arXiv subjects

Yu Wang

Publications and source records attributed to Yu Wang.

At least 721 records · Page 40Linked to original sources

arXiv2024

DiP-GO: A Diffusion Pruner via Few-step Gradient Optimization

Diffusion models have achieved remarkable progress in the field of image generation due to their outstanding capabilities. However, these models require substantial computing resources because of the multi-step denoising process during inference. While traditional pruning methods have been employed to optimize these models, the retraining process necessitates large-scale training datasets and extensive computational costs to maintain generalization ability, making it neither convenient nor efficient. Recent studies attempt to utilize the similarity of features across adjacent denoising stages to reduce computational costs through simple and static strategies. However, these strategies cannot fully harness the potential of the similar feature patterns across adjacent timesteps. In this work, we propose a novel pruning method that derives an efficient diffusion model via a more intelligent and differentiable pruner. At the core of our approach is casting the model pruning process into a SubNet search process. Specifically, we first introduce a SuperNet based on standard diffusion via adding some backup connections built upon the similar features. We then construct a plugin pruner network and design optimization losses to identify redundant computation. Finally, our method can identify an optimal SubNet through few-step gradient optimization and a simple post-processing procedure. We conduct extensive experiments on various diffusion models including Stable Diffusion series and DiTs. Our DiP-GO approach achieves 4.4 x speedup for SD-1.5 without any loss of accuracy, significantly outperforming the previous state-of-the-art methods.

cs.CV↗

arXiv2024

Error estimates between SGD with momentum and underdamped Langevin diffusion

Stochastic gradient descent with momentum is a popular variant of stochastic gradient descent, which has recently been reported to have a close relationship with the underdamped Langevin diffusion. In this paper, we establish a quantitative error estimate between them in the 1-Wasserstein and total variation distances.

stat.ML↗

arXiv2024

Towards Trustworthy Knowledge Graph Reasoning: An Uncertainty Aware Perspective

Recently, Knowledge Graphs (KGs) have been successfully coupled with Large Language Models (LLMs) to mitigate their hallucinations and enhance their reasoning capability, such as in KG-based retrieval-augmented frameworks. However, current KG-LLM frameworks lack rigorous uncertainty estimation, limiting their reliable deployment in high-stakes applications. Directly incorporating uncertainty quantification into KG-LLM frameworks presents challenges due to their complex architectures and the intricate interactions between the knowledge graph and language model components. To address this gap, we propose a new trustworthy KG-LLM framework, Uncertainty Aware Knowledge-Graph Reasoning (UAG), which incorporates uncertainty quantification into the KG-LLM framework. We design an uncertainty-aware multi-step reasoning framework that leverages conformal prediction to provide a theoretical guarantee on the prediction set. To manage the error rate of the multi-step process, we additionally introduce an error rate control module to adjust the error rate within the individual components. Extensive experiments show that our proposed UAG can achieve any pre-defined coverage rate while reducing the prediction set/interval size by 40% on average over the baselines.

cs.AI↗

arXiv2024

DiTFastAttn: Attention Compression for Diffusion Transformer Models

Diffusion Transformers (DiT) excel at image and video generation but face computational challenges due to the quadratic complexity of self-attention operators. We propose DiTFastAttn, a post-training compression method to alleviate the computational bottleneck of DiT. We identify three key redundancies in the attention computation during DiT inference: (1) spatial redundancy, where many attention heads focus on local information; (2) temporal redundancy, with high similarity between the attention outputs of neighboring steps; (3) conditional redundancy, where conditional and unconditional inferences exhibit significant similarity. We propose three techniques to reduce these redundancies: (1) Window Attention with Residual Sharing to reduce spatial redundancy; (2) Attention Sharing across Timesteps to exploit the similarity between steps; (3) Attention Sharing across CFG to skip redundant computations during conditional generation. We apply DiTFastAttn to DiT, PixArt-Sigma for image generation tasks, and OpenSora for video generation tasks. Our results show that for image generation, our method reduces up to 76% of the attention FLOPs and achieves up to 1.8x end-to-end speedup at high-resolution (2k x 2k) generation.

cs.CV↗

arXiv2024

CSKV: Training-Efficient Channel Shrinking for KV Cache in Long-Context Scenarios

