arXiv subjects

Wei Huang

Publications and source records attributed to Wei Huang.

At least 199 records · Page 11Linked to original sources

arXiv2025

Decision SpikeFormer: Spike-Driven Transformer for Decision Making

Offline reinforcement learning (RL) enables policy training solely on pre-collected data, avoiding direct environment interaction - a crucial benefit for energy-constrained embodied AI applications. Although Artificial Neural Networks (ANN)-based methods perform well in offline RL, their high computational and energy demands motivate exploration of more efficient alternatives. Spiking Neural Networks (SNNs) show promise for such tasks, given their low power consumption. In this work, we introduce DSFormer, the first spike-driven transformer model designed to tackle offline RL via sequence modeling. Unlike existing SNN transformers focused on spatial dimensions for vision tasks, we develop Temporal Spiking Self-Attention (TSSA) and Positional Spiking Self-Attention (PSSA) in DSFormer to capture the temporal and positional dependencies essential for sequence modeling in RL. Additionally, we propose Progressive Threshold-dependent Batch Normalization (PTBN), which combines the benefits of LayerNorm and BatchNorm to preserve temporal dependencies while maintaining the spiking nature of SNNs. Comprehensive results in the D4RL benchmark show DSFormer's superiority over both SNN and ANN counterparts, achieving 78.4% energy savings, highlighting DSFormer's advantages not only in energy efficiency but also in competitive performance. Code and models are public at https://wei-nijuan.github.io/DecisionSpikeFormer.

cs.LG↗

arXiv2025

Temporal-contextual Event Learning for Pedestrian Crossing Intent Prediction

Ensuring the safety of vulnerable road users through accurate prediction of pedestrian crossing intention (PCI) plays a crucial role in the context of autonomous and assisted driving. Analyzing the set of observation video frames in ego-view has been widely used in most PCI prediction methods to forecast the cross intent. However, they struggle to capture the critical events related to pedestrian behaviour along the temporal dimension due to the high redundancy of the video frames, which results in the sub-optimal performance of PCI prediction. Our research addresses the challenge by introducing a novel approach called \underline{T}emporal-\underline{c}ontextual Event \underline{L}earning (TCL). The TCL is composed of the Temporal Merging Module (TMM), which aims to manage the redundancy by clustering the observed video frames into multiple key temporal events. Then, the Contextual Attention Block (CAB) is employed to adaptively aggregate multiple event features along with visual and non-visual data. By synthesizing the temporal feature extraction and contextual attention on the key information across the critical events, TCL can learn expressive representation for the PCI prediction. Extensive experiments are carried out on three widely adopted datasets, including PIE, JAAD-beh, and JAAD-all. The results show that TCL substantially surpasses the state-of-the-art methods. Our code can be accessed at https://github.com/dadaguailhb/TCL.

cs.CV↗

arXiv2025

Non-parametric Quantile Regression and Uniform Inference with Unknown Error Distribution

This paper studies the non-parametric estimation and uniform inference for the conditional quantile regression function (CQRF) with covariates exposed to measurement errors. We consider the case that the distribution of the measurement error is unknown and allowed to be either ordinary or super smooth. We estimate the density of the measurement error by the repeated measurements and propose the deconvolution kernel estimator for the CQRF. We derive the uniform Bahadur representation of the proposed estimator and construct the uniform confidence bands for the CQRF, uniformly in the sense for all covariates and a set of quantile indices, and establish the theoretical validity of the proposed inference. A data-driven approach for selecting the tuning parameter is also included. Monte Carlo simulations and a real data application demonstrate the usefulness of the proposed method.

stat.ME↗

arXiv2025

ANNs-SaDE: A Machine-Learning-Based Design Automation Framework for Microwave Branch-Line Couplers

The traditional method for designing branch-line couplers involves a trial-and-error optimization process that requires multiple design iterations through electromagnetic (EM) simulations. Thus, it is extremely time consuming and labor intensive. In this paper, a novel machine-learning-based framework is proposed to tackle this issue. It integrates artificial neural networks with a self-adaptive differential evolution algorithm (ANNs-SaDE). This framework enables the self-adaptive design of various types of microwave branch-line couplers by precisely optimizing essential electrical properties, such as coupling factor, isolation, and phase difference between output ports. The effectiveness of the ANNs-SaDE framework is demonstrated by the designs of folded single-stage branch-line couplers and multi-stage wideband branch-line couplers.

