arXiv subjects

Hui Liu

Publications and source records attributed to Hui Liu.

At least 289 records · Page 16Linked to original sources

arXiv2024

Towards Understanding Jailbreak Attacks in LLMs: A Representation Space Analysis

Large language models (LLMs) are susceptible to a type of attack known as jailbreaking, which misleads LLMs to output harmful contents. Although there are diverse jailbreak attack strategies, there is no unified understanding on why some methods succeed and others fail. This paper explores the behavior of harmful and harmless prompts in the LLM's representation space to investigate the intrinsic properties of successful jailbreak attacks. We hypothesize that successful attacks share some similar properties: They are effective in moving the representation of the harmful prompt towards the direction to the harmless prompts. We leverage hidden representations into the objective of existing jailbreak attacks to move the attacks along the acceptance direction, and conduct experiments to validate the above hypothesis using the proposed objective. We hope this study provides new insights into understanding how LLMs understand harmfulness information.

cs.CL↗

arXiv2024

Large Language Models for Lossless Image Compression: Next-Pixel Prediction in Language Space is All You Need

We have recently witnessed that ``Intelligence" and `` Compression" are the two sides of the same coin, where the language large model (LLM) with unprecedented intelligence is a general-purpose lossless compressor for various data modalities. This attribute particularly appeals to the lossless image compression community, given the increasing need to compress high-resolution images in the current streaming media era. Consequently, a spontaneous envision emerges: Can the compression performance of the LLM elevate lossless image compression to new heights? However, our findings indicate that the naive application of LLM-based lossless image compressors suffers from a considerable performance gap compared with existing state-of-the-art (SOTA) codecs on common benchmark datasets. In light of this, we are dedicated to fulfilling the unprecedented intelligence (compression) capacity of the LLM for lossless image compression tasks, thereby bridging the gap between theoretical and practical compression performance. Specifically, we propose P$^{2}$-LLM, a next-pixel prediction-based LLM, which integrates various elaborated insights and methodologies, \textit{e.g.,} pixel-level priors, the in-context ability of LLM, and a pixel-level semantic preservation strategy, to enhance the understanding capacity of pixel sequences for better next-pixel predictions. Extensive experiments on benchmark datasets demonstrate that P$^{2}$-LLM can beat SOTA classical and learned codecs.

cs.CV↗

arXiv2024

Towards Knowledge Checking in Retrieval-augmented Generation: A Representation Perspective

Retrieval-Augmented Generation (RAG) systems have shown promise in enhancing the performance of Large Language Models (LLMs). However, these systems face challenges in effectively integrating external knowledge with the LLM's internal knowledge, often leading to issues with misleading or unhelpful information. This work aims to provide a systematic study on knowledge checking in RAG systems. We conduct a comprehensive analysis of LLM representation behaviors and demonstrate the significance of using representations in knowledge checking. Motivated by the findings, we further develop representation-based classifiers for knowledge filtering. We show substantial improvements in RAG performance, even when dealing with noisy knowledge databases. Our study provides new insights into leveraging LLM representations for enhancing the reliability and effectiveness of RAG systems.

cs.LG↗

arXiv2024

Realization of Kagome Kondo lattice

The Kondo lattice, describing a grid of the local magnetic moments coupling to itinerant electrons, is a fertile ground of strongly correlated states in condensed matter physics. While the Kagome lattice has long been predicted to host Kondo physics with exotic magnetism and nontrivial topology, no experimental realization has been achieved. Here, we report the discovery of CsCr6Sb6, a van der Waals-like Kagome Kondo lattice featuring extremely flat, isolated bands at the Fermi level (EF) that composed entirely of Cr-3d electrons. We observe heavy fermions with the effective mass over 100 times greater than those of its vanadium counterpart. We also observe Kondo insulating behavior in an ultra-low carrier density of 1019 cm-3 and dimensionality-induced Kondo breakdown. More interestingly, the frustrated magnetism observed in the bulk give way to a hidden A-type antiferromagnetic ordering in few layers, in sharp contrast to the common sense of weakened magnetism with thinning. The realization of Kondo physics in Kagome lattice opens avenues for exploring diverse quantum criticalities in a strongly-correlated frustrated system.

