arXiv subjects

Zixuan Zhang

Publications and source records attributed to Zixuan Zhang.

At least 37 records · Page 2Linked to original sources

arXiv2025

Orbital chiral lasing in twisted bilayer metasurfaces

Chirality is a fundamental concept in physics that underpins various phenomena in nonlinear optics, quantum physics, and topological photonics. Although the spin of a photon naturally brings chirality, orbital angular momentum can also become chirally active in the structures with a broken mirror symmetry. Here, we observe orbital chiral lasing from a twisted bilayer photonic structure leveraging its inherent structural chirality. Specifically, we design and fabricate a Moire-type optical structure by bonding and rotating two separate semiconductor membrane metasurfaces. We achieve single-mode lasing over a broad spectral range of 250 nm by optically pumping the twisted structure. The lasing emission exhibits orbital chiral characteristics, arising from helical and non-Hermitian couplings between clockwise and counter-clockwise rotating collective guided resonances, confirmed by polarization-resolved imaging and self-interference patterns. Our results provide the first observation of orbital chiral lasing in twisted photonics, and they can contribute to diverse applications of chiral light in diagnostics, optical manipulation, and communication with light.

physics.optics↗

arXiv2025

SlimMoE: Structured Compression of Large MoE Models via Expert Slimming and Distillation

The Mixture of Experts (MoE) architecture has emerged as a powerful paradigm for scaling large language models (LLMs) while maintaining inference efficiency. However, their enormous memory requirements make them prohibitively expensive to fine-tune or deploy in resource-constrained environments. To address this challenge, we introduce SlimMoE, a multi-stage compression framework for transforming large MoE models into much smaller, efficient variants without incurring the prohibitive costs of training from scratch. Our method systematically reduces parameter counts by slimming experts and transferring knowledge through intermediate stages, effectively mitigating the performance degradation common in one-shot pruning approaches. Using this framework, we compress Phi 3.5-MoE (41.9B total/6.6B activated parameters) to create Phi-mini-MoE (7.6B total/2.4B activated parameters) and Phi-tiny-MoE (3.8B total/1.1B activated parameters) using only 400B tokens--less than 10% of the original model's training data. These compressed models can be fine-tuned on a single GPU (A100 for Phi-mini-MoE, A6000 for Phi-tiny-MoE), making them highly suitable for academic and resource-limited settings. Our experiments demonstrate that these compressed models outperform others of similar size and remain competitive with larger models. For instance, Phi-mini-MoE achieves similar or better performance to Phi-3-mini using only 2/3 of the activated parameters and yields comparable MMLU scores to Llama 3.1 8B despite having significantly lower latency. Our findings demonstrate that structured pruning combined with staged distillation offers an effective path to creating high-quality, compact MoE models, paving the way for broader adoption of MoE architectures. We make our models publicly available at https://huggingface.co/microsoft/Phi-mini-MoE-instruct and https://huggingface.co/microsoft/Phi-tiny-MoE-instruct .

cs.LG↗

arXiv2025

LLMs Can Generate a Better Answer by Aggregating Their Own Responses

Large Language Models (LLMs) have shown remarkable capabilities across tasks, yet they often require additional prompting techniques when facing complex problems. While approaches like self-correction and response selection have emerged as popular solutions, recent studies have shown these methods perform poorly when relying on the LLM itself to provide feedback or selection criteria. We argue this limitation stems from the fact that common LLM post-training procedures lack explicit supervision for discriminative judgment tasks. In this paper, we propose Generative Self-Aggregation (GSA), a novel prompting method that improves answer quality without requiring the model's discriminative capabilities. GSA first samples multiple diverse responses from the LLM, then aggregates them to obtain an improved solution. Unlike previous approaches, our method does not require the LLM to correct errors or compare response quality; instead, it leverages the model's generative abilities to synthesize a new response based on the context of multiple samples. While GSA shares similarities with the self-consistency (SC) approach for response aggregation, SC requires specific verifiable tokens to enable majority voting. In contrast, our approach is more general and can be applied to open-ended tasks. Empirical evaluation demonstrates that GSA effectively improves response quality across various tasks, including mathematical reasoning, knowledge-based problems, and open-ended generation tasks such as code synthesis and conversational responses.