Large Language Models (LLMs) have been widely adopted to process long-context tasks. However, the large memory overhead of the key-value (KV) cache poses significant challenges in long-context scenarios. Existing training-free KV cache compression methods typically focus on quantization and token pruning, which have compression limits, and excessive sparsity can lead to severe performance degradation. Other methods design new architectures with less KV overhead but require significant training overhead. To address the above two drawbacks, we further explore the redundancy in the channel dimension and apply an architecture-level design with minor training costs. Therefore, we introduce CSKV, a training-efficient Channel Shrinking technique for KV cache compression: (1) We first analyze the singular value distribution of the KV cache, revealing significant redundancy and compression potential along the channel dimension. Based on this observation, we propose using low-rank decomposition for key and value layers and storing the low-dimension features. (2) To preserve model performance, we introduce a bi-branch KV cache, including a window-based full-precision KV cache and a low-precision compressed KV cache. (3) To reduce the training costs, we minimize the layer-wise reconstruction loss for the compressed KV cache instead of retraining the entire LLMs. Extensive experiments show that CSKV can reduce the memory overhead of the KV cache by 80% while maintaining the model's long-context capability. Moreover, we show that our method can be seamlessly combined with quantization to further reduce the memory overhead, achieving a compression ratio of up to 95%. Code is available at https://github.com/wln20/CSKV.

cs.LG↗

arXiv2024

TAIA: Large Language Models are Out-of-Distribution Data Learners

Fine-tuning on task-specific question-answer pairs is a predominant method for enhancing the performance of instruction-tuned large language models (LLMs) on downstream tasks. However, in certain specialized domains, such as healthcare or harmless content generation, it is nearly impossible to obtain a large volume of high-quality data that matches the downstream distribution. To improve the performance of LLMs in data-scarce domains with domain-mismatched data, we re-evaluated the Transformer architecture and discovered that not all parameter updates during fine-tuning contribute positively to downstream performance. Our analysis reveals that within the self-attention and feed-forward networks, only the fine-tuned attention parameters are particularly beneficial when the training set's distribution does not fully align with the test set. Based on this insight, we propose an effective inference-time intervention method: Training All parameters but Inferring with only Attention (\trainallInfAttn). We empirically validate \trainallInfAttn using two general instruction-tuning datasets and evaluate it on seven downstream tasks involving math, reasoning, and knowledge understanding across LLMs of different parameter sizes and fine-tuning techniques. Our comprehensive experiments demonstrate that \trainallInfAttn achieves superior improvements compared to both the fully fine-tuned model and the base model in most scenarios, with significant performance gains. The high tolerance of \trainallInfAttn to data mismatches makes it resistant to jailbreaking tuning and enhances specialized tasks using general data. Code is available in \url{https://github.com/pixas/TAIA_LLM}.

cs.CL↗

arXiv2024

MedCare: Advancing Medical LLMs through Decoupling Clinical Alignment and Knowledge Aggregation

Large language models (LLMs) have shown substantial progress in natural language understanding and generation, proving valuable especially in the medical field. Despite advancements, challenges persist due to the complexity and diversity inherent in medical tasks, which can be categorized as knowledge-intensive tasks and alignment-required tasks. Previous approaches either ignore the latter task or focus on a minority of tasks and hence lose generalization. To address these drawbacks, we propose a progressive fine-tuning pipeline. This pipeline employs a Knowledge Aggregator and a Noise aggregator to encode diverse knowledge in the first stage and filter out detrimental information. In the second stage, we drop the Noise Aggregator to avoid the interference of suboptimal representation and leverage an additional alignment module optimized towards an orthogonal direction to the knowledge space to mitigate knowledge forgetting. Based on this two-stage paradigm, we proposed a Medical LLM through decoupling Clinical Alignment and Knowledge Aggregation (MedCare), which is designed to achieve state-of-the-art (SOTA) performance on over 20 medical tasks, as well as SOTA results on specific medical alignment tasks. Various model sizes of MedCare (1.8B, 7B, 14B) all demonstrate significant improvements over existing models with similar model sizes.

cs.CL↗

arXiv2024

SPF-EMPC Planner: A real-time multi-robot trajectory planner for complex environments with uncertainties

