Author name cluster

Yihong Liu

Possible papers associated with this exact author name in Arrow. This page groups case-insensitive exact name matches and is not a full identity disambiguation profile.

3 papers

2 author rows

ICRA Conference 2025 Conference Paper

Learning a High-Quality Robotic Wiping Policy Using Systematic Reward Analysis and Visual-Language Model Based Curriculum

Yihong Liu
Dongyeop Kang
Sehoon Ha

Autonomous robotic wiping is an important task in various industries, ranging from industrial manufacturing to sanitization in healthcare. Deep reinforcement learning (Deep RL) has emerged as a promising algorithm, however, it often suffers from a high demand for repetitive reward engineering. Instead of relying on manual tuning, we first analyze the convergence of quality-critical robotic wiping, which requires both high-quality wiping and fast task completion, to show the poor convergence of the problem and propose a new bounded reward formulation to make the problem feasible. Then, we further improve the learning process by proposing a novel visual-language model (VLM) based curriculum, which actively monitors the progress and suggests hyperparameter tuning. We demonstrate that the combined method can find a desirable wiping policy on surfaces with various curvatures, frictions, and waypoints, which cannot be learned with the baseline formulation. The demo of this project can be found at: https://sites.google.com/view/highqualitywiping

Details

NeurIPS Conference 2025 Conference Paper

Refusal Direction is Universal Across Safety-Aligned Languages

Xinpeng Wang
Mingyang Wang
Yihong Liu
Hinrich Schuetze
Barbara Plank

Refusal mechanisms in large language models (LLMs) are essential for ensuring safety. Recent research has revealed that refusal behavior can be mediated by a single direction in activation space, enabling targeted interventions to bypass refusals. While this is primarily demonstrated in an English-centric context, appropriate refusal behavior is important for any language, but poorly understood. In this paper, we investigate the refusal behavior in LLMs across 14 languages using \textit{PolyRefuse}, a multilingual safety dataset created by translating malicious and benign English prompts into these languages. We uncover the surprising cross-lingual universality of the refusal direction: a vector extracted from English can bypass refusals in other languages with near-perfect effectiveness, without any additional fine-tuning. Even more remarkably, refusal directions derived from any safety-aligned language transfer seamlessly to others. We attribute this transferability to the parallelism of refusal vectors across languages in the embedding space and identify the underlying mechanism behind cross-lingual jailbreaks. These findings provide actionable insights for building more robust multilingual safety defenses and pave the way for a deeper mechanistic understanding of cross-lingual vulnerabilities in LLMs.

PDF Details

ECAI Conference 2024 Conference Paper

Segmentation-Driven Image Enhancement Based on Deep Reinforcement Learning

Yihong Liu
Zishang Chen
Yukang Cui 0002
Piji Li

The rise of large models, often referred to as foundational models, has led to considerable progress in the field of artificial intelligence research. Our empirical findings indicate that the large models might struggle or deliver poor performance when it comes to specific surface segmentation challenges, including the identification and segmentation of defects on strip steel surfaces (S3D) and the detection of imperfections on magnetic tile surfaces. To apply the large model to defects segmentation, rather than fine-tuning the large model, we propose Segmentation-Driven Image Enhancement (SDIE), using several classic filters to enhance the input images. In this case, the weights of the filters in multiple layers are controlled by reinforcement learning. Then, we test our method on two S3D datasets with different few-shot settings. Our method accomplishes the task brilliantly compared with other methods for S3D such as CPANet. We believe that our work not only opens up opportunities for downstream tasks such as segmenting industrial defects using large models, but may also have potential applications in various fields in the future, including medical image processing, remote sensing image analysis, agriculture and more.

Details