eess.SP↗

arXiv2025

DiffGED: Computing Graph Edit Distance via Diffusion-based Graph Matching

The Graph Edit Distance (GED) problem, which aims to compute the minimum number of edit operations required to transform one graph into another, is a fundamental challenge in graph analysis with wide-ranging applications. However, due to its NP-hard nature, traditional A* approaches often suffer from scalability issue, making them computationally intractable for large graphs. Many recent deep learning frameworks address GED by formulating it as a regression task, which, while efficient, fails to recover the edit path -- a central interest in GED. Furthermore, recent hybrid approaches that combine deep learning with traditional methods to recover the edit path often yield poor solution quality. These methods also struggle to generate candidate solutions in parallel, resulting in increased running times.In this paper, we present a novel approach, DiffGED, that leverages generative diffusion model to solve GED and recover the corresponding edit path. Specifically, we first generate multiple diverse node matching matrices in parallel through a diffusion-based graph matching model. Next, node mappings are extracted from each generated matching matrices in parallel, and each extracted node mapping can be simply transformed into an edit path. Benefiting from the generative diversity provided by the diffusion model, DiffGED is less likely to fall into local sub-optimal solutions, thereby achieving superior overall solution quality close to the exact solution. Experimental results on real-world datasets demonstrate that DiffGED can generate multiple diverse edit paths with exceptionally high accuracy comparable to exact solutions while maintaining a running time shorter than most of hybrid approaches.

cs.LG↗

arXiv2025

High-rate continuous-variable quantum key distribution over 100 km fiber with composable security

Quantum key distribution (QKD), providing a way to generate secret keys with information-theoretic security,is arguably one of the most significant achievements in quantum information. The continuous-variable QKD (CV-QKD) offers the potential advantage of achieving a higher secret key rate (SKR) within a metro area, as well as being compatible with the mature telecom industry. However, the SKR and transmission distance of state-of-the-art CV-QKD systems are currently limited. Here, based on the novelly proposed orthogonal-frequency-division-multiplexing (OFDM) CV-QKD protocol, we demonstrate for the first time a high-rate multi-carrier (MC) CV-QKD with a 10 GHz symbol rate that chieves Gbps SKR within 10km and Mbps SKR over 100 km in the finite-size regime under composable security against collective attacks. The record-breaking results are achieved by suitable optimization of subcarrier number and modulation variance, well-controlled excess noise induced by both OFDM mechanism and efficient DSP scheme, and high-performance post-processing capacity realized by heterogeneous computing scheme. The composable finite-size SKR reaches 1779.45 Mbps@5km, 1025.49 Mbps@10km, 370.50 Mbps@25km, 99.93 Mbps@50km, 25.70 Mbps@75km,and 2.25 Mbps@100km, which improves the SKR by two orders of magnitude and quintuples the maximal transmission distance compared to most recently reported CV-QKD results [Nature Communications, 13, 4740 (2022)]. Interestingly, it is experimentally verified that the SKR of the proposed MC CV-QKD can approach five times larger than that of the single-carrier CV-QKD with the same symbol rate without additional hardware costs. Our work constitutes a critical step towards future high-speed quantum metropolitan and access networks.

quant-ph↗

arXiv2025

High-rate discrete-modulated continuous-variable quantum key distribution with composable security

Continuous-variable quantum key distribution holds the potential to generate high secret key rates, making it a prime candidate for high-rate metropolitan quantum network applications. However, despite these promising opportunities, the realization of high-rate continuous-variable quantum key distribution systems with composable security remains an elusive goal. Here, we report a discrete-modulated continuous-variable quantum key distribution system with a composable secret key rate of 18.93 Mbps against collective attacks over a 25 km fiber channel. This record breaking rate is achieved through the probability shaped 16QAM-modulated protocol, which employs semidefinite programming to ensure its composable security. Furthermore, we have employed a fully digital and precise quantum signal processing technique to reduce excess noise to extremely low levels, thereby facilitating efficient broadband system operation. While ensuring low complexity and cost,our system achieves a performance advantage of over an order of magnitude compared to previous continuous-variable quantum key distribution systems, providing a promising solution for future deployment of quantum key distribution.

quant-ph↗

arXiv2025

High-rate discrete-modulated continuous-variable quantum key distribution with composable security