In practical applications, the unpredictable movement of obstacles and the imprecise state observation of robots introduce significant uncertainties for the swarm of robots, especially in cluster environments. However, existing methods are difficult to realize safe navigation, considering uncertainties, complex environmental structures, and robot swarms. This paper introduces an extended state model predictive control planner with a safe probability field to address the multi-robot navigation problem in complex, dynamic, and uncertain environments. Initially, the safe probability field offers an innovative approach to model the uncertainty of external dynamic obstacles, combining it with an unconstrained optimization method to generate safe trajectories for multi-robot online. Subsequently, the extended state model predictive controller can accurately track these generated trajectories while considering the robots' inherent model constraints and state uncertainty, thus ensuring the practical feasibility of the planned trajectories. Simulation experiments show a success rate four times higher than that of state-of-the-art algorithms. Physical experiments demonstrate the method's ability to operate in real-time, enabling safe navigation for multi-robot in uncertain environments.

cs.RO↗

arXiv2024

Reward-Robust RLHF in LLMs

As Large Language Models (LLMs) continue to progress toward more advanced forms of intelligence, Reinforcement Learning from Human Feedback (RLHF) is increasingly seen as a key pathway toward achieving Artificial General Intelligence (AGI). However, the reliance on reward-model-based (RM-based) alignment methods introduces significant challenges due to the inherent instability and imperfections of Reward Models (RMs), which can lead to critical issues such as reward hacking and misalignment with human intentions. In this paper, we introduce a reward-robust RLHF framework aimed at addressing these fundamental challenges, paving the way for more reliable and resilient learning in LLMs. Our approach introduces a novel optimization objective that carefully balances performance and robustness by incorporating Bayesian Reward Model Ensembles (BRME) to model the uncertainty set of reward functions. This allows the framework to integrate both nominal performance and minimum reward signals, ensuring more stable learning even with imperfect RMs. Empirical results demonstrate that our framework consistently outperforms baselines across diverse benchmarks, showing improved accuracy and long-term stability. We also provide a theoretical analysis, demonstrating that reward-robust RLHF approaches the stability of constant reward settings, which proves to be acceptable even in a stochastic-case analysis. Together, these contributions highlight the framework potential to enhance both the performance and stability of LLM alignment.

cs.LG↗

arXiv2024

Soft-Matter-Based Topological Vertical Cavity Surface Emitting Lasers

Polarized topological vertical cavity surface-emitting lasers (VCSELs), as stable and efficient on-chip light sources, play an important role in the next generation of optical storage and optical communications. However, most current topological lasers demand complex design and expensive fabrication processes, and their semiconductor-based structures pose challenges for flexible device applications. By use of an analogy with two-dimensional Semenov insulators in synthetic parametric space, we design and realize a one-dimensional optical superlattice (stacked polymerized cholesteric liquid crystal films and Mylar films), thereby we demonstrate a flexible, low threshold, circularly polarized topological VCSEL with high slope efficiency. We show that such a laser maintains a good single-mode property under low pump power and inherits the transverse spatial profile of the pump laser. Thanks to the soft-matter-based flexibility, our topological VCSEL can be "attached" to substrates of various shapes, enabling desired laser properties and robust beam steering even after undergoing hundreds of bends. Our results may find applications in consumer electronics, laser scanning and displays, as well as wearable devices.

physics.optics↗

arXiv2024

Detecting Subtle Differences between Human and Model Languages Using Spectrum of Relative Likelihood

Human and model-generated texts can be distinguished by examining the magnitude of likelihood in language. However, it is becoming increasingly difficult as language model's capabilities of generating human-like texts keep evolving. This study provides a new perspective by using the relative likelihood values instead of absolute ones, and extracting useful features from the spectrum-view of likelihood for the human-model text detection task. We propose a detection procedure with two classification methods, supervised and heuristic-based, respectively, which results in competitive performances with previous zero-shot detection methods and a new state-of-the-art on short-text detection. Our method can also reveal subtle differences between human and model languages, which find theoretical roots in psycholinguistics studies. Our code is available at https://github.com/CLCS-SUSTech/FourierGPT

cs.CL↗

arXiv2024

Hybrid Fourier Score Distillation for Efficient One Image to 3D Object Generation