cs.CL↗

arXiv2025

IHEval: Evaluating Language Models on Following the Instruction Hierarchy

The instruction hierarchy, which establishes a priority order from system messages to user messages, conversation history, and tool outputs, is essential for ensuring consistent and safe behavior in language models (LMs). Despite its importance, this topic receives limited attention, and there is a lack of comprehensive benchmarks for evaluating models' ability to follow the instruction hierarchy. We bridge this gap by introducing IHEval, a novel benchmark comprising 3,538 examples across nine tasks, covering cases where instructions in different priorities either align or conflict. Our evaluation of popular LMs highlights their struggle to recognize instruction priorities. All evaluated models experience a sharp performance decline when facing conflicting instructions, compared to their original instruction-following performance. Moreover, the most competitive open-source model only achieves 48% accuracy in resolving such conflicts. Our results underscore the need for targeted optimization in the future development of LMs.

cs.CL↗

arXiv2025

A Minimalist Example of Edge-of-Stability and Progressive Sharpening

Recent advances in deep learning optimization have unveiled two intriguing phenomena under large learning rates: Edge of Stability (EoS) and Progressive Sharpening (PS), challenging classical Gradient Descent (GD) analyses. Current research approaches, using either generalist frameworks or minimalist examples, face significant limitations in explaining these phenomena. This paper advances the minimalist approach by introducing a two-layer network with a two-dimensional input, where one dimension is relevant to the response and the other is irrelevant. Through this model, we rigorously prove the existence of progressive sharpening and self-stabilization under large learning rates, and establish non-asymptotic analysis of the training dynamics and sharpness along the entire GD trajectory. Besides, we connect our minimalist example to existing works by reconciling the existence of a well-behaved ``stable set" between minimalist and generalist analyses, and extending the analysis of Gradient Flow Solution sharpness to our two-dimensional input scenario. These findings provide new insights into the EoS phenomenon from both parameter and input data distribution perspectives, potentially informing more effective optimization strategies in deep learning practice.

cs.LG↗

arXiv2025

Multiple magnetic states, valley electronics, and topological phase transitions in two-dimensional Janus XYZH (X = Sc, Y, La, Y = Cl, Br, I, and Z = S, Se, Te): From monolayers to bilayers

Exploring the coupling between layer, magnetism, valley, and topology in two-dimensional (2D) materials is an important approach to deepen our understanding of materials properties. We propose 27 stable ferromagnetic semiconductor monolayers of Janus XYZH (X = Sc, Y, La, Y = Cl, Br, I, and Z = S, Se, Te). All these monolayers exhibit spontaneous valley polarization, forming ferrovalley (FV) monolayers, showing anomalous valley Hall (AVH) effect. By applying external strain, the topological phase transitions including quantum anomalous Hall (QAH) effect can be introduced. In the ScBrSH bilayer system, AA and AB stacking configurations were constructed through interlayer sliding and rotational operation. The bilayer system exhibits interlayer antiferromagnetic (AFM) ordering with spontaneous valley polarization differing from the FV observed in monolayers. The sliding ferroelectricity observed in the AA stacking indicates that the system exhibits significant multiferroic characteristics. Further analysis shows that interlayer sliding can introduce a layer polarization anomalous valley Hall (LPAVH) effect, which can be precisely controlled by tuning the direction of the ferroelectric polarization. Upon applying external strain, the quantum layer spin Hall (QLSH) effect observed during the topological phase transition in the bilayer can be regarded as the superposition of two QAH monolayers. Furthermore, applying a twisting operation to the bilayer induces unexpected altermagnetism. Our study systematically reveals the rich physical properties of 2D XYZH materials, providing important theoretical foundations and guidance for the design and development of next-generation quantum devices.

cond-mat.mtrl-sci↗

arXiv2025

Double Distillation Network for Multi-Agent Reinforcement Learning

Multi-agent reinforcement learning typically employs a centralized training-decentralized execution (CTDE) framework to alleviate the non-stationarity in environment. However, the partial observability during execution may lead to cumulative gap errors gathered by agents, impairing the training of effective collaborative policies. To overcome this challenge, we introduce the Double Distillation Network (DDN), which incorporates two distillation modules aimed at enhancing robust coordination and facilitating the collaboration process under constrained information. The external distillation module uses a global guiding network and a local policy network, employing distillation to reconcile the gap between global training and local execution. In addition, the internal distillation module introduces intrinsic rewards, drawn from state information, to enhance the exploration capabilities of agents. Extensive experiments demonstrate that DDN significantly improves performance across multiple scenarios.

