arXiv subjects

Zhuo Li

Publications and source records attributed to Zhuo Li.

At least 109 records · Page 6Linked to original sources

arXiv2025

Atoxia: Red-teaming Large Language Models with Target Toxic Answers

Despite the substantial advancements in artificial intelligence, large language models (LLMs) remain being challenged by generation safety. With adversarial jailbreaking prompts, one can effortlessly induce LLMs to output harmful content, causing unexpected negative social impacts. This vulnerability highlights the necessity for robust LLM red-teaming strategies to identify and mitigate such risks before large-scale application. To detect specific types of risks, we propose a novel red-teaming method that $\textbf{A}$ttacks LLMs with $\textbf{T}$arget $\textbf{Toxi}$c $\textbf{A}$nswers ($\textbf{Atoxia}$). Given a particular harmful answer, Atoxia generates a corresponding user query and a misleading answer opening to examine the internal defects of a given LLM. The proposed attacker is trained within a reinforcement learning scheme with the LLM outputting probability of the target answer as the reward. We verify the effectiveness of our method on various red-teaming benchmarks, such as AdvBench and HH-Harmless. The empirical results demonstrate that Atoxia can successfully detect safety risks in not only open-source models but also state-of-the-art black-box models such as GPT-4o.

cs.CL↗

arXiv2025

Topologically protected edge states in time photonic crystals with chiral symmetry

Time photonic crystals are media in which their electromagnetic parameters are modulated periodically in time, showing promising applications in non-resonant lasers and particle accelerators, among others. Traditionally utilized to study space photonic crystals, topological band theory has also been translated recently to analyze time photonic crystals with time inversion symmetry, enabling the construction of the temporal version of topological edge states. However, temporal disorder can readily break time inversion symmetry in practice, hence likely destroying the edge states associated with this type of time photonic crystals. To overcome this limitation, here we propose a new class of time photonic crystals presenting chiral symmetry instead, whose edge states exhibit superior robustness over the time-reversal-symmetry-protected counterparts. Our time photonic crystal is equivalent to a temporal version of the Su-Schrieffer-Heeger model, and the chiral symmetry of this type of time photonic crystals quantizes the winding number defined in the Bloch frequency band. Remarkably, random temporal disorders do not impact the eigenfrequencies of these chiral-symmetry-protected edge states, while instead enhancing their temporal localizations. Our findings thus provide a promising paradigm to control field amplification with exceptional robustness as well as being a feasible platform to investigate various topological phases in time-varying media.

physics.optics↗

arXiv2024

Human-Humanoid Robots Cross-Embodiment Behavior-Skill Transfer Using Decomposed Adversarial Learning from Demonstration

Humanoid robots are envisioned as embodied intelligent agents capable of performing a wide range of human-level loco-manipulation tasks, particularly in scenarios requiring strenuous and repetitive labor. However, learning these skills is challenging due to the high degrees of freedom of humanoid robots, and collecting sufficient training data for humanoid is a laborious process. Given the rapid introduction of new humanoid platforms, a cross-embodiment framework that allows generalizable skill transfer is becoming increasingly critical. To address this, we propose a transferable framework that reduces the data bottleneck by using a unified digital human model as a common prototype and bypassing the need for re-training on every new robot platform. The model learns behavior primitives from human demonstrations through adversarial imitation, and the complex robot structures are decomposed into functional components, each trained independently and dynamically coordinated. Task generalization is achieved through a human-object interaction graph, and skills are transferred to different robots via embodiment-specific kinematic motion retargeting and dynamic fine-tuning. Our framework is validated on five humanoid robots with diverse configurations, demonstrating stable loco-manipulation and highlighting its effectiveness in reducing data requirements and increasing the efficiency of skill transfer across platforms.

cs.RO↗

arXiv2024

Targeting the Core: A Simple and Effective Method to Attack RAG-based Agents via Direct LLM Manipulation

AI agents, powered by large language models (LLMs), have transformed human-computer interactions by enabling seamless, natural, and context-aware communication. While these advancements offer immense utility, they also inherit and amplify inherent safety risks such as bias, fairness, hallucinations, privacy breaches, and a lack of transparency. This paper investigates a critical vulnerability: adversarial attacks targeting the LLM core within AI agents. Specifically, we test the hypothesis that a deceptively simple adversarial prefix, such as \textit{Ignore the document}, can compel LLMs to produce dangerous or unintended outputs by bypassing their contextual safeguards. Through experimentation, we demonstrate a high attack success rate (ASR), revealing the fragility of existing LLM defenses. These findings emphasize the urgent need for robust, multi-layered security measures tailored to mitigate vulnerabilities at the LLM level and within broader agent-based architectures.

