paper-with-me

Papers

Inference-Time Consensus for Mitigating Hidden Behaviors from LLM Fine-Tuning

2026-07-25 · Adhyyan Narang, Artin Tajdini, Claire Zhang, Jamie Morgenstern arxiv

Recent work shows that fine-tuning language models on even a small amount of poisoned data can install targeted misbehavior, and ostensibly benign data can transmit hidden preferences that generalize broadly. Standard defenses, such as data filtering, mixing in harmless data, and regularization, attenuate these effects but do not eliminate them. We instead pursue robustness through redundancy: collecting multiple datasets from different sources and only learning what is common between them. Thus, if only a subset of sources are malicious, the misbehavior will be blocked. In order to implement this defense strategy, we fine-tune a separate reference model on each source's dataset and aggregate their next-token distributions at decoding time. We introduce two consensus decoders: a token-wise minimum, which caps each token at the lowest probability any source assigns, and a base-relative variant, which reverts to the base probability on any token the sources move in opposing directions. We further relax exact agreement to tolerate partial support across sources and different surface expressions of the same intention. Across controlled poisoning tasks, subliminal learning, and emergent misalignment, consensus decoding suppresses source-specific misbehavior while preserving shared desirable behavior, including cases where union training and weight averaging retain the unwanted behavior.

📄 PDF Abstract BibTeX arXiv:2607.23394

Code (0)

등록된 구현이 없습니다.

Similar Papers 제목 키워드 기반

Resilient Average Consensus: A Detection and Compensation Approach

2022-02-22 · Wenzhe Zheng, Zhiyu He, Jianping He, Chengcheng Zhao 외

We study the problem of resilient average consensus for multi-agent systems with misbehaving nodes. To protect consensus valuefrom being influenced by misbehaving nodes, we address this problem by detecting misbehaviors,…

Mitigating Backdoor Threats to Large Language Models: Advancement and Challenges

2024-09-30 · Qin Liu, Wenjie Mo, Terry Tong, Jiashu Xu 외

The advancement of Large Language Models (LLMs) has significantly impacted various domains, including Web search, healthcare, and software development. However, as these models scale, they become more vulnerable to cyber…

Memorization

Self-Improving Code Generation via Semantic Entropy and Behavioral Consensus

2026-03-31 · Huan Zhang, Wei Cheng, Wei Hu arxiv

Improving the code generation capabilities of large language models (LLMs) typically relies on supervised fine-tuning or preference optimization, both of which require costly external resources such as powerful teacher m…

Code Generation

Causal Intervention for Fairness in Multi-behavior Recommendation

2022-09-10 · Xi Wang, Wenjie Wang, Fuli Feng, Wenge Rong 외

Recommender systems usually learn user interests from various user behaviors, including clicks and post-click behaviors (e.g., like and favorite). However, these behaviors inevitably exhibit popularity bias, leading to s…

FairnessRecommendation Systems

Steering Frozen LLMs: Adaptive Social Alignment via Online Prompt Routing

2026-03-04 · Zeyu Zhang, Xiangxiang Dai, Ziyi Han, Xutong Liu 외 arxiv

Large language models (LLMs) are typically governed by post-training alignment (e.g., RLHF or DPO), which yields a largely static policy during deployment and inference. However, real-world safety is a full-lifecycle pro…