Continuous-variable quantum key distribution holds the potential to generate high secret key rates, making it a prime candidate for high-rate metropolitan quantum network applications. However, despite these promising opportunities, the realization of high-rate continuous-variable quantum key distribution systems with composable security remains an elusive goal. Here, we report a discrete-modulated continuous-variable quantum key distribution system with a composable secret key rate of 18.93 Mbps against collective attacks over a 25 km fiber channel. This record-breaking rate is achieved through the probability shaped 16QAM-modulated protocol, which employs semidefinite programming to ensure its composable security. Furthermore, we have employed a fully digital and precise quantum signal processing technique to reduce excess noise to extremely low levels, thereby facilitating efficient broadband system operation. While ensuring low complexity and cost, our system achieves a performance advantage of over an order of magnitude compared to previous continuous-variable quantum key distribution systems, providing a promising solution for future deployment of quantum key distribution.

quant-ph↗

arXiv2025

Large Language Model Enhanced Recommender Systems: A Survey

Large Language Model (LLM) has transformative potential in various domains, including recommender systems (RS). There have been a handful of research that focuses on empowering the RS by LLM. However, previous efforts mainly focus on LLM as RS, which may face the challenge of intolerant inference costs by LLM. Recently, the integration of LLM into RS, known as LLM-Enhanced Recommender Systems (LLMERS), has garnered significant interest due to its potential to address latency and memory constraints in real-world applications. This paper presents a comprehensive survey of the latest research efforts aimed at leveraging LLM to enhance RS capabilities. We identify a critical shift in the field with the move towards incorporating LLM into the online system, notably by avoiding their use during inference. Our survey categorizes the existing LLMERS approaches into three primary types based on the component of the RS model being augmented: Knowledge Enhancement, Interaction Enhancement, and Model Enhancement. We provide an in-depth analysis of each category, discussing the methodologies, challenges, and contributions of recent studies. Furthermore, we highlight several promising research directions that could further advance the field of LLMERS.

cs.IR↗

arXiv2025

On the Feature Learning in Diffusion Models

The predominant success of diffusion models in generative modeling has spurred significant interest in understanding their theoretical foundations. In this work, we propose a feature learning framework aimed at analyzing and comparing the training dynamics of diffusion models with those of traditional classification models. Our theoretical analysis demonstrates that diffusion models, due to the denoising objective, are encouraged to learn more balanced and comprehensive representations of the data. In contrast, neural networks with a similar architecture trained for classification tend to prioritize learning specific patterns in the data, often focusing on easy-to-learn components. To support these theoretical insights, we conduct several experiments on both synthetic and real-world datasets, which empirically validate our findings and highlight the distinct feature learning dynamics in diffusion models compared to classification.

stat.ML↗

arXiv2025

On the Optimization and Generalization of Two-layer Transformers with Sign Gradient Descent

The Adam optimizer is widely used for transformer optimization in practice, which makes understanding the underlying optimization mechanisms an important problem. However, due to the Adam's complexity, theoretical analysis of how it optimizes transformers remains a challenging task. Fortunately, Sign Gradient Descent (SignGD) serves as an effective surrogate for Adam. Despite its simplicity, theoretical understanding of how SignGD optimizes transformers still lags behind. In this work, we study how SignGD optimizes a two-layer transformer -- consisting of a softmax attention layer with trainable query-key parameterization followed by a linear layer -- on a linearly separable noisy dataset. We identify four stages in the training dynamics, each exhibiting intriguing behaviors. Based on the training dynamics, we prove the fast convergence but poor generalization of the learned transformer on the noisy dataset. We also show that Adam behaves similarly to SignGD in terms of both optimization and generalization in this setting. Additionally, we find that the poor generalization of SignGD is not solely due to data noise, suggesting that both SignGD and Adam requires high-quality data for real-world tasks. Finally, experiments on synthetic and real-world datasets empirically support our theoretical results.

