Publications (88)
Strong Positive recurrence for potential and exponential mixing of equilibrium states of surface diffeomorphisms
Chiyi Luo, Dawei Yang
In this paper, we study the strong positive recurrence property for a large class of potentials of surface diffeomorphisms with positive entropy. We establish several…
Hyperbolicity versus non-hyperbolic ergodic measures inside homoclinic classes
Cheng Cheng, Sylvain Crovisier, Shaobo Gan +2
We prove that, for -generic diffeomorphisms, if a homoclinic class is not hyperbolic, then there is a non-hyperbolic ergodic measure supported on it. This proves a conjecture…
On the existence of attractors
Christian Bonatti, Ming Li, Dawei Yang
On every compact 3-manifold, we build a non-empty open set $\cU$ of $\Diff^1(M)$ such that, for every , every -generic diffeomorphism $f\in\cU\cap \Diff^r(M)$ has no…
I-LLM: Efficient Integer-Only Inference for Fully-Quantized Low-Bit Large Language Models
Xing Hu, Yuan Cheng, Dawei Yang +4
Post-training quantization (PTQ) serves as a potent technique to accelerate the inference of large language models (LLMs). Nonetheless, existing works still necessitate a considera…
Learning to Generate Synthetic 3D Training Data through Hybrid Gradient
Dawei Yang, Jia Deng
Synthetic images rendered by graphics engines are a promising source for training deep networks. However, it is challenging to ensure that they can help train a network to perform…
FQ-PETR: Fully Quantized Position Embedding Transformation for Multi-View 3D Object Detection
Jiangyong Yu, Changyong Shu, Sifan Zhou +4
Camera-based multi-view 3D detection is crucial for autonomous driving. PETR and its variants (PETRs) excel in benchmarks but face deployment challenges due to high computational c…
Pushing the Limits of BFP on Narrow Precision LLM Inference
Hui Wang, Yuan Cheng, Xiaomeng Han +3
The substantial computational and memory demands of Large Language Models (LLMs) hinder their deployment. Block Floating Point (BFP) has proven effective in accelerating linear ope…
Lyapunov optimizing measures and periodic measures for expanding maps
Wen Huang, Leiye Xu, Dawei Yang
We prove that there exists an open and dense subset in the space of expanding self-maps of the circle such that the Lyapunov minimizing measures…
BBAL: A Bidirectional Block Floating Point-Based Quantisation Accelerator for Large Language Models
Xiaomeng Han, Yuan Cheng, Jing Wang +6
Large language models (LLMs), with their billions of parameters, pose substantial challenges for deployment on edge devices, straining both memory capacity and computational resour…
Non-hyperbolic ergodic measures and horseshoes in partially hyperbolic homoclinic classes
Dawei Yang, Jinhua Zhang
We study a rich family of robustly non-hyperbolic transitive diffeomorphisms and we show that each ergodic measure is approached by hyperbolic sets in weak-topology and in entro…
FlashOcc: Fast and Memory-Efficient Occupancy Prediction via Channel-to-Height Plugin
Zichen Yu, Changyong Shu, Jiajun Deng +6
Given the capability of mitigating the long-tail deficiencies and intricate-shaped absence prevalent in 3D object detection, occupancy prediction has become a pivotal component in…
GSQ-Tuning: Group-Shared Exponents Integer in Fully Quantized Training for LLMs On-Device Fine-tuning
Sifan Zhou, Shuo Wang, Zhihang Yuan +3
Large Language Models (LLMs) fine-tuning technologies have achieved remarkable results. However, traditional LLM fine-tuning approaches face significant challenges: they require la…
MVCTrack: Boosting 3D Point Cloud Tracking via Multimodal-Guided Virtual Cues
Zhaofeng Hu, Sifan Zhou, Zhihang Yuan +3
3D single object tracking is essential in autonomous driving and robotics. Existing methods often struggle with sparse and incomplete point cloud scenarios. To address these limita…
Rel3D: A Minimally Contrastive Benchmark for Grounding Spatial Relations in 3D
Ankit Goyal, Kaiyu Yang, Dawei Yang +1
Understanding spatial relations (e.g., "laptop on table") in visual input is important for both humans and robots. Existing datasets are insufficient as they lack large-scale, high…
MoEQuant: Enhancing Quantization for Mixture-of-Experts Large Language Models via Expert-Balanced Sampling and Affinity Guidance
Xing Hu, Zhixuan Chen, Dawei Yang +5
Mixture-of-Experts (MoE) large language models (LLMs), which leverage dynamic routing and sparse activation to enhance efficiency and scalability, have achieved higher performance…
Robust transitivity of singular hyperbolic attractors
Sylvain Crovisier, Dawei Yang
Singular hyperbolicity is a weakened form of hyperbolicity that has been introduced for vector fields in order to allow non-isolated singularities inside the non-wandering set. A t…
Expansive homoclinic classes
Dawei Yang, Shaobo Gan
We prove that for generic diffeomorphisms, every expansive homoclinic class is hyperbolic.
Multi-parameter estimating photometric redshifts with artificial neural networks
Lili Li, Yanxia Zhang, Yongheng Zhao +1
We calculate photometric redshifts from the Sloan Digital Sky Survey Data Release 2 Galaxy Sample using artificial neural networks (ANNs). Different input patterns based on various…
Empirical measures of partially hyperbolic attractors
Sylvain Crovisier, Dawei Yang, Jinhua Zhang
In this paper, we study the limit measures of the empirical measures of Lebesgue almost every point in the basin of a partially hyperbolic attractor. They are strongly related to a…
R3-VAE: Reference Vector-Guided Rating Residual Quantization VAE for Generative Recommendation
Qiang Wan, Ze Yang, Dawei Yang +8
Generative Recommendation (GR) has gained traction for its merits of superior performance and cold-start capability. As the vital role in GR, Semantic Identifiers (SIDs) represent…
Effective SPR property for surface diffeomorphisms and three-dimensional vector fields
David Burguet, Chiyi Luo, Dawei Yang
In this paper, we prove that ergodic measures with large entropy give uniformly large measure to the set of points with simultaneously long unstable and long stable manifolds. As a…
On the abundance of SRB measures
Yongluo Cao, Zeya Mi, Dawei Yang
We prove the abundance of Sinai-Ruelle-Bowen measures for diffeomorphisms away from ones with a homoclinic tangency. This is motivated by conjectures of Palis on the existence of p…
MoBiE: Efficient Inference of Mixture of Binary Experts under Post-Training Quantization
Zhixiong Zhao, Zukang Xu, Zhixuan Chen +1
Mixture-of-Experts (MoE) based large language models (LLMs) offer strong performance but suffer from high memory and computation costs. Weight binarization provides extreme efficie…
Dominated chain recurrent class with singularities
Christian Bonatti, Shaobo Gan, Dawei Yang
We prove that for generic three-dimensional vector fields, domination implies singular hyperbolicity.
BWLA: Breaking the Barrier of W1AX Post-Training Quantization for LLMs
Zhixiong Zhao, Zukang Xu, Dawei Yang
Large language models (LLMs) have driven major progress in NLP, yet their substantial memory and compute demands still hinder practical deployment. Binarization can compress weight…
DLLMQuant: Quantizing Diffusion-based Large Language Models
Chen Xu, Dawei Yang
Diffusion-based large language models (DLLMs) have shown promise for non-autoregressive text generation, but their deployment is constrained by large model sizes and heavy computat…
Topologically stable manifolds for index- singular dominated splittings
Sylvain Crovisier, Dawei Yang