cs.AI↗

arXiv2024

M$^3$GPT: An Advanced Multimodal, Multitask Framework for Motion Comprehension and Generation

This paper presents M$^3$GPT, an advanced $\textbf{M}$ultimodal, $\textbf{M}$ultitask framework for $\textbf{M}$otion comprehension and generation. M$^3$GPT operates on three fundamental principles. The first focuses on creating a unified representation space for various motion-relevant modalities. We employ discrete vector quantization for multimodal conditional signals, such as text, music and motion/dance, enabling seamless integration into a large language model (LLM) with a single vocabulary. The second involves modeling motion generation directly in the raw motion space. This strategy circumvents the information loss associated with a discrete tokenizer, resulting in more detailed and comprehensive motion generation. Third, M$^3$GPT learns to model the connections and synergies among various motion-relevant tasks. Text, the most familiar and well-understood modality for LLMs, is utilized as a bridge to establish connections between different motion tasks, facilitating mutual reinforcement. To our knowledge, M$^3$GPT is the first model capable of comprehending and generating motions based on multiple signals. Extensive experiments highlight M$^3$GPT's superior performance across various motion-relevant tasks and its powerful zero-shot generalization capabilities for extremely challenging tasks. Project page: \url{https://github.com/luomingshuang/M3GPT}.

cs.CV↗

arXiv2024

Generalized coherent wave control at dynamic interfaces

Coherent wave control is of key importance across a broad range of fields such as electromagnetics, photonics, and acoustics. It enables us to amplify or suppress the outgoing waves via engineering amplitudes and phases of multiple incidences. However, within a purely spatially (temporally) engineered medium, coherent wave control requires the frequency of the associated incidences to be identical (opposite). In this work, we break this conventional constraint by generalizing coherent wave control into a spatiotemporally engineered medium, i.e., the system featuring a dynamic interface. Owing to the broken translational symmetry in space and time, both the subluminal and superluminal interfaces allow interference between scattered waves regardless of their different frequencies and wavevectors. Hence, one can flexibly eliminate the backward- or forward-propagating waves scattered from the dynamic interfaces by controlling the incident amplitudes and phases. Our work not only presents a generalized way for reshaping arbitrary waveforms but also provides a promising paradigm to generate ultrafast pulses using low-frequency signals. We have also implemented suppression of forward-propagating waves in microstrip transmission lines with fast photodiode switches.

physics.optics↗

arXiv2024

Precision Knowledge Editing: Enhancing Safety in Large Language Models

Large language models (LLMs) have demonstrated remarkable capabilities, but they also pose risks related to the generation of toxic or harmful content. This work introduces Precision Knowledge Editing (PKE), an advanced technique that builds upon existing knowledge editing methods to more effectively identify and modify toxic parameter regions within LLMs. By leveraging neuron weight tracking and activation pathway tracing, PKE achieves finer granularity in toxic content management compared to previous methods like Detoxifying Instance Neuron Modification (DINM). Our experiments demonstrate that PKE significantly reduces the attack success rate (ASR) across various models, including Llama2-7b and Llama-3-8b-instruct, while maintaining overall model performance. Additionally, we also compared the performance of some closed-source models (gpt-4-0613 and Claude 3 Sonnet) in our experiments, and found that models adjusted using our method far outperformed the closed-source models in terms of safety. This research contributes to the ongoing efforts to make LLMs safer and more reliable for real-world applications.

cs.CL↗

arXiv2024

Implication of Jet Physics from MeV Line Emission of GRB 221009A

Ultrarelativistic jets are believed to play an important role in producing prompt emission and afterglow of gamma-ray bursts (GRBs), but the nature of the jet is poorly known owing to the lack of decisive features observed in the prompt emission. The discovery of an emission line evolving from about 37 to 6 MeV in the brightest-of-all-time GRB 221009A provides an unprecedented opportunity to probe GRB jet physics. The time evolution of the central energy of the line with power-law index $-1$ is naturally explained by the high-latitude curvature effect. Under the assumption that the line emission is generated in the prompt emission by $e^\pm$ pair production, cooling, and annihilation in the jet, we can strictly constrain jet physics with observed line emission properties. We find that the radius of the emission region is $r\gtrsim10^{16}$ cm. The narrow line width of $\sim10\%$ requires that the line emission occurs within $\sim10\%$ of the dynamical time, which further implies short timescales of pair cooling to the nonrelativistic state and pair annihilation, as well as a slightly clumpy emission region. If the jet's Lorentz factor is $Γ\gtrsim400$, the fast cooling requirement needs an energy density of magnetic field in the jet much larger than that of prompt gamma rays, i.e., a magnetically dominated jet. The temporal behavior of line flux suggests some angle dependence of line emission. We also discuss the difficulties of other scenarios for the observed emission line.