Single image-to-3D generation is pivotal for crafting controllable 3D assets. Given its under-constrained nature, we attempt to leverage 3D geometric priors from a novel view diffusion model and 2D appearance priors from an image generation model to guide the optimization process. We note that there is a disparity between the generation priors of these two diffusion models, leading to their different appearance outputs. Specifically, image generation models tend to deliver more detailed visuals, whereas novel view models produce consistent yet over-smooth results across different views. Directly combining them leads to suboptimal effects due to their appearance conflicts. Hence, we propose a 2D-3D hybrid Fourier Score Distillation objective function, hy-FSD. It optimizes 3D Gaussians using 3D priors in spatial domain to ensure geometric consistency, while exploiting 2D priors in the frequency domain through Fourier transform for better visual quality. hy-FSD can be integrated into existing 3D generation methods and produce significant performance gains. With this technique, we further develop an image-to-3D generation pipeline to create high-quality 3D objects within one minute, named Fourier123. Extensive experiments demonstrate that Fourier123 excels in efficient generation with rapid convergence speed and visually-friendly generation results.

cs.CV↗

arXiv2024

Complexity factor for a static self-gravitating sphere in Rastall-Rainbow gravity

We generalized Herrera's definition of complexity factor for static spherically symmetric fluid distributions to Rastall-Rainbow theory of gravity. For this purpose, an energy-dependent equation of motion is employed in accordance with the principle of gravity's rainbow. It is found that the complexity factor appears in the orthogonal splitting of the Riemann curvature tensor, and measures the deviation of the value of the active gravitational mass from the simplest system under the combined corrections of Rastall and rainbow. In the low-energy limit, all the results we have obtained reduce to the counterparts of general relativity when the non-conserved parameter is taken to be one. We also demonstrate how to build an anisotropic or isotropic star model using complexity approach. In particular, the vanishing complexity factor condition in Rastall-Rainbow gravity is exactly the same as that derived in general relativity. This fact may imply a deeper geometric foundation for the complexity factor.

gr-qc↗

arXiv2024

Atmospheric Pressure Ammonia Synthesis on AuRu Catalysts Enabled by Plasmon-Controlled Hydrogenation and Nitrogen-species Desorption

Ammonia is a key component of fertilizer and a potential clean fuel and hydrogen carrier. The Haber-Bosch process for ammonia synthesis consumes more than half of industrial hydrogen and contributes up to ~3% of global greenhouse gas emissions. Light-driven reactions via surface plasmon resonances offer a less energy-intensive pathway for ammonia production by altering reaction intermediates. Here, we report gold-ruthenium plasmonic bimetallic alloys for ammonia synthesis at room temperature and pressure, driven by visible light. We use colloidal synthesis to create AuRu$_x$ alloys (x=0.1, 0.2, 0.3) and disperse these nanoparticles on MgO supports for gas-phase ammonia synthesis. We observe a ~60 $μ$mol/g/h reactivity and ~0.12% external quantum efficiency on a AuRu$_0$$_.$$_2$ sample under 100 mW/cm$^2$ visible light. In-situ diffuse reflective infrared Fourier transform spectroscopic measurements show that hydrogenation of nitrogen adsorbates is accelerated under light compared to thermocatalysis. Combining wavelength-dependent reactivity and spectroscopic findings with semi-classical electromagnetic modeling, we show plasmonic bimetallic alloys expedite ammonia synthesis by aiding hydrogenation of adsorbed nitrogen species via plasmon-mediated hot electrons. Quantum mechanical calculations reveal hydrogen-assisted N$_2$ splitting in the excited state is key to activating the reaction under ambient conditions. Therefore, light or H$_2$ alone cannot dissociate N$_2$ -- the key bottleneck to breaking N$_2$'s triple bond. Our findings are consistent with recent hypotheses on how nitrogenase enzymes catalyze ammonia production at mild conditions and provide insights for sustainable photochemical transformations.

