Papers by Xiaotian Pan
Pruning Weights but Not Truth: Safeguarding Truthfulness While Pruning LLMs (2025.findings-emnlp)
Copied to clipboard
| Challenge: | Neural network pruning disrupts LLMs’ internal activation features crucial for lie detection . layer-wise pruning sparsity inadvertently removes crucial weights, failing to improve lie detection performance despite its reliance on the most crucial LLM layer. |
| Approach: | They propose a pruning approach that places greater emphasis on layers with more activation outliers and stronger discriminative features simultaneously. |
| Outcome: | The proposed approach improves the hallucination detection for pruned LLMs (achieving 88% accuracy at 50% sparsity) and enhances their performance on TruthfulQA. |
GOBench: Stage-Wise Diagnostics and the Visual Paradox in Multimodal Graph Optimization (2026.findings-acl)
Copied to clipboard
| Challenge: | Existing benchmarks fail to represent multimodal problem specifications, score outcomes only and cannot localize where failures occur along the modeling pipeline. |
| Approach: | They propose a Graph Optimization benchmark that aligns multiple modalities with solver-derived oracles and a diagnostic protocol that evaluates intermediate artifacts as well as end results. |
| Outcome: | Graph Optimization benchmark (GOBench) evaluates intermediate artifacts as well as end results . vision reliably increases inference cost, while reliability impact is regime-dependent . current benchmarks fail to represent multimodal problem specifications, fail to localize failures . |
When Truthful Representations Flip Under Deceptive Instructions? (2025.emnlp-main)
Copied to clipboard
| Challenge: | Large language models (LLMs) follow maliciously crafted instructions to generate deceptive responses, posing safety challenges. |
| Approach: | They use Sparse Autoencoders to analyze LLM's internal representations to determine when and how they "flip" from truthful to deceptive under deceptively crafted instructions. |
| Outcome: | The proposed model's True/False output is predictable across all conditions based on the model''s representation, and the Deceptive instructions induce significant representational shifts compared to Truthful/Neutral representations. |
Quantized but Deceptive? A Multi-Dimensional Truthfulness Evaluation of Quantized LLMs (2025.emnlp-main)
Copied to clipboard
| Challenge: | Quantization enables efficient deployment of large language models in resource-constrained environments . but impact on truthfulness remains largely unexplored . |
| Approach: | They propose a framework to assess the truthfulness of quantized large language models . they find quantized models retain internally truthful representations but produce false outputs . |
| Outcome: | The framework assesses the truthfulness of quantized models across three dimensions . it finds that quantized model models retain internally truthful representations but are more susceptible to false outputs . |