cs.LG↗

arXiv2025

Mixture Compressor for Mixture-of-Experts LLMs Gains More

Mixture-of-Experts large language models (MoE-LLMs) marks a significant step forward of language models, however, they encounter two critical challenges in practice: 1) expert parameters lead to considerable memory consumption and loading latency; and 2) the current activated experts are redundant, as many tokens may only require a single expert. Motivated by these issues, we investigate the MoE-LLMs and make two key observations: a) different experts exhibit varying behaviors on activation reconstruction error, routing scores, and activated frequencies, highlighting their differing importance, and b) not all tokens are equally important -- only a small subset is critical. Building on these insights, we propose MC, a training-free Mixture-Compressor for MoE-LLMs, which leverages the significance of both experts and tokens to achieve an extreme compression. First, to mitigate storage and loading overheads, we introduce Pre-Loading Mixed-Precision Quantization, which formulates the adaptive bit-width allocation as a Linear Programming problem, where the objective function balances multi-factors reflecting the importance of each expert. Additionally, we develop Online Dynamic Pruning, which identifies important tokens to retain and dynamically select activated experts for other tokens during inference to optimize efficiency while maintaining performance. Our MC integrates static quantization and dynamic pruning to collaboratively achieve extreme compression for MoE-LLMs with less accuracy loss, ensuring an optimal trade-off between performance and efficiency. Extensive experiments confirm the effectiveness of our approach. For instance, at 2.54 bits, MC compresses 76.6% of the model, with only a 3.8% average accuracy loss. During dynamic inference, we further reduce activated parameters by 15%, with a performance drop of less than 0.6%.

cs.LG↗

arXiv2025

From Layers to States: A State Space Model Perspective to Deep Neural Network Layer Dynamics

The depth of neural networks is a critical factor for their capability, with deeper models often demonstrating superior performance. Motivated by this, significant efforts have been made to enhance layer aggregation - reusing information from previous layers to better extract features at the current layer, to improve the representational power of deep neural networks. However, previous works have primarily addressed this problem from a discrete-state perspective which is not suitable as the number of network layers grows. This paper novelly treats the outputs from layers as states of a continuous process and considers leveraging the state space model (SSM) to design the aggregation of layers in very deep neural networks. Moreover, inspired by its advancements in modeling long sequences, the Selective State Space Models (S6) is employed to design a new module called Selective State Space Model Layer Aggregation (S6LA). This module aims to combine traditional CNN or transformer architectures within a sequential framework, enhancing the representational capabilities of state-of-the-art vision networks. Extensive experiments show that S6LA delivers substantial improvements in both image classification and detection tasks, highlighting the potential of integrating SSMs with contemporary deep learning techniques.

cs.LG↗

arXiv2025

Can Diffusion Models Learn Hidden Inter-Feature Rules Behind Images?

Despite the remarkable success of diffusion models (DMs) in data generation, they exhibit specific failure cases with unsatisfactory outputs. We focus on one such limitation: the ability of DMs to learn hidden rules between image features. Specifically, for image data with dependent features ($\mathbf{x}$) and ($\mathbf{y}$) (e.g., the height of the sun ($\mathbf{x}$) and the length of the shadow ($\mathbf{y}$)), we investigate whether DMs can accurately capture the inter-feature rule ($p(\mathbf{y}|\mathbf{x})$). Empirical evaluations on mainstream DMs (e.g., Stable Diffusion 3.5) reveal consistent failures, such as inconsistent lighting-shadow relationships and mismatched object-mirror reflections. Inspired by these findings, we design four synthetic tasks with strongly correlated features to assess DMs' rule-learning abilities. Extensive experiments show that while DMs can identify coarse-grained rules, they struggle with fine-grained ones. Our theoretical analysis demonstrates that DMs trained via denoising score matching (DSM) exhibit constant errors in learning hidden rules, as the DSM objective is not compatible with rule conformity. To mitigate this, we introduce a common technique - incorporating additional classifier guidance during sampling, which achieves (limited) improvements. Our analysis reveals that the subtle signals of fine-grained rules are challenging for the classifier to capture, providing insights for future exploration.