astro-ph.HE↗

arXiv2024

CodeScore: Evaluating Code Generation by Learning Code Execution

A proper code evaluation metric (CEM) profoundly impacts the evolution of code generation, which is an important research field in NLP and software engineering. Prevailing match-based CEMs (e.g., BLEU, Accuracy, and CodeBLEU) suffer from two significant drawbacks. 1. They primarily measure the surface differences between codes without considering their functional equivalence. However, functional equivalence is pivotal in evaluating the effectiveness of code generation, as different codes can perform identical operations. 2. They are predominantly designed for the Ref-only input format. However, code evaluation necessitates versatility in input formats. Aside from Ref-only, there are NL-only and Ref\&NL formats, which existing match-based CEMs cannot effectively accommodate. In this paper, we propose CodeScore, a large language model (LLM)-based CEM, which estimates the functional correctness of generated code on three input types. To acquire CodeScore, we present UniCE, a unified code generation learning framework, for LLMs to learn code execution (i.e., learning PassRatio and Executability of generated code) with unified input. Extensive experimental results on multiple code evaluation datasets demonstrate that CodeScore absolutely improves up to 58.87% correlation with functional correctness compared to other CEMs, achieves state-of-the-art performance, and effectively handles three input formats.

cs.SE↗

arXiv2024

Self-Instructed Derived Prompt Generation Meets In-Context Learning: Unlocking New Potential of Black-Box LLMs

Large language models (LLMs) have shown success in generating high-quality responses. In order to achieve better alignment with LLMs with human preference, various works are proposed based on specific optimization process, which, however, is not suitable to Black-Box LLMs like GPT-4, due to inaccessible parameters. In Black-Box LLMs case, their performance is highly dependent on the quality of the provided prompts. Existing methods to enhance response quality often involve a prompt refinement model, yet these approaches potentially suffer from semantic inconsistencies between the refined and original prompts, and typically overlook the relationship between them. To address these challenges, we introduce a self-instructed in-context learning framework that empowers LLMs to deliver more effective responses by generating reliable derived prompts to construct informative contextual environments. Our approach incorporates a self-instructed reinforcement learning mechanism, enabling direct interaction with the response model during derived prompt generation for better alignment. We then formulate querying as an in-context learning task, using responses from LLMs combined with the derived prompts to establish a contextual demonstration for the original prompt. This strategy ensures alignment with the original query, reduces discrepancies from refined prompts, and maximizes the LLMs' in-context learning capability. Extensive experiments demonstrate that the proposed method not only generates more reliable derived prompts but also significantly enhances LLMs' ability to deliver more effective responses, including Black-Box models such as GPT-4.

cs.CL↗

arXiv2024

Relation between the keV-MeV and TeV emission of GRB 221009A and its implications

Gamma-ray bursts (GRBs) are believed to launch relativistic jets, which generate prompt emission by internal processes, and produce long-lasting afterglows by driving external shocks into surrounding medium. However, how the jet powers the external shock is poorly known. The unprecedented observations of the keV-MeV emission with GECAM and the TeV emission with LHAASO of the brightest-of-all-time GRB 221009A offer a great opportunity to study the prompt-to-afterglow transition and the impact of jet on the early dynamics of external shock. In this letter, we find that the cumulative light curve of keV-MeV emission could well fit the rising stage of the TeV light curve of GRB 221009A, with a time delay, $4.45^{+0.26}_{-0.26}$\,s, of TeV emission. Moreover, both the rapid increase in the initial stage and the excess from about \T+260\,s to 270\,s in the TeV light curve are tracking the light-curve bumps in the prompt keV-MeV emission. The close relation between the keV-MeV and TeV emission reveals the continuous energy-injection into the external shock. Assuming an energy-injection rate exactly following the keV-MeV flux of GRB 221009A, including the very early precursor, we build a continuous energy-injection model where the jet Lorentz factor is derived from the TeV time delay, and the TeV data is well fitted, with the TeV excesses interpreted by inverse Compton (IC) scatterings of the inner-coming prompt emission by the energetic electrons in external shock.

