10 papers
Quantifying the Stability of Multi-Step Reasoning via Error Amplification
Dongyue Li, Ziniu Zhang, Minxuan Duan +1
We consider the stability of multi-step reasoning processes, which have extensive applications in language models, including chain-of-thought and algorithmic reasoning. While longe…
MoPLEx: Estimating Plackett-Luce Mixture Models for Multi-Objective Alignment
Dongyue Li, Ziniu Zhang, Lu Wang +1
We study learning a mixture of Plackett-Luce models from multi-way ranking responses from annotators that may represent heterogeneous underlying preferences. This problem has m…
Efficient Estimation of Kernel Surrogate Models for Task Attribution
Zhenshuo Zhang, Minxuan Duan, Hongyang R. Zhang
Modern AI agents such as large language models are trained on diverse tasks -- translation, code generation, mathematical reasoning, and text prediction -- simultaneously. A key qu…
One-Sided Matrix Completion from Ultra-Sparse Samples
Hongyang R. Zhang, Zhenshuo Zhang, Huy L. Nguyen +1
Matrix completion is a classical problem that has received recurring interest across a wide range of fields. In this paper, we revisit this problem in an ultra-sparse sampling regi…
Learning Multimodal Embeddings for Traffic Accident Prediction and Causal Estimation
Ziniu Zhang, Minxuan Duan, Haris N. Koutsopoulos +1
We consider analyzing traffic accident patterns using both road network data and satellite images aligned to road graph nodes. Previous work for predicting accident occurrences rel…
Efficiently Learning Branching Networks for Multitask Algorithmic Reasoning
Dongyue Li, Zhenshuo Zhang, Minxuan Duan +2
Algorithmic reasoning -- the ability to perform step-by-step logical inference -- is a synthetic benchmark for evaluating multi-step reasoning abilities, designed for graph neural…
Scalable Multi-Objective and Meta Reinforcement Learning via Gradient Estimation
Zhenshuo Zhang, Minxuan Duan, Youran Ye +1
We study the problem of efficiently estimating policies that simultaneously optimize multiple objectives in reinforcement learning (RL). Given objectives (or tasks), we seek th…
Linear-Time Demonstration Selection for In-Context Learning via Gradient Estimation
Ziniu Zhang, Zhenshuo Zhang, Dongyue Li +3
This paper introduces an algorithm to select demonstration examples for in-context learning of a query set. Given a set of examples, how can we quickly select out of to…
Efficient Ensemble for Fine-tuning Language Models on Multiple Datasets
Dongyue Li, Ziniu Zhang, Lu Wang +1
This paper develops an ensemble method for fine-tuning a language model to multiple datasets. Existing methods, such as quantized LoRA (QLoRA), are efficient when adapting to a sin…
Scalable Fine-tuning from Multiple Data Sources: A First-Order Approximation Approach
Dongyue Li, Ziniu Zhang, Lu Wang +1
We study the problem of fine-tuning a language model (LM) for a target task by optimally using the information from auxiliary tasks. This problem has broad applications in NLP,…