Papers by Thomas Lu
Efficient Knowledge Editing via Minimal Precomputation (2025.acl-short)
Copied to clipboard
| Challenge: | Knowledge editing methods like MEMIT require a one-time but significant computational cost. |
| Approach: | They propose to pre-compute 44 million hidden vectors per edited layer . authors show that this precomputation step is unnecessary . |
| Outcome: | The proposed methods can be performed by pre-computing a small portion of 44 million hidden vectors. |
Learned Incremental Representations for Parsing (2022.acl-long)
Copied to clipboard
| Challenge: | a new syntactic representation that commits to syntakic choices is proposed for humans . we use a system that uses only incremental processing of a prefix to predict the word in a sentence . |
| Approach: | They propose a syntactic representation that commits to syntakic choices incrementally . they say the system can achieve 93.72 F1 on the Penn Treebank with as few as 5 bits per word . |
| Outcome: | The proposed representation achieves 93.72 F1 on the Penn Treebank with as few as 5 bits per word . the analysis of the representations shows they have interpretable features and deferred resolution of syntactic ambiguities. |
ToolSandbox: A Stateful, Conversational, Interactive Evaluation Benchmark for LLM Tool Use Capabilities (2025.findings-naacl)
Copied to clipboard
Jiarui Lu, Thomas Holleis, Yizhe Zhang, Bernhard Aumayer, Feng Nan, Haoping Bai, Shuang Ma, Shen Ma, Mengyu Li, Guoli Yin, Zirui Wang, Ruoming Pang
| Challenge: | Recent advances in large language models have led to a growing interest in tool assisted LLMs . toolSandbox includes stateful tool execution, implicit state dependencies between tools . |
| Approach: | a new tool-based evaluation tool is released to help LLMs evaluate their tool-use capabilities. a tool-driven evaluation tool includes stateful tool execution, implicit state dependencies between tools and a built-in user simulator. |
| Outcome: | the toolSandbox evaluation benchmark shows that open source and proprietary models have a performance gap . the benchmarks show that even the most capable LLMs are challenged by state dependent tasks . |
Causal Distillation for Language Models (2022.naacl-main)
Copied to clipboard
Zhengxuan Wu, Atticus Geiger, Joshua Rozner, Elisa Kreiss, Hanson Lu, Thomas Icard, Christopher Potts, Noah Goodman
| Challenge: | Distillation efforts have led to language models that are more compact and efficient without serious drops in performance. |
| Approach: | They propose to augment distillation with a third objective that encourages the student model to imitate the causal dynamics of the teacher through a distillation interchange intervention training objective (DIITO). |
| Outcome: | The proposed method lowers perplexity on the WikiText-103M corpus and improves on the GLUE benchmark, SQuAD, and CoNLL-2003. |
Paired by the Teacher: Turning Unpaired Data into High-Fidelity Pairs for Low-Resource Text Generation (2025.emnlp-main)
Copied to clipboard
| Challenge: | a low-resource natural language generation task requires a large number of examples to generate outputs and outputs. |
| Approach: | They propose a teacher-student pipeline that synthesizes accurate input–output pairs without human labels or parallel data. |
| Outcome: | The proposed pipeline synthesizes accurate input–output pairs without human labels or parallel data. |
Utilizing Semantic Textual Similarity for Clinical Survey Data Feature Selection (2025.findings-acl)
Copied to clipboard
| Challenge: | Survey data often contains many features that lead to high-dimensional inputs for machine learning models. |
| Approach: | They propose to use semantic textual similarity scores to score features in surveys . they compare STS scores to a statistically-based scoring approach . |
| Outcome: | The proposed method can be used in clinical surveys to score features . it can be applied to a clinical dataset and to NIH All of Us dataset . |
Lifelong Knowledge Editing requires Better Regularization (2025.findings-emnlp)
Copied to clipboard
Akshat Gupta, Phudish Prateepamornkul, Maochuan Lu, Ahmed Alaa, Thomas Hartvigsen, Gopala Anumanchipalli
| Challenge: | Knowledge editing is a promising way to improve factuality in large language models, but recent studies have shown significant model degradation during sequential editing. |
| Approach: | They formalize locate-then-edit methods as a two-step fine-tuning process . they show that model degradation occurs due to over-optimization of internal activations . |
| Outcome: | The proposed methods reduce time and improve factuality by 42-61%. |