cs.CV↗

arXiv2025

Accelerating Large Language Models through Partially Linear Feed-Forward Network

Large language models (LLMs) demonstrate remarkable capabilities but face deployment challenges due to their massive parameter counts. While existing compression techniques like pruning can reduce model size, it leads to significant accuracy degradation under high compression ratios. We present a novel perspective inspired by constant folding in compiler optimization. Our approach enables parameter reduction by treating activation functions in LLMs as linear functions. However, recent LLMs use complex non-linear activations like GELU that prevent direct application of this technique. We propose TARDIS, which enables optimization of LLMs with non-linear activations by partially approximating them with linear functions in frequently occurring input ranges. For outlier inputs, TARDIS employs an online predictor to dynamically fall back to original computations. Our experiments demonstrate that TARDIS achieves 80% parameter reduction in feed-forward networks, while significantly outperforming state-of-the-art pruning methods Wanda and RIA with up to 65% higher accuracy. In practical deployments for a 7B model, TARDIS achieves 1.6x end-to-end inference speedup when integrated with the vLLM serving system, and 1.4x speedup with the widely adopted HuggingFace implementation, while incurring only a 10.9% accuracy trade-off.

cs.LG↗

arXiv2025

An empirical study of LLaMA3 quantization: from LLMs to MLLMs

The LLaMA family, a collection of foundation language models ranging from 7B to 65B parameters, has become one of the most powerful open-source large language models (LLMs) and the popular LLM backbone of multi-modal large language models (MLLMs), widely used in computer vision and natural language understanding tasks. In particular, LLaMA3 models have recently been released and have achieved impressive performance in various domains with super-large scale pre-training on over 15T tokens of data. Given the wide application of low-bit quantization for LLMs in resource-constrained scenarios, we explore LLaMA3's capabilities when quantized to low bit-width. This exploration can potentially provide new insights and challenges for the low-bit quantization of LLaMA3 and other future LLMs, especially in addressing performance degradation issues that suffer in LLM compression. Specifically, we comprehensively evaluate the 10 existing post-training quantization and LoRA fine-tuning (LoRA-FT) methods of LLaMA3 on 1-8 bits and various datasets to reveal the low-bit quantization performance of LLaMA3. To uncover the capabilities of low-bit quantized MLLM, we assessed the performance of the LLaMA3-based LLaVA-Next-8B model under 2-4 ultra-low bits with post-training quantization methods. Our experimental results indicate that LLaMA3 still suffers from non-negligible degradation in linguistic and visual contexts, particularly under ultra-low bit widths. This highlights the significant performance gap at low bit-width that needs to be addressed in future developments. We expect that this empirical study will prove valuable in advancing future models, driving LLMs and MLLMs to achieve higher accuracy at lower bit to enhance practicality. Our project is released on https://github.com/Macaronlin/LLaMA3-Quantization , and quantized models are released at https://huggingface.co/Efficient-ML .

cs.LG↗

arXiv2024

Analogue of collectively induced transparency in metamaterials

Most recently, a brand new optical phenomenon, collectively induced transparency (CIT) has already been proposed in the cavity quantum electrodynamics system, which comes from the coupling between the cavity and ions and the quantum interference of collective ions. In this paper, we propose the CIT in terahertz (THz) metamaterial device by employing the coupling between bright mode and interference of dark modes for the first time. We give the theoretical analysis, analytical calculations and simulations to present the transmission spectrum of CIT metamaterials. Furthermore, we can observe the tendency of CIT's transmission spectrum by experiments which well verify our idea. Ideal CIT metamaterial device can produce a very high Q peak in the middle of transmission spectrum of Electromagnetically induced transparency (EIT), which can be useful for highly sensitive metamaterial sensors, optical switches and photo-memory.

physics.optics↗

arXiv2024

Breaking the Stage Barrier: A Novel Single-Stage Approach to Long Context Extension for Large Language Models

Recently, Large language models (LLMs) have revolutionized Natural Language Processing (NLP). Pretrained LLMs, due to limited training context size, struggle with handling long token sequences, limiting their performance on various downstream tasks. Current solutions toward long context modeling often employ multi-stage continual pertaining, which progressively increases the effective context length through several continual pretraining stages. However, those approaches require extensive manual tuning and human expertise. In this paper, we introduce a novel single-stage continual pretraining method, Head-Adaptive Rotary Position Encoding (HARPE), to equip LLMs with long context modeling capabilities while simplifying the training process. Our HARPE leverages different Rotary Position Encoding (RoPE) base frequency values across different attention heads and directly trains LLMs on the target context length. Extensive experiments on 4 language modeling benchmarks, including the latest RULER benchmark, demonstrate that HARPE excels in understanding and integrating long-context tasks with single-stage training, matching and even outperforming existing multi-stage methods. Our results highlight that HARPE successfully breaks the stage barrier for training LLMs with long context modeling capabilities.

cs.CL↗