physics.chem-ph↗

arXiv2024

Topological one-way Weyl fiber

Topological photonics enables unprecedented photon manipulation by realizing various topological states, such as corner states, edge states, and surface states. However, achieving a topological fiber state has remained elusive. Here, we demonstrate a topological fiber state in a Weyl gyromagnetic photonic crystal fiber. By applying an in-plane magnetic bias to a gyromagnetic photonic crystal fiber with broken parity-inversion symmetry, we create an asymmetrical Weyl bandgap that supports one-way fiber states associated with type-II Weyl points. Dispersion and topological invariant calculations reveal the transition from Weyl surface states to one-way Weyl fiber states. Electromagnetic field simulations confirm the existence of one-way Weyl fiber states and their robust transport in the presence of metallic obstacle along the transport path. Our findings offer an intriguing pathway for exploring novel topological states and guiding the design of topological fibers.

physics.optics↗

arXiv2024

Multi-Designated Detector Watermarking for Language Models

In this paper, we initiate the study of \emph{multi-designated detector watermarking (MDDW)} for large language models (LLMs). This technique allows model providers to generate watermarked outputs from LLMs with two key properties: (i) only specific, possibly multiple, designated detectors can identify the watermarks, and (ii) there is no perceptible degradation in the output quality for ordinary users. We formalize the security definitions for MDDW and present a framework for constructing MDDW for any LLM using multi-designated verifier signatures (MDVS). Recognizing the significant economic value of LLM outputs, we introduce claimability as an optional security feature for MDDW, enabling model providers to assert ownership of LLM outputs within designated-detector settings. To support claimable MDDW, we propose a generic transformation converting any MDVS to a claimable MDVS. Our implementation of the MDDW scheme highlights its advanced functionalities and flexibility over existing methods, with satisfactory performance metrics.

cs.CR↗

arXiv2024

Cross-Domain Few-Shot Object Detection via Enhanced Open-Set Object Detector

This paper studies the challenging cross-domain few-shot object detection (CD-FSOD), aiming to develop an accurate object detector for novel domains with minimal labeled examples. While transformer-based open-set detectors, such as DE-ViT, show promise in traditional few-shot object detection, their generalization to CD-FSOD remains unclear: 1) can such open-set detection methods easily generalize to CD-FSOD? 2) If not, how can models be enhanced when facing huge domain gaps? To answer the first question, we employ measures including style, inter-class variance (ICV), and indefinable boundaries (IB) to understand the domain gap. Based on these measures, we establish a new benchmark named CD-FSOD to evaluate object detection methods, revealing that most of the current approaches fail to generalize across domains. Technically, we observe that the performance decline is associated with our proposed measures: style, ICV, and IB. Consequently, we propose several novel modules to address these issues. First, the learnable instance features align initial fixed instances with target categories, enhancing feature distinctiveness. Second, the instance reweighting module assigns higher importance to high-quality instances with slight IB. Third, the domain prompter encourages features resilient to different styles by synthesizing imaginary domains without altering semantic contents. These techniques collectively contribute to the development of the Cross-Domain Vision Transformer for CD-FSOD (CD-ViTO), significantly improving upon the base DE-ViT. Experimental results validate the efficacy of our model.

cs.CV↗

arXiv2024

Towards the Mitigation of Confirmation Bias in Semi-supervised Learning: a Debiased Training Perspective

Semi-supervised learning (SSL) commonly exhibits confirmation bias, where models disproportionately favor certain classes, leading to errors in predicted pseudo labels that accumulate under a self-training paradigm. Unlike supervised settings, which benefit from a rich, static data distribution, SSL inherently lacks mechanisms to correct this self-reinforced bias, necessitating debiased interventions at each training step. Although the generation of debiased pseudo labels has been extensively studied, their effective utilization remains underexplored. Our analysis indicates that data from biased classes should have a reduced influence on parameter updates, while more attention should be given to underrepresented classes. To address these challenges, we introduce TaMatch, a unified framework for debiased training in SSL. TaMatch employs a scaling ratio derived from both a prior target distribution and the model's learning status to estimate and correct bias at each training step. This ratio adjusts the raw predictions on unlabeled data to produce debiased pseudo labels. In the utilization phase, these labels are differently weighted according to their predicted class, enhancing training equity and minimizing class bias. Additionally, TaMatch dynamically adjust the target distribution in response to the model's learning progress, facilitating robust handling of practical scenarios where the prior distribution is unknown. Empirical evaluations show that TaMatch significantly outperforms existing state-of-the-art methods across a range of challenging image classification tasks, highlighting the critical importance of both the debiased generation and utilization of pseudo labels in SSL.

cs.LG↗