cond-mat.str-el↗

arXiv2024

Constructing A Finite Tension Domain Wall in $ϕ^4_4$

We have recently claimed that the domain wall in the 3+1 dimensional $ϕ^4$ double-well model can be constructed as a squeezed, coherent state and that at one loop it has a finite tension given general, but unspecified, renormalization conditions. In the present note, we justify this claim by showing that the tadpole is finite and the infrared divergences cancel exactly. Also we carefully treat the renormalization of the normal ordering mass scale. Faddeev and Korepin have stressed that ultraviolet divergences cancel in the soliton sector if they cancel in the vacuum sector when the corresponding calculations are identical in the ultraviolet. We therefore renormalize the divergences in the vacuum sector using a Schrodinger picture prescription, which mirrors closely the analogous calculations in the domain wall sector.

hep-th↗

arXiv2024

A Finite Tension for the $ϕ^4_4$ Domain Wall

In 1+1 dimensions, it is well known that the quantum states corresponding to solitons are well described by coherent states. In his 1975 Erice lectures, Coleman observed that this construction does not extend to higher dimensions, as the coherent states have infinite energy density. He challenged the students to construct the quantum states corresponding to solitons in higher dimensions, a problem which remains unsolved today. However, even in 1+1 dimensions, the correct quantum states are actually given by deformations of coherent states. In the 3+1 dimensional $ϕ^4$ double-well model, we show that the leading deformation, which is just a squeeze, already cancels the one-loop divergence in the energy density of the domain wall soliton.

hep-th↗

arXiv2024

An Empirical Study on the Potential of LLMs in Automated Software Refactoring

Recent advances in large language models (LLMs), make it potentially feasible to automatically refactor source code with LLMs. However, it remains unclear how well LLMs perform compared to human experts in conducting refactorings automatically and accurately. To fill this gap, in this paper, we conduct an empirical study to investigate the potential of LLMs in automated software refactoring, focusing on the identification of refactoring opportunities and the recommendation of refactoring solutions. We first construct a high-quality refactoring dataset comprising 180 real-world refactorings from 20 projects, and conduct the empirical study on the dataset. With the to-be-refactored Java documents as input, ChatGPT and Gemini identified only 28 and 7 respectively out of the 180 refactoring opportunities. However, explaining the expected refactoring subcategories and narrowing the search space in the prompts substantially increased the success rate of ChatGPT from 15.6% to 86.7%. Concerning the recommendation of refactoring solutions, ChatGPT recommended 176 refactoring solutions for the 180 refactorings, and 63.6% of the recommended solutions were comparable to (even better than) those constructed by human experts. However, 13 out of the 176 solutions suggested by ChatGPT and 9 out of the 137 solutions suggested by Gemini were unsafe in that they either changed the functionality of the source code or introduced syntax errors, which indicate the risk of LLM-based refactoring. To this end, we propose a detect-and-reapply tactic, called RefactoringMirror, to avoid such unsafe refactorings. By reapplying the identified refactorings to the original code using thoroughly tested refactoring engines, we can effectively mitigate the risks associated with LLM-based automated refactoring while still leveraging LLM's intelligence to obtain valuable refactoring recommendations.

cs.SE↗

arXiv2024

HC$^3$L-Diff: Hybrid conditional latent diffusion with high frequency enhancement for CBCT-to-CT synthesis

