Papers by Dylan Zhang

6 papers
Diversification Catalyzes Language Models’ Instruction Generalization To Unseen Semantics (2025.findings-acl)

Copied to clipboard

Challenge: Instruction-tuned language models excel in knowledge, reasoning, and instruction-following . however, the factors enabling generalization to unseen instructions remain underexplored .
Approach: They propose to model instruction-following as a computational process and design controlled experiments inspired by the Turing-complete Markov algorithm to disentangle its dynamics.
Outcome: The proposed model outperforms scaling up data volumes in generalist models by combining in-domain and diverse out-of-domain tasks.
Improving Influence-based Instruction Tuning Data Selection for Balanced Learning of Diverse Capabilities (2025.findings-emnlp)

Copied to clipboard

Challenge: Influence-based methods show promise in achieving (1), but often struggle with (2) . data selection is often biased towards high-influence tasks, harming performance on them .
Approach: They propose a Balanced and Influential Data Selection algorithm that normalizes influence scores of training data and iteratively chooses the training example with the highest influence on the most underrepresented task.
Outcome: The proposed model outperforms both state-of-the-art influence-based methods and non-influence-based frameworks on seven benchmarks spanning five diverse capabilities.
Visual Prompting in LLMs for Enhancing Emotion Recognition (2024.emnlp-main)

Copied to clipboard

Challenge: Existing methods for enhancing in-context emotion classification fail to include spatial relationships between different people and facial features within a single face.
Approach: They propose a set-of-vision prompting approach that uses spatial information to mark targets precisely.
Outcome: The proposed approach improves face count and emotion categorization while preserving the enriched image context.
ScaleBiO: Scalable Bilevel Optimization for LLM Data Reweighting (2025.acl-long)

Copied to clipboard

Challenge: Existing paradigms for bilevel optimization require second-order information, making it difficult to scale them up.
Approach: They propose a scalable instantiation of a bilevel optimization paradigm for large-scale LLMs by using a memory-efficient training technique.
Outcome: The proposed paradigm scales to 30B-sized LLMs on 8H100 GPUs.
Building A Proof-Oriented Programmer That Is 64% Better Than GPT-4o Under Data Scarcity (2025.findings-acl)

Copied to clipboard

Challenge: Existing proof-oriented programming languages struggle due to data scarcity . existing proof-based models struggle due a lack of sufficient corpora for proof-orientated programming languages such as F* .
Approach: They propose a method that synthesizes proof-oriented programming problems for proficiency in a language and incorporates diverse coding data for reasoning capability elicitation.
Outcome: The proposed method outperforms existing proof-oriented models in function- and repository-level proof-based programming by 64% relative margin and improves GPT-4o's performance by 54% by repairing outputs over GPT-4)
LMOD: A Large Multimodal Ophthalmology Dataset and Benchmark for Large Vision-Language Models (2025.findings-naacl)

Copied to clipboard

Challenge: Existing benchmarks for large vision-language models (LVLMs) are limited to ophthalmology-specific applications.
Approach: They introduce a large-scale multimodal ophthalmology benchmark consisting of 21,993 instances across five ocular imaging modalities and 13 state-of-the-art LVLM representatives from closed-source, open-source and medical domains.
Outcome: The proposed model shows significant performance drop in ophthalmology compared to other domains.

What is GenGO?

GenGO is an NLP powered publication search system. It currenctly indexes 30k+ papers from ACL Anthology, and implements multi-aspect summarization, semantic search, and more!

Information

About
Limitations