Searcharxiv⌕ Search

arXiv subjects

Ziqi Zhong

Publications and source records attributed to Ziqi Zhong.

4 recordsLinked to original sources

Hallucination-Guided Unlearning: Using Hallucination Traces to Reveal Overfitted Memories

Selective unlearning is increasingly important for the safety and compliance of large language models. Harmful or memorized content is often entangled with useful knowledge and difficult to enumerate. We use the model's own failures to identify candidates for unlearning. We propose Hallucination-Guided Unlearning (HGU), which uses hallucinations as diagnostic signals. HGU localizes unsupported spans via evidence and internal consensus, traces a sparse set of high-responsibility MLP neurons, and edits them with compact gates under a utility-preserving KL constraint. On TOFU, HGU achieves the best MU-FE trade-off, with Avg scores of 0.7221 (forget05) and 0.7699 (forget10), and the highest SepS H-Avg (0.4382/0.4715) under mixed forget/retain prompts. On the WMDP auxiliary task, HGU yields the strongest forgetting (e.g., 14.2 on Economics) while improving Retain (52.7) and MMLU (57.2). These results suggest that tracing hallucinations enables targeted, post-hoc unlearning that better balances forgetting efficacy and general utility without full retraining.

cs.AI↗

Seeing through the Conflict: Transparent Knowledge Conflict Handling in Retrieval-Augmented Generation

Large language models (LLMs) equipped with retrieval--the Retrieval-Augmented Generation (RAG) paradigm--should combine their parametric knowledge with external evidence, yet in practice they often hallucinate, over-trust noisy snippets, or ignore vital context. We introduce TCR (Transparent Conflict Resolution), a plug-and-play framework that makes this decision process observable and controllable. TCR (i) disentangles semantic match and factual consistency via dual contrastive encoders, (ii) estimates self-answerability to gauge confidence in internal memory, and (iii) feeds the three scalar signals to the generator through a lightweight soft-prompt with SNR-based weighting. Across seven benchmarks TCR improves conflict detection (+5-18 F1), raises knowledge-gap recovery by +21.4 pp and cuts misleading-context overrides by -29.3 pp, while adding only 0.3% parameters. The signals align with human judgements and expose temporal decision patterns.

cs.AI↗

ReasonBridge: Efficient Reasoning Transfer from Closed to Open-Source Language Models

Recent advancements in Large Language Models (LLMs) have revealed a significant performance gap between closed-source and open-source models, particularly in tasks requiring complex reasoning and precise instruction following. This paper introduces ReasonBridge, a methodology that efficiently transfers reasoning capabilities from powerful closed-source to open-source models through a novel hierarchical knowledge distillation framework. We develop a tailored dataset Reason1K with only 1,000 carefully curated reasoning traces emphasizing difficulty, diversity, and quality. These traces are filtered from across multiple domains using a structured multi-criteria selection algorithm. Our transfer learning approach incorporates: (1) a hierarchical distillation process capturing both strategic abstraction and tactical implementation patterns, (2) a sparse reasoning-focused adapter architecture requiring only 0.3% additional trainable parameters, and (3) a test-time compute scaling mechanism using guided inference interventions. Comprehensive evaluations demonstrate that ReasonBridge improves reasoning capabilities in open-source models by up to 23% on benchmark tasks, significantly narrowing the gap with closed-source models. Notably, the enhanced Qwen2.5-14B outperforms Claude-Sonnet3.5 on MATH500 and matches its performance on competition-level AIME problems. Our methodology generalizes effectively across diverse reasoning domains and model architectures, establishing a sample-efficient approach to reasoning enhancement for instruction following.

cs.AI↗

MANTA: Cross-Modal Semantic Alignment and Information-Theoretic Optimization for Long-form Multimodal Understanding

While multi-modal learning has advanced significantly, current approaches often treat modalities separately, creating inconsistencies in representation and reasoning. We introduce MANTA (Multi-modal Abstraction and Normalization via Textual Alignment), a theoretically-grounded framework that unifies visual and auditory inputs into a structured textual space for seamless processing with large language models. MANTA addresses four key challenges: (1) semantic alignment across modalities with information-theoretic optimization, (2) adaptive temporal synchronization for varying information densities, (3) hierarchical content representation for multi-scale understanding, and (4) context-aware retrieval of sparse information from long sequences. We formalize our approach within a rigorous mathematical framework, proving its optimality for context selection under token constraints. Extensive experiments on the challenging task of Long Video Question Answering show that MANTA improves state-of-the-art models by up to 22.6% in overall accuracy, with particularly significant gains (27.3%) on videos exceeding 30 minutes. Additionally, we demonstrate MANTA's superiority on temporal reasoning tasks (23.8% improvement) and cross-modal understanding (25.1% improvement). Our framework introduces novel density estimation techniques for redundancy minimization while preserving rare signals, establishing new foundations for unifying multimodal representations through structured text.

cs.CV↗