Background: Cone-beam computed tomography (CBCT) plays a crucial role in image-guided radiotherapy, but artifacts and noise make them unsuitable for accurate dose calculation. Artificial intelligence methods have shown promise in enhancing CBCT quality to produce synthetic CT (sCT) images. However, existing methods either produce images of suboptimal quality or incur excessive time costs, failing to satisfy clinical practice standards. Methods and materials: We propose a novel hybrid conditional latent diffusion model for efficient and accurate CBCT-to-CT synthesis, named HC$^3$L-Diff. We employ the Unified Feature Encoder (UFE) to compress images into a low-dimensional latent space, thereby optimizing computational efficiency. Beyond the use of CBCT images, we propose integrating its high-frequency knowledge as a hybrid condition to guide the diffusion model in generating sCT images with preserved structural details. This high-frequency information is captured using our designed High-Frequency Extractor (HFE). During inference, we utilize denoising diffusion implicit model to facilitate rapid sampling. We construct a new in-house prostate dataset with paired CBCT and CT to validate the effectiveness of our method. Result: Extensive experimental results demonstrate that our approach outperforms state-of-the-art methods in terms of sCT quality and generation efficiency. Moreover, our medical physicist conducts the dosimetric evaluations to validate the benefit of our method in practical dose calculation, achieving a remarkable 93.8% gamma passing rate with a 2%/2mm criterion, superior to other methods. Conclusion: The proposed HC$^3$L-Diff can efficiently achieve high-quality CBCT-to-CT synthesis in only over 2 mins per patient. Its promising performance in dose calculation shows great potential for enhancing real-world adaptive radiotherapy.

eess.IV↗

arXiv2024

PrefPaint: Aligning Image Inpainting Diffusion Model with Human Preference

In this paper, we make the first attempt to align diffusion models for image inpainting with human aesthetic standards via a reinforcement learning framework, significantly improving the quality and visual appeal of inpainted images. Specifically, instead of directly measuring the divergence with paired images, we train a reward model with the dataset we construct, consisting of nearly 51,000 images annotated with human preferences. Then, we adopt a reinforcement learning process to fine-tune the distribution of a pre-trained diffusion model for image inpainting in the direction of higher reward. Moreover, we theoretically deduce the upper bound on the error of the reward model, which illustrates the potential confidence of reward estimation throughout the reinforcement alignment process, thereby facilitating accurate regularization. Extensive experiments on inpainting comparison and downstream tasks, such as image extension and 3D reconstruction, demonstrate the effectiveness of our approach, showing significant improvements in the alignment of inpainted images with human preference compared with state-of-the-art methods. This research not only advances the field of image inpainting but also provides a framework for incorporating human preference into the iterative refinement of generative models based on modeling reward accuracy, with broad implications for the design of visually driven AI applications. Our code and dataset are publicly available at https://prefpaint.github.io.

cs.CV↗

arXiv2024

Evaluating and Improving ChatGPT-Based Expansion of Abbreviations

Source code identifiers often contain abbreviations. Such abbreviations may reduce the readability of the source code, which in turn hinders the maintenance of the software applications. To this end, accurate and automated approaches to expanding abbreviations in source code are desirable and abbreviation expansion has been intensively investigated. However, to the best of our knowledge, most existing approaches are heuristics, and none of them has even employed deep learning techniques, let alone the most advanced large language models (LLMs). LLMs have demonstrated cutting-edge performance in various software engineering tasks, and thus it has the potential to expand abbreviation automatically. To this end, in this paper, we present the first empirical study on LLM-based abbreviation expansion. Our evaluation results on a public benchmark suggest that ChatGPT is substantially less accurate than the state-of-the-art approach, reducing precision and recall by 28.2\% and 27.8\%, respectively. We manually analyzed the failed cases, and discovered the root causes for the failures: 1) Lack of contexts and 2) Inability to recognize abbreviations. In response to the first cause, we investigated the effect of various contexts and found surrounding source code is the best selection. In response to the second cause, we designed an iterative approach that identifies and explicitly marks missed abbreviations in prompts. Finally, we proposed a post-condition checking to exclude incorrect expansions that violate commonsense. All such measures together make ChatGPT-based abbreviation expansion comparable to the state of the art while avoiding expensive source code parsing and deep analysis that are indispensable for state-of-the-art approaches.

cs.SE↗

arXiv2024

Exploring Social Desirability Response Bias in Large Language Models: Evidence from GPT-4 Simulations

Large language models (LLMs) are employed to simulate human-like responses in social surveys, yet it remains unclear if they develop biases like social desirability response (SDR) bias. To investigate this, GPT-4 was assigned personas from four societies, using data from the 2022 Gallup World Poll. These synthetic samples were then prompted with or without a commitment statement intended to induce SDR. The results were mixed. While the commitment statement increased SDR index scores, suggesting SDR bias, it reduced civic engagement scores, indicating an opposite trend. Additional findings revealed demographic associations with SDR scores and showed that the commitment statement had limited impact on GPT-4's predictive performance. The study underscores potential avenues for using LLMs to investigate biases in both humans and LLMs themselves.