cs.MA↗

arXiv2025

Feasible Path SQP Algorithm for Simulation-based Optimization Surrogated with Differentiable Machine Learning Models

With the development of artificial intelligence, simulation-based optimization problems, which present a significant challenge in the process systems engineering community, are increasingly being addressed with the surrogate-based framework. In this work, we propose a deterministic algorithm framework based on feasible path sequential quadratic programming for optimizing differentiable machine learning models embedded problems. The proposed framework effectively addresses two key challenges: (i) achieving the computation of first- and second-order derivatives of machine learning models' outputs with respect to inputs; and (ii) by introducing the feasible path method, the massive intermediate variables resulting from the algebraic formulation of machine learning models eliminated. Surrogate models for six test functions and two process simulations were established and optimized. All six test functions were successfully optimized to the global optima, demonstrating the framework's effectiveness. The optimization time for all cases did not exceed 2s, highlighting the efficiency of the algorithm.

math.OC↗

arXiv2024

Nonparametric Classification on Low Dimensional Manifolds using Overparameterized Convolutional Residual Networks

Convolutional residual neural networks (ConvResNets), though overparameterized, can achieve remarkable prediction performance in practice, which cannot be well explained by conventional wisdom. To bridge this gap, we study the performance of ConvResNeXts, which cover ConvResNets as a special case, trained with weight decay from the perspective of nonparametric classification. Our analysis allows for infinitely many building blocks in ConvResNeXts, and shows that weight decay implicitly enforces sparsity on these blocks. Specifically, we consider a smooth target function supported on a low-dimensional manifold, then prove that ConvResNeXts can adapt to the function smoothness and low-dimensional structures and efficiently learn the function without suffering from the curse of dimensionality. Our findings partially justify the advantage of overparameterized ConvResNeXts over conventional machine learning models.

cs.LG↗

arXiv2024

Miniaturized spectrometer enabled by end-to-end deep learning on large-scale radiative cavity array

Miniaturized (mini-) spectrometers are highly desirable tools for chemical, biological, and medical diagnostics because of their potential for portable and in situ spectral detection. In this work, we propose and demonstrate a mini-spectrometer that combines a large-scale radiative cavity array with end-to-end deep learning networks. Specifically, we utilize high-Q bound states in continuum cavities with distinct radiation characteristics as the fundamental units to achieve parallel spectral detection. We realize a 36 $\times$ 30 cavity array that spans a wide spectral range from 1525 to 1605 nm with quality factors above 10^4. We further train a deep network with 8000 outputs to directly map arbitrary spectra to array responses excited by the out-of-plane incident. Experimental results demonstrate that the proposed mini-spectrometer can resolve unknown spectra with a resolution of 0.048 nm in a bandwidth of 80 nm and fidelity exceeding 95%, thus offering a promising method for compact, high resolution, and broadband spectroscopy.

physics.optics↗

arXiv2024

Chiral emission of vortex microlasers enabled by collective modes of guided resonances

Vortex lasers have attracted substantial attention in recent years owing to their wide array of applications such as micromanipulation, optical multiplexing, and quantum cryptography. In this work, we propose and demonstrate chiral emission of vortex microlaser leveraging the collective modes from omnidirectionally hybridizing the guided mode resonances (GMRs) within photonic crystal (PhC) slabs. Specifically, we encircle a central uniform PhC with a heterogeneous PhC that features a circular lateral boundary. Consequently, the bulk GMRs hybridize into a series of collective modes due to boundary scatterings, resulting in a vortex pattern in real space with a spiral phase front in its radiation. Benefiting from the long lifetime of GMRs as quasi-bound state in the continuum and using asymmetric pumping to lift the chiral symmetry, we demonstrate stable single-mode lasing oscillation with a low optical pumping threshold of $18~\mathrm{kW/cm^2}$ at room temperature. We identify the real-space vortex through polarization-resolved imaging and self-interference patterns, showing a vivid example of applying collective modes to realize compact and energy-efficient vortex microlasers.

