arXiv subjects

Chongyang Liu

Publications and source records attributed to Chongyang Liu.

8 recordsLinked to original sources

arXiv2026

ROBOSHACKLES: A Safety Dataset for Human-Injury Prevention in Embodied Foundation Models

Embodied Foundation Models (EFMs) integrate multimodal understanding, future-state reasoning, and executable robot actions. Yet their safety alignment for human-injury prevention remains underexplored, primarily because real-world data of robots harming humans or creating hazardous household situations cannot be safely or ethically collected. To address this challenge, we propose a safety-critical data construction pipeline for human-injury prevention in EFMs.Starting from real DROID observations, our construction pipeline proceeds through scene understanding, hazard-aware image editing, temporal prompt generation, and single-pass rollout synthesis. The temporal prompts specify the expected scene evolution, while Wan2.7 synthesizes realistic robotic rollouts from the edited hazardous states in a single pass. Using this pipeline, we construct ROBOSHACKLES, a 10,000-clip robotic video dataset derived from real DROID observations, spanning two direct-harm and four indirect-harm categories. To ensure dataset quality, we assess task completion and visual quality with automatic metrics, and evaluate six representative EFMs under a refusal-based safety criterion. Results show that all evaluated models produce unsafe actions in the tested safety-critical scenarios, yielding a 100% unsafe action generation rate. ROBOSHACKLES serves as a scalable benchmark and training resource for refusal learning and hazard anticipation before robot action execution.The dataset is publicly available at https://huggingface.co/datasets/YZW00/RoboShackles.

cs.RO↗

arXiv2025

Improving Code Understanding in Large Language Models through Concept-Aware Consistency Learning

Large language models (LLMs) have recently shown impressive results on diverse code-related tasks, benefiting from large-scale training and instruction tuning. However, studies reveal that their grasp of fundamental programming concepts, such as data flow and control flow, remains shallow, leading to fragile performance when code requires deeper reasoning. This limitation restricts the practical adoption of LLMs in real-world software development. To address this issue, this work introduces a counterfactual code augmentation framework combined with concept-aware tuning, designed to guide LLMs toward stronger conceptual understanding. Comprehensive evaluation across multiple models and benchmarks demonstrates the effectiveness of the proposed approach.

cs.SE↗

arXiv2024

Self and Cross-Model Distillation for LLMs: Effective Methods for Refusal Pattern Alignment

Large Language Models (LLMs) like OpenAI's GPT series, Anthropic's Claude, and Meta's LLaMa have shown remarkable capabilities in text generation. However, their susceptibility to toxic prompts presents significant security challenges. This paper investigates alignment techniques, including Supervised Fine-Tuning (SFT) and Reinforcement Learning from Human Feedback (RLHF), to mitigate these risks. We conduct an empirical study on refusal patterns across nine LLMs, revealing that models with uniform refusal patterns, such as Claude3, exhibit higher security. Based on these findings, we propose self-distilling and cross-model distilling methods to enhance LLM security. Our results show that these methods significantly improve refusal rates and reduce unsafe content, with cross-model distilling achieving refusal rates close to Claude3's 94.51%. These findings underscore the potential of distillation-based alignment in securing LLMs against toxic prompts.

cs.CR↗

arXiv2024

Scalable single-microring hybrid III-V/Si lasers for emerging narrow-linewidth applications

Silicon photonics, compatible with large-scale silicon manufacturing, is a disruptive photonic platform that has indicated significant implications in industry and research areas (e.g., quantum, neuromorphic computing, LiDAR). Cutting-edge applications such as high-capacity coherent optical communication and heterodyne LiDAR have escalated the demand for integrated narrow-linewidth laser sources. To that effect, this work seeks to address this requirement through the development of a high-performance hybrid III-V/silicon laser. The developed integrated laser, utilizes a single microring resonator (MRR), demonstrating single-mode operation with a side mode suppression ratio (SMSR) exceeding 40 dB, with laser output power as high as 16.4 mW. Moving away from current hybrid/heterogeneous laser architectures that necessitate multiple complex control, the developed laser architecture requires only two control parameters. Importantly, this serves to streamline industrial adoption by reducing the complexity involved in characterizing these lasers, at-scale. Through the succinct structure and control framework, a narrow laser linewidth of 2.79 kHz and low relative intensity noise (RIN) of -135 dB/Hz are achieved. Furthermore, optical data transmission at 12.5 Gb/s is demonstrated where a signal-to-noise ratio (SNR) of 10 dB is measured.