cs.AI↗

arXiv2024

Deep Learning-based Software Engineering: Progress, Challenges, and Opportunities

Researchers have recently achieved significant advances in deep learning techniques, which in turn has substantially advanced other research disciplines, such as natural language processing, image processing, speech recognition, and software engineering. Various deep learning techniques have been successfully employed to facilitate software engineering tasks, including code generation, software refactoring, and fault localization. Many papers have also been presented in top conferences and journals, demonstrating the applications of deep learning techniques in resolving various software engineering tasks. However, although several surveys have provided overall pictures of the application of deep learning techniques in software engineering, they focus more on learning techniques, that is, what kind of deep learning techniques are employed and how deep models are trained or fine-tuned for software engineering tasks. We still lack surveys explaining the advances of subareas in software engineering driven by deep learning techniques, as well as challenges and opportunities in each subarea. To this end, in this paper, we present the first task-oriented survey on deep learning-based software engineering. It covers twelve major software engineering subareas significantly impacted by deep learning techniques. Such subareas spread out the through the whole lifecycle of software development and maintenance, including requirements engineering, software development, testing, maintenance, and developer collaboration. As we believe that deep learning may provide an opportunity to revolutionize the whole discipline of software engineering, providing one survey covering as many subareas as possible in software engineering can help future research push forward the frontier of deep learning-based software engineering more systematically.

cs.SE↗

arXiv2024

Fate of localization features in a one-dimensional non-Hermitian flat-band lattice with quasiperiodic modulations

We investigate the influence of quasiperiodic modulations on one-dimensional non-Hermitian diamond lattices with an artificial magnetic flux $θ$ that possess flat bands. Our study shows that the symmetry of these modulations and the magnetic flux $θ$ play a pivotal role in shaping the localization properties of the system. When $θ=0$, the non-Hermitian lattice exhibits a single flat band in the crystalline case, and symmetric as well as antisymmetric modulations can induce accurate mobility edges. In contrast, when $θ=π$, the clean diamond lattice manifests three dispersionless bands referred to as an "all-band-flat" (ABF) structure, irrespective of the non-Hermitian parameter. The ABF structure restricts the transition from delocalized to localized states, as all states remain localized for any finite symmetric modulation. Our numerical calculations further unveil that the ABF system subjected to antisymmetric modulations exhibits multifractal-to-localized edges. Multifractal states are predominantly concentrated in the internal region of the spectrum. Additionally, we explore the case where $θ$ lies within the range of $(0, π)$, revealing a diverse array of complex localization features. Finally, we propose a classical electrical circuit scheme to realize the non-Hermitian flat-band chain with quasiperiodic modulations.

cond-mat.dis-nn↗

arXiv2024

Unraveling the Mechanics of Learning-Based Demonstration Selection for In-Context Learning

Large Language Models (LLMs) have demonstrated impressive in-context learning (ICL) capabilities from few-shot demonstration exemplars. While recent learning-based demonstration selection methods have proven beneficial to ICL by choosing more useful exemplars, their underlying mechanisms are opaque, hindering efforts to address limitations such as high training costs and poor generalization across tasks. These methods generally assume the selection process captures similarities between the exemplar and the target instance, however, it remains unknown what kinds of similarities are captured and vital to performing ICL. To dive into this question, we analyze the working mechanisms of the learning-based demonstration selection methods and empirically identify two important factors related to similarity measurement: 1) The ability to integrate different levels of task-agnostic text similarities between the input of exemplars and test cases enhances generalization power across different tasks. 2) Incorporating task-specific labels when measuring the similarities significantly improves the performance on each specific task. We validate these two findings through extensive quantitative and qualitative analyses across ten datasets and various LLMs. Based on our findings, we introduce two effective yet simplified exemplar selection methods catering to task-agnostic and task-specific demands, eliminating the costly LLM inference overhead.

