arXiv subjects

Yonghong Deng

Publications and source records attributed to Yonghong Deng.

4 recordsLinked to original sources

arXiv2026

How Do LLMs and VLMs Understand Viewpoint Rotation Without Vision? An Interpretability Study

Over the past year, spatial intelligence has drawn increasing attention. Many prior works study it from the perspective of visual-spatial intelligence, where models have access to visuospatial information from visual inputs. However, in the absence of visual information, whether linguistic intelligence alone is sufficient to endow models with spatial intelligence, and how models perform relevant tasks with text-only inputs still remain unexplored. Therefore, in this paper, we focus on a fundamental and critical capability in spatial intelligence from a linguistic perspective: viewpoint rotation understanding (VRU). Specifically, LLMs and VLMs are asked to infer their final viewpoint and predict the corresponding observation in an environment given textual description of viewpoint rotation and observation over multiple steps. We find that both LLMs and VLMs perform poorly on our proposed dataset while human can easily achieve 100% accuracy, indicating a substantial gap between current model capabilities and the requirements of spatial intelligence. To uncover the underlying mechanisms, we conduct a layer-wise probing analysis and head-wise causal intervention. Our findings reveal that although models encode viewpoint information in the hidden states, they appear to struggle to bind the viewpoint position with corresponding observation, resulting in a hallucination in final layers. Finally, we selectively fine-tune the key attention heads identified by causal intervention to improve VRU performance. Experimental results demonstrate that such selective fine-tuning achieves improved VRU performance while avoiding catastrophic forgetting of generic abilities. Our dataset and code will be released at https://github.com/Young-Zhen/VRU_Interpret .

cs.AI↗

arXiv2026

The Struggle Between Continuation and Refusal: A Mechanistic Analysis of the Continuation-Triggered Jailbreak in LLMs

With the rapid advancement of large language models (LLMs), the safety of LLMs has become a critical concern. Despite significant efforts in safety alignment, current LLMs remain vulnerable to jailbreaking attacks. However, the root causes of such vulnerabilities are still poorly understood, necessitating a rigorous investigation into jailbreak mechanisms across both academic and industrial communities. In this work, we focus on a continuation-triggered jailbreak phenomenon, whereby simply relocating a continuation-triggered instruction suffix can substantially increase jailbreak success rates. To uncover the intrinsic mechanisms of this phenomenon, we conduct a comprehensive mechanistic interpretability analysis at the level of attention heads. Through causal interventions and activation scaling, we show that this jailbreak behavior primarily arises from an inherent competition between the model's intrinsic continuation drive and the safety defenses acquired through alignment training. Furthermore, we perform a detailed behavioral analysis of the identified safety-critical attention heads, revealing notable differences in the behaviors of safety heads across different model architectures. Grounded in these mechanistic findings, we propose Head Competition Steering (HCS), a mechanistically grounded inference-time strategy that explicitly leverages the competition between safety heads and continuation heads to suppress harmful generation, and further distill its behavioral signal into a student model via knowledge distillation, achieving inference-time safety improvements without additional computational overhead.

cs.AI↗

arXiv2025

Spontaneous Donor Defects and Voltage-Assisted Hole Doping in Beta-Gallium Oxides under Multiple Epitaxy Conditions

Beta-phase gallium oxide (beta-Ga2O3) is prone to the spontaneous formation of donor defects but poses a formidable challenge in achieving high-quality p-type doping, mainly due to its exceptionally low valence band maximum (VBM). In this study, we utilize first-principles computations to investigate the origin of spontaneous donor defects in beta-Ga2O3 grown by three typical techniques: molecular beam epitaxy (MBE), metal organic chemical vapor deposition (MOCVD), and halide vapor phase epitaxy (HVPE). Our findings elucidate that the primary donor defects vary with the growth techniques, specifically Gai3+ for MBE, Hi+ and CGa+ for MOCVD, and (2VGa+Gai+2VO)+ and ClO+ for HVPE under unintentionally doped conditions. Employing a theoretically proposed voltage-assisted doping method, we computationally demonstrate that the dominant spontaneous donors can be significantly reduced accompanied by a noticeable increase in acceptors, leading to a stepwise reduction of Fermi level to 0.52, 0.88, and 2.10 eV above VBM for the MOCVD, HVPE, and MBE methods, and a hole concentration of 8.5*10^17, 8.7*10^11, and 2.7*10^-9 cm-3, respectively, at room temperature without the use of external dopants. By introducing Mg doping, we further reduce the Fermi level for both the MBE and HVPE experiments.

cond-mat.mtrl-sci↗

arXiv2018

Spontaneous Repairing Liquid Metal/Si Nanocomposite as a Smart Conductive-Additive-Free Anode for Lithium-ion Battery

Silicon is a promising candidate for negative electrodes due to its high theoretical specific capacity (~3579 mAh g-1) and low lithiation potential (~0.40 V vs Li). However, its practical applications in battery have been inhibited by the large volume change (~400%) induced by Li+-insertion into Si lattices. Here, we attempt to resolve this issue at a fundamental level, and report for the first time a novel liquid metal (LM)-mediated spontaneous repairing conductive-additive-free Si anode for Li-ion battery. The fluidity of LM ensures the eternal contact between Si and the conducting-network during its repeated electrochemical reactions. The as-prepared nano-composite of LM/Si leads to superior performances as characterized by high capacity utilization (2300 mAh g-1 at 500 mA g-1), long-term stability (968 mAh g-1 after 1500 charge-discharge cycles at 8 A g-1 with 81.3% retention), high rate capability (360 mAh g-1 at 20 A g-1, equivalence of 55 C, or full charge/discharge in 65 seconds), and, in particular, an extra-ordinarily high initial coulombic efficiency (95.92%), which is not only the highest reported for Si to the best of our knowledge, but also higher than the mature graphitic carbon anodes. The unique approach described in this work not only resolves the basic stress challenges faced by the promising but often problematic alloy-type materials; in broader context it also provides a universal inspiration to all electrode materials whose electric properties suffer from extreme mechanic upheavals induced by the electrochemical strains during the cell reactions.

physics.app-ph↗