astro-ph.HE↗

arXiv2024

Approach for composition measurement of cosmic rays using the muon-to-electron ratio observed by LHAASO-KM2A

Composition measurement of cosmic rays (CRs) around the knee of the CR energy spectrum is crucial for studying the processes of particle acceleration and propagation of Galactic CRs. The Square Kilometer Array (KM2A) of Large High Altitude Air Shower Observatory (LHAASO) can provide precise measurement of the muonic and electromagnetic (em.) components in CR-induced extensive air showers, and hence a good chance to disentangle the CR composition. Here we propose an approach of decomposing CR compositions with the number ratio between muons and em. particles ($N_μ$/$N_{\rm e}$) observed by LHAASO-KM2A: we reconstruct the energy spectra of individual CR compositions by fitting $N_μ$/$N_{\rm e}$ distributions in each reconstructed energy bin using the template shapes of $N_μ$/$N_{\rm e}$ distributions of individual CR compositions based on Monte Carlo (MC) simulation. We evaluate the performance of this approach with MC tests where mock data of LHAASO-KM2A are generated by MC simulation. We show that the input composition model can be well recovered in this approach, independent of the CR composition model adopted in the MC simulation for the template distributions. The uncertainties of the reconstructed spectra at < 20 PeV, mainly limited by simulation statistics, are $\le$ 7% for proton, He, and Fe groups, and $\le$ 8% and $\le$ 16% for CNO and MgAlSi groups, respectively.

astro-ph.HE↗

arXiv2024

DIM: Dynamic Integration of Multimodal Entity Linking with Large Language Model

Our study delves into Multimodal Entity Linking, aligning the mention in multimodal information with entities in knowledge base. Existing methods are still facing challenges like ambiguous entity representations and limited image information utilization. Thus, we propose dynamic entity extraction using ChatGPT, which dynamically extracts entities and enhances datasets. We also propose a method: Dynamically Integrate Multimodal information with knowledge base (DIM), employing the capability of the Large Language Model (LLM) for visual understanding. The LLM, such as BLIP-2, extracts information relevant to entities in the image, which can facilitate improved extraction of entity features and linking them with the dynamic entity representations provided by ChatGPT. The experiments demonstrate that our proposed DIM method outperforms the majority of existing methods on the three original datasets, and achieves state-of-the-art (SOTA) on the dynamically enhanced datasets (Wiki+, Rich+, Diverse+). For reproducibility, our code and collected datasets are released on \url{https://github.com/season1blue/DIM}.

cs.CL↗

arXiv2024

Calibration of Time-Series Forecasting: Detecting and Adapting Context-Driven Distribution Shift

Recent years have witnessed the success of introducing deep learning models to time series forecasting. From a data generation perspective, we illustrate that existing models are susceptible to distribution shifts driven by temporal contexts, whether observed or unobserved. Such context-driven distribution shift (CDS) introduces biases in predictions within specific contexts and poses challenges for conventional training paradigms. In this paper, we introduce a universal calibration methodology for the detection and adaptation of CDS with a trained model. To this end, we propose a novel CDS detector, termed the "residual-based CDS detector" or "Reconditionor", which quantifies the model's vulnerability to CDS by evaluating the mutual information between prediction residuals and their corresponding contexts. A high Reconditionor score indicates a severe susceptibility, thereby necessitating model adaptation. In this circumstance, we put forth a straightforward yet potent adapter framework for model calibration, termed the "sample-level contextualized adapter" or "SOLID". This framework involves the curation of a contextually similar dataset to the provided test sample and the subsequent fine-tuning of the model's prediction layer with a limited number of steps. Our theoretical analysis demonstrates that this adaptation strategy can achieve an optimal bias-variance trade-off. Notably, our proposed Reconditionor and SOLID are model-agnostic and readily adaptable to a wide range of models. Extensive experiments show that SOLID consistently enhances the performance of current forecasting models on real-world datasets, especially on cases with substantial CDS detected by the proposed Reconditionor, thus validating the effectiveness of the calibration approach.

cs.LG↗

arXiv2024

PEMT: Multi-Task Correlation Guided Mixture-of-Experts Enables Parameter-Efficient Transfer Learning