cs.LG↗

arXiv2024

Learning Fair Models without Sensitive Attributes: A Generative Approach

Most existing fair classifiers rely on sensitive attributes to achieve fairness. However, for many scenarios, we cannot obtain sensitive attributes due to privacy and legal issues. The lack of sensitive attributes challenges many existing fair classifiers. Though we lack sensitive attributes, for many applications, there usually exists features or information of various formats that are relevant to sensitive attributes. For example, purchase history of a person can reflect his or her race, which would help for learning fair classifiers on race. However, the work on exploring relevant features for learning fair models without sensitive attributes is rather limited. Therefore, in this paper, we study a novel problem of learning fair models without sensitive attributes by exploring relevant features. We propose a probabilistic generative framework to effectively estimate the sensitive attribute from the training data with relevant features in various formats and utilize the estimated sensitive attribute information to learn fair models. Experimental results on real-world datasets show the effectiveness of our framework in terms of both accuracy and fairness.

cs.LG↗

arXiv2024

TCDformer-based Momentum Transfer Model for Long-term Sports Prediction

Accurate sports prediction is a crucial skill for professional coaches, which can assist in developing effective training strategies and scientific competition tactics. Traditional methods often use complex mathematical statistical techniques to boost predictability, but this often is limited by dataset scale and has difficulty handling long-term predictions with variable distributions, notably underperforming when predicting point-set-game multi-level matches. To deal with this challenge, this paper proposes TM2, a TCDformer-based Momentum Transfer Model for long-term sports prediction, which encompasses a momentum encoding module and a prediction module based on momentum transfer. TM2 initially encodes momentum in large-scale unstructured time series using the local linear scaling approximation (LLSA) module. Then it decomposes the reconstructed time series with momentum transfer into trend and seasonal components. The final prediction results are derived from the additive combination of a multilayer perceptron (MLP) for predicting trend components and wavelet attention mechanisms for seasonal components. Comprehensive experimental results show that on the 2023 Wimbledon men's tournament datasets, TM2 significantly surpasses existing sports prediction models in terms of performance, reducing MSE by 61.64% and MAE by 63.64%.

cs.LG↗

arXiv2024

Quantum walks of correlated photons in non-Hermitian photonic lattices

Entanglement entropy characterizes the correlation of multi-particles and unveils the crucial features of open quantum systems. However, the experimental realization of exploring entanglement in non-Hermitian systems remains a challenge. In parallel, quantum walks have offered the possibility of studying the underlying mechanisms of non-Hermitian physics, which includes exceptional points, the non-Hermitian skin effect, and non-Bloch phase transitions. Unfortunately, these studies have only involved and prevailingly focused on the behavior of a single particle. Here, we propose and experimentally realize quantum walks of two indistinguishable photons in engineered non-Hermitian photonic lattices. We have successfully observed the unidirectional behavior of quantum walks in the bulk far from the edges induced by the skin effect. Moreover, we experimentally reveal the suppression of entanglement that is caused by the skin effect in non-Hermitian systems. Our study may facilitate a deep understanding of entanglement in open quantum many-body systems that are far from thermal equilibrium.

quant-ph↗

arXiv2024

Interpretable Multimodal Misinformation Detection with Logic Reasoning

Multimodal misinformation on online social platforms is becoming a critical concern due to increasing credibility and easier dissemination brought by multimedia content, compared to traditional text-only information. While existing multimodal detection approaches have achieved high performance, the lack of interpretability hinders these systems' reliability and practical deployment. Inspired by NeuralSymbolic AI which combines the learning ability of neural networks with the explainability of symbolic learning, we propose a novel logic-based neural model for multimodal misinformation detection which integrates interpretable logic clauses to express the reasoning process of the target task. To make learning effective, we parameterize symbolic logical elements using neural representations, which facilitate the automatic generation and evaluation of meaningful logic clauses. Additionally, to make our framework generalizable across diverse misinformation sources, we introduce five meta-predicates that can be instantiated with different correlations. Results on three public datasets (Twitter, Weibo, and Sarcasm) demonstrate the feasibility and versatility of our model.

cs.MM↗