physics.optics↗

arXiv2024

Robust Reinforcement Learning from Corrupted Human Feedback

Reinforcement learning from human feedback (RLHF) provides a principled framework for aligning AI systems with human preference data. For various reasons, e.g., personal bias, context ambiguity, lack of training, etc, human annotators may give incorrect or inconsistent preference labels. To tackle this challenge, we propose a robust RLHF approach -- $R^3M$, which models the potentially corrupted preference label as sparse outliers. Accordingly, we formulate the robust reward learning as an $\ell_1$-regularized maximum likelihood estimation problem. Computationally, we develop an efficient alternating optimization algorithm, which only incurs negligible computational overhead compared with the standard RLHF approach. Theoretically, we prove that under proper regularity conditions, $R^3M$ can consistently learn the underlying reward and identify outliers, provided that the number of outlier labels scales sublinearly with the preference sample size. Furthermore, we remark that $R^3M$ is versatile and can be extended to various preference optimization methods, including direct preference optimization (DPO). Our experiments on robotic control and natural language generation with large language models (LLMs) show that $R^3M$ improves robustness of the reward against several types of perturbations to the preference data.

cs.LG↗

arXiv2024

TextEE: Benchmark, Reevaluation, Reflections, and Future Challenges in Event Extraction

Event extraction has gained considerable interest due to its wide-ranging applications. However, recent studies draw attention to evaluation issues, suggesting that reported scores may not accurately reflect the true performance. In this work, we identify and address evaluation challenges, including inconsistency due to varying data assumptions or preprocessing steps, the insufficiency of current evaluation frameworks that may introduce dataset or data split bias, and the low reproducibility of some previous approaches. To address these challenges, we present TextEE, a standardized, fair, and reproducible benchmark for event extraction. TextEE comprises standardized data preprocessing scripts and splits for 16 datasets spanning eight diverse domains and includes 14 recent methodologies, conducting a comprehensive benchmark reevaluation. We also evaluate five varied large language models on our TextEE benchmark and demonstrate how they struggle to achieve satisfactory performance. Inspired by our reevaluation results and findings, we discuss the role of event extraction in the current NLP era, as well as future challenges and insights derived from TextEE. We believe TextEE, the first standardized comprehensive benchmarking tool, will significantly facilitate future event extraction research.

cs.CL↗

arXiv2024

Chem-FINESE: Validating Fine-Grained Few-shot Entity Extraction through Text Reconstruction

Fine-grained few-shot entity extraction in the chemical domain faces two unique challenges. First, compared with entity extraction tasks in the general domain, sentences from chemical papers usually contain more entities. Moreover, entity extraction models usually have difficulty extracting entities of long-tailed types. In this paper, we propose Chem-FINESE, a novel sequence-to-sequence (seq2seq) based few-shot entity extraction approach, to address these two challenges. Our Chem-FINESE has two components: a seq2seq entity extractor to extract named entities from the input sentence and a seq2seq self-validation module to reconstruct the original input sentence from extracted entities. Inspired by the fact that a good entity extraction system needs to extract entities faithfully, our new self-validation module leverages entity extraction results to reconstruct the original input sentence. Besides, we design a new contrastive loss to reduce excessive copying during the extraction process. Finally, we release ChemNER+, a new fine-grained chemical entity extraction dataset that is annotated by domain experts with the ChemNER schema. Experiments in few-shot settings with both ChemNER+ and CHEMET datasets show that our newly proposed framework has contributed up to 8.26% and 6.84% absolute F1-score gains respectively.

cs.CL↗

arXiv2024

Towards Better Generalization in Open-Domain Question Answering by Mitigating Context Memorization

