Selected Publications

Towards Personalized Language Models via Inference-time Human Preference Optimization

(NeurIPS 2024 AFM) We introduce a novel approach to LLM alignment for personalized preference based on decode-time frameworks.

Nikki Lijing Kuang, Wei Sun, Scott McFaddin, Markus Ettl, Yi-An Ma

Towards Personalized Language Models via Inference-time Human Preference Optimization

Diff-BBO: Diffusion-Based Inverse Modeling for Black-Box Optimization

(NeurIPS 2024 BDU) We propose an inverse modeling approach for efficient online black-box optimization by resorting to classifier-free conditional diffusion models with a novel uncertainty-aware acquisition function.

Dongxia Wu*, Nikki Lijing Kuang*, Ruijia Niu, Yi-An Ma, Rose Yu

Log-concave Sampling from a Convex Body with a Barrier: a Robust and Unified Dikin Walk

(NeurIPS 2024) We design a Dikin walk for log-concave sampling over polytopes and spectrahedra with optimal mixing time and efficient per-iteration cost.

(Alphabetical) Yuzhou Gu, Nikki Lijing Kuang, Yi-An Ma, Zhao Song, Lichen Zhang

Log-concave Sampling from a Convex Body with a Barrier: a Robust and Unified Dikin Walk

Towards Robust Estimation of Intention Hierarchy in Robot Teleoperation

(NeurIPS 2024 Workshop) We propose a novel architecture to reason about the intentions of the human partner in assistive robot teleoperation through non-verbal observations.

Nikki Lijing Kuang, Soshi Iba, Songpo Li

Posterior sampling with delayed feedback for reinforcement learning with linear function approximation

(NeurIPS 2023) We provide the first theoretical analysis for the class of posterior sampling algorithms to handle delayed feedback in RL frameworks.

Nikki Lijing Kuang*, Ming Yin*, Mengdi Wang, Yu-Xiang Wang, Yi-An Ma

Langevin Thompson sampling with logarithmic communication: bandits and reinforcement learning

(ICML 2023) We study approximate Thompson Sampling with Markov Chain Monte Carlo in bandit and reinforcement learning frameworks, providing algorithms that achieve optimal performance with low computation and communication cost.

Nikki Lijing Kuang*, Siddharth Mitra*, Amin Karbasi, Yi-An Ma

Statistical and computational trade-offs in variational inference: A case study in inferential model selection

(Preprint) We provide a rigorous study of the statistical and computational trade-offs for variational inference (VI).

Nikki Lijing Kuang*, Kush Bhatia*, Yian Ma*, Yixin Wang*

Statistical and computational trade-offs in variational inference: A case study in inferential model selection

Sample Efficient Constrained Reinforcement Learning

We propose sample efficient algorithms for constrained RL frameworks.

Nikki Lijing Kuang

Sample Efficient Constrained Reinforcement Learning