we leverage the capabilities of ChatGPT in a qualitative study to annotate an educational GitHub repository based on maliciousness of its metadata contents. Our contribution is twofold. First, we demonstrate the employment of ChatGPT to understand and annotate the content published in software repositories. Second, we provide evidence of hidden risk in educational repositories contributing to the opportunities of potential threats and malicious intents. We carry out a systematic study on a collection of 35.2K GitHub repositories claimed to be created for educational purposes only. First, our study finds an increasing trend in the number of such repositories published every year. Second, 9294 of them are labeled by ChatGPT as malicious, and further categorization of the malicious ones detects 14 different malware families including DDoS, keylogger, ransomware and so on. Overall, this exploratory study flags a wake-up call for the community for better understanding and analysis of software platforms.

cs.SE↗