Open-domain Question Answering (OpenQA) aims at answering factual questions with an external large-scale knowledge corpus. However, real-world knowledge is not static; it updates and evolves continually. Such a dynamic characteristic of knowledge poses a vital challenge for these models, as the trained models need to constantly adapt to the latest information to make sure that the answers remain accurate. In addition, it is still unclear how well an OpenQA model can transfer to completely new knowledge domains. In this paper, we investigate the generalization performance of a retrieval-augmented QA model in two specific scenarios: 1) adapting to updated versions of the same knowledge corpus; 2) switching to completely different knowledge domains. We observe that the generalization challenges of OpenQA models stem from the reader's over-reliance on memorizing the knowledge from the external corpus, which hinders the model from generalizing to a new knowledge corpus. We introduce Corpus-Invariant Tuning (CIT), a simple but effective training strategy, to mitigate the knowledge over-memorization by controlling the likelihood of retrieved contexts during training. Extensive experimental results on multiple OpenQA benchmarks show that CIT achieves significantly better generalizability without compromising the model's performance in its original corpus and domain.

cs.CL↗

arXiv2024

Large topological Hall effect arising from spin reorientation in kagome magnet Fe3Ge

Materials systems with spin chirality can provide ultra-high-density, ultra-fast, and ultralow-power information carriers for digital transformation. These material systems include magnetic skyrmions, chiral domain walls, spin reorientation,and so on. The topological Hall effect (THE) has been identified as the most convenient and effective tool for detecting the presence of spin chirality in these systems. The research on the THE that may arise from spin reorientation and specifically in Fe3Ge with spin reorientation remains an unexplored area, so we study the THE in Fe3Ge Conduct systematic research. X-Ray Diffraction (XRD) results indicate that our Fe3Ge ribbon sample has a D019 structure. First-principles calculations and magnetic and electrical testing confirm spin reorientation in the Fe3Ge ribbon sample at 350 K.The Hall resistivity test results are consistent with our expectations, indicating the presence of the THE in the Fe3Ge ribbon sample. The topological Hall resistivity reaches a maximum value of 0.69 mΩ cm at 400 K. For the first time, a detailed experimental study of the THE in Fe3Ge with spin reorientation has been conducted, introducing a new member to the family of THE.

cond-mat.mtrl-sci↗

arXiv2024

Observation of Berry curvature in non-Hermitian system from far-field radiation

Berry curvature that describes local geometrical properties of energy bands can elucidate many fascinating phenomena in solid-state, photonic, and phononic systems, given its connection to global topological invariants such as the Chern number. Despite its significance, the observation of Berry curvature poses a substantial challenging since wavefunctions are deeply embedded within the system. Here, we theoretically propose a correspondence between the geometry of far-field radiation and the underneath band topology of non-Hermitian systems, thus providing a general method to fully capture the Berry curvature without strongly disturbing the eigenstates. We further experimentally observe the Berry curvature in a honeycomb photonic crystal slab from polarimetry measurements and quantitatively obtain the non-trivial valley Chern number. Our work reveals the feasibility of retrieving the bulk band topology from escaping photons and paves the way to exploring intriguing topological landscapes in non-Hermitian systems.

physics.optics↗

arXiv2024

EVEDIT: Event-based Knowledge Editing with Deductive Editing Boundaries

The dynamic nature of real-world information necessitates efficient knowledge editing (KE) in large language models (LLMs) for knowledge updating. However, current KE approaches, which typically operate on (subject, relation, object) triples, ignore the contextual information and the relation among different knowledge. Such editing methods could thus encounter an uncertain editing boundary, leaving a lot of relevant knowledge in ambiguity: Queries that could be answered pre-edit cannot be reliably answered afterward. In this work, we analyze this issue by introducing a theoretical framework for KE that highlights an overlooked set of knowledge that remains unchanged and aids in knowledge deduction during editing, which we name as the deduction anchor. We further address this issue by proposing a novel task of event-based knowledge editing that pairs facts with event descriptions. This task manifests not only a closer simulation of real-world editing scenarios but also a more logically sound setting, implicitly defining the deduction anchor to address the issue of indeterminate editing boundaries. We empirically demonstrate the superiority of event-based editing over the existing setting on resolving uncertainty in edited models, and curate a new benchmark dataset EvEdit derived from the CounterFact dataset. Moreover, while we observe that the event-based setting is significantly challenging for existing approaches, we propose a novel approach Self-Edit that showcases stronger performance, achieving 55.6% consistency improvement while maintaining the naturalness of generation.

cs.CL↗