5 papers
Drawback of Enforcing Equivariance and its Compensation via the Lens of Expressive Power
Yuzhu Chen, Tian Qin, Xinmei Tian +2
Equivariant neural networks encode the intrinsic symmetry of data as an inductive bias, which has achieved impressive performance in wide domains. However, the understanding to the…
Generalisation of RLHF under Reward Shift and Clipped KL Regularisation
Kenton Tang, Yuzhu Chen, Fengxiang He
Alignment and adaptation in large language models heavily rely on reinforcement learning from human feedback (RLHF); yet, theoretical understanding of its generalisability remains…
CoVeR: Conformal Calibration for Versatile and Reliable Autoregressive Next-Token Prediction
Yuzhu Chen, Yingjie Wang, Shunyu Liu +2
Autoregressive pre-trained models combined with decoding methods have achieved impressive performance on complex reasoning tasks. While mainstream decoding strategies such as beam…
HRP: High-Rank Preheating for Superior LoRA Initialization
Yuzhu Chen, Yingjie Wang, Shi Fu +4
This paper studies the crucial impact of initialization in Low-Rank Adaptation (LoRA). Through theoretical analysis, we demonstrate that the fine-tuned result of LoRA is highly sen…
A Theoretical Perspective: How to Prevent Model Collapse in Self-consuming Training Loops
Shi Fu, Yingjie Wang, Yuzhu Chen +2
High-quality data is essential for training large generative models, yet the vast reservoir of real data available online has become nearly depleted. Consequently, models increasin…