Papers by Kevin Du
Activation Scaling for Steering and Interpreting Language Models (2024.findings-emnlp)
Copied to clipboard
| Challenge: | a successful intervention should flip the correct with the wrong token, while remaining sparse. |
| Approach: | They propose to use activation scaling to flip the correct with the wrong token . they use gradient-based optimization to learn and evaluate a specific kind of efficient intervention . |
| Outcome: | The proposed method performs comparable with steering vectors but is much less minimal. |
Context versus Prior Knowledge in Language Models (2024.acl-long)
Copied to clipboard
| Challenge: | Existing studies have investigated how often a model will rely on prior knowledge over conflicting contextual information in answering questions. |
| Approach: | They propose two mutual information-based metrics to measure a model’s dependency on a context and on its prior about an entity. |
| Outcome: | The proposed metrics show that language models can integrate prior knowledge and new information in a predictable way across different questions and contexts. |
It’s Not What You Say, It’s How You Say It: Evaluating LLM Responses to Expressions of Belief (2026.acl-long)
Copied to clipboard
| Challenge: | a typology is grounded in four linguistically motivated dimensions: form, evidentiality, epistemic stance, and tone. |
| Approach: | They propose a typology to evaluate how different EoBs affect whether models follow context versus prior knowledge. |
| Outcome: | The proposed model systematically evaluates 16 LLMs that differ in architecture, scale, and training stages . human listeners subconsciously interpret the belief based on how it is expressed, i.e., its explicitness, tone, or contextual cues. |
Speech-to-Speech Translation for a Real-world Unwritten Language (2023.findings-acl)
Copied to clipboard
Peng-Jen Chen, Kevin Tran, Yilin Yang, Jingfei Du, Justine Kao, Yu-An Chung, Paden Tomasello, Paul-Ambroise Duquenne, Holger Schwenk, Hongyu Gong, Hirofumi Inaguma, Sravya Popuri, Changhan Wang, Juan Pino, Wei-Ning Hsu, Ann Lee
| Challenge: | a new study examines speech-to-speech translation (S2ST) that translates speech from one language into another . the research area for unwritten languages remains a research area with little exploration due to the lack of training data. |
| Approach: | They propose a system that translates speech from one language into another . they use Taiwanese Hokkien as an example of an unwritten language . |
| Outcome: | The proposed system can be used to train models in languages without standard writing systems. |
How Persuasive Is Your Context? (2025.emnlp-main)
Copied to clipboard
| Challenge: | Empirically, through aseries of experiments, we show that TPS captures a more nuanced notion of persuasiveness than previously proposed metrics. |
| Approach: | They introduce a targeted persuasion score to quantify how persuasive a given context is to an LM. |
| Outcome: | Empirically, the proposed model captures a more nuanced notion of persuasiveness than previously proposed metrics. |
Efficiently Computing Susceptibility to Context in Language Models (2024.findings-emnlp)
Copied to clipboard
| Challenge: | a current language model is able to incorporate information from a user-input context when answering queries, but it is not equally sensitive to subtle changes to that context. |
| Approach: | They propose a metric to quantify the degree to which contexts can influence a model’s response to a query at a distributional level. |
| Outcome: | The proposed method is comparable to Monte Carlo's estimated susceptibility across a diverse set of query domains despite being 70 faster. |
An Experimental Design Framework for Label-Efficient Supervised Finetuning of Large Language Models (2024.findings-acl)
Copied to clipboard
Gantavya Bhatt, Yifang Chen, Arnav Das, Jifan Zhang, Sang Truong, Stephen Mussmann, Yinglun Zhu, Jeff Bilmes, Simon Du, Kevin Jamieson, Jordan Ash, Robert Nowak
| Challenge: | Supervised finetuning (SFT) on instruction datasets has shown immense potential in improving the zero-shot generalization capabilities observed in large language models (LLMs). |
| Approach: | They propose to use experimental design to minimize the computational cost of active learning by identifying useful subsets of samples to annotate from an unlabeled pool. |
| Outcome: | The proposed methods save 50% of the annotation cost compared to random sampling on generative tasks. |
Generalizing Backpropagation for Gradient-Based Interpretability (2023.acl-long)
Copied to clipboard
| Challenge: | Several feature-attribution methods for interpreting deep neural networks rely on computing the gradients of a model’s output with respect to its inputs, but they reveal little about the inner workings of the model itself. |
| Approach: | They propose a generalized backpropagation algorithm that generalizes the gradient computation of a model to efficiently compute other interpretable statistics about the gradient graph of neural networks. |
| Outcome: | The proposed generalized algorithm can be used to compute other interpretable statistics about the gradient graph of a neural network, such as the highest-weighted path and entropy. |