physics.optics↗

arXiv2024

A Cross-Language Investigation into Jailbreak Attacks in Large Language Models

Large Language Models (LLMs) have become increasingly popular for their advanced text generation capabilities across various domains. However, like any software, they face security challenges, including the risk of 'jailbreak' attacks that manipulate LLMs to produce prohibited content. A particularly underexplored area is the Multilingual Jailbreak attack, where malicious questions are translated into various languages to evade safety filters. Currently, there is a lack of comprehensive empirical studies addressing this specific threat. To address this research gap, we conducted an extensive empirical study on Multilingual Jailbreak attacks. We developed a novel semantic-preserving algorithm to create a multilingual jailbreak dataset and conducted an exhaustive evaluation on both widely-used open-source and commercial LLMs, including GPT-4 and LLaMa. Additionally, we performed interpretability analysis to uncover patterns in Multilingual Jailbreak attacks and implemented a fine-tuning mitigation method. Our findings reveal that our mitigation strategy significantly enhances model defense, reducing the attack success rate by 96.2%. This study provides valuable insights into understanding and mitigating Multilingual Jailbreak attacks.

cs.CR↗

arXiv2023

Crown ether decorated silicon photonics for safeguarding against lead poisoning

Lead (Pb2+) toxification in society is one of the most concerning public health crisis that remains unaddressed. The exposure to Pb2+ poisoning leads to a multitude of enduring health issues, even at the part-per-billion scale (ppb). Yet, public action dwarfs its impact. Pb2+ poisoning is estimated to account for 1 million deaths per year globally, which is in addition to its chronic impact on children. With their ring-shaped cavities, crown ethers are uniquely capable of selectively binding to specific ions. In this work, for the first time, the synergistic integration of highly-scalable silicon photonics, with crown ether amine conjugation via Fischer esterification in an environmentally-friendly fashion is demonstrated. This realises a photonic platform that enables the in-situ, highly-selective and quantitative detection of various ions. The development dispels the existing notion that Fischer esterification is restricted to organic compounds, laying the ground for subsequent amine conjugation for various crown ethers. In this work, the platform is engineered for Pb2+ detection, demonstrating a large dynamic detection range of 1 - 262000 ppb with high selectivity against a wide range of relevant ions. These results indicate the potential for the pervasive implementation of the technology to safeguard against ubiquitous lead poisoning in our society.

physics.optics↗

arXiv2022

Linear Optical Modulators for Prospective Communications at the 2 μm Waveband

The 2 μm waveband is an area that could have significant technological consequences, with applications ranging from spectroscopy, LIDAR and free-space communications. The development of the thulium-doped fiber amplifier, hollow-core photonic bandgap fiber and 2 μm GaSb-based diode lasers has highlighted the ability of the waveband in alleviating the fiber capacity crisis in the incumbent communication infrastructure. The above has initiated vibrant development in the silicon photonic-space at 2 μm, where the area is capable of enabling highly-integrated photonic circuits, and potentially at low-cost and high-volumes. However, as of now, modulator linearity at 2 μm has not been addressed. The metric, as characterized by spurious free dynamic range is imperative for numerous applications such as RF photonic links for 5G and digital analog transmission in coherent communications. The development of linear optical modulators will be crucial in bringing these applications to the 2 μm. In view of that, this work is the first to address modulator linearity at the 2 μm, where the ring-assisted Mach-Zehnder modulator is developed, indicating spurious free dynamic range as high as 95 dB.Hz^2/3. It is found that that modulator spurious free dynamic range has a strong dependence on modulator bias voltage where it impacts the linearity of the transfer function in which the input RF signal is applied upon. The demonstrated modulator indicates favorable performance within silicon photonic modulators developed at 2 μm with bandwidth exceeding 17.5 GHz and modulation efficiency ranging from 0.70 to 1.25 V.cm.

physics.app-ph↗

arXiv2021

RelationRS: Relationship Representation Network for Object Detection in Aerial Images

Object detection is a basic and important task in the field of aerial image processing and has gained much attention in computer vision. However, previous aerial image object detection approaches have insufficient use of scene semantic information between different regions of large-scale aerial images. In addition, complex background and scale changes make it difficult to improve detection accuracy. To address these issues, we propose a relationship representation network for object detection in aerial images (RelationRS): 1) Firstly, multi-scale features are fused and enhanced by a dual relationship module (DRM) with conditional convolution. The dual relationship module learns the potential relationship between features of different scales and learns the relationship between different scenes from different patches in a same iteration. In addition, the dual relationship module dynamically generates parameters to guide the fusion of multi-scale features. 2) Secondly, The bridging visual representations module (BVR) is introduced into the field of aerial images to improve the object detection effect in images with complex backgrounds. Experiments with a publicly available object detection dataset for aerial images demonstrate that the proposed RelationRS achieves a state-of-the-art detection performance.

cs.CV↗