Parameter-efficient fine-tuning (PEFT) has emerged as an effective method for adapting pre-trained language models to various tasks efficiently. Recently, there has been a growing interest in transferring knowledge from one or multiple tasks to the downstream target task to achieve performance improvements. However, current approaches typically either train adapters on individual tasks or distill shared knowledge from source tasks, failing to fully exploit task-specific knowledge and the correlation between source and target tasks. To overcome these limitations, we propose PEMT, a novel parameter-efficient fine-tuning framework based on multi-task transfer learning. PEMT extends the mixture-of-experts (MoE) framework to capture the transferable knowledge as a weighted combination of adapters trained on source tasks. These weights are determined by a gated unit, measuring the correlation between the target and each source task using task description prompt vectors. To fully exploit the task-specific knowledge, we also propose the Task Sparsity Loss to improve the sparsity of the gated unit. We conduct experiments on a broad range of tasks over 17 datasets. The experimental results demonstrate our PEMT yields stable improvements over full fine-tuning, and state-of-the-art PEFT and knowledge transferring methods on various tasks. The results highlight the effectiveness of our method which is capable of sufficiently exploiting the knowledge and correlation features across multiple tasks.

cs.CL↗

arXiv2024

Identifiability Matters: Revealing the Hidden Recoverable Condition in Unbiased Learning to Rank

Unbiased Learning to Rank (ULTR) aims to train unbiased ranking models from biased click logs, by explicitly modeling a generation process for user behavior and fitting click data based on examination hypothesis. Previous research found empirically that the true latent relevance is mostly recoverable through click fitting. However, we demonstrate that this is not always achievable, resulting in a significant reduction in ranking performance. This research investigates the conditions under which relevance can be recovered from click data in the first principle. We initially characterize a ranking model as identifiable if it can recover the true relevance up to a scaling transformation, a criterion sufficient for the pairwise ranking objective. Subsequently, we investigate an equivalent condition for identifiability, articulated as a graph connectivity test problem: the recovery of relevance is feasible if and only if the identifiability graph (IG), derived from the underlying structure of the dataset, is connected. The presence of a disconnected IG may lead to degenerate cases and suboptimal ranking performance. To tackle this challenge, we introduce two methods, namely node intervention and node merging, designed to modify the dataset and restore the connectivity of the IG. Empirical results derived from a simulated dataset and two real-world LTR benchmark datasets not only validate our proposed theory but also demonstrate the effectiveness of our methods in alleviating data bias when the relevance model is unidentifiable.

cs.IR↗

arXiv2024

Electrically Programmable Pixelated Graphene-Integrated Plasmonic Metasurfaces for Coherent Mid-Infrared Emission

Active metasurfaces have recently emerged as compact, lightweight, and efficient platforms for dynamic control of electromagnetic fields and optical responses. However, the complexities associated with their post-fabrication tunability significantly hinder their widespread applications, especially for the mid-infrared range due to material scarcity and design intricacy. Here, we experimentally demonstrate highly dynamic, pixelated modulations of coherent mid-infrared emission based on an electrically programmable plasmonic metasurface integrated with graphene field effect transistors (Gr-FETs). The ultrabroad infrared transparency of graphene allows for free-form control over plasmonic meta-atoms, thus achieving coherent mid-infrared states across a broad range of wavelengths and polarizations. The spatial temperature modulation generated by Gr-FETs is effectively synergized with the emissivity control by the localized surface plasmon polaritons from gold nanoantennas. This integrated temperature-emissivity modulation of metasurfaces is systematically extended to form a pixelated 2D array, envisioning new approaches toward scalable 2D electrical wiring for densely packed, independently controlled pixels.

physics.optics↗

arXiv2024

Extracting Clean and Balanced Subset for Noisy Long-tailed Classification

Real-world datasets usually are class-imbalanced and corrupted by label noise. To solve the joint issue of long-tailed distribution and label noise, most previous works usually aim to design a noise detector to distinguish the noisy and clean samples. Despite their effectiveness, they may be limited in handling the joint issue effectively in a unified way. In this work, we develop a novel pseudo labeling method using class prototypes from the perspective of distribution matching, which can be solved with optimal transport (OT). By setting a manually-specific probability measure and using a learned transport plan to pseudo-label the training samples, the proposed method can reduce the side-effects of noisy and long-tailed data simultaneously. Then we introduce a simple yet effective filter criteria by combining the observed labels and pseudo labels to obtain a more balanced and less noisy subset for a robust model training. Extensive experiments demonstrate that our method can extract this class-balanced subset with clean labels, which brings effective performance gains for long-tailed classification with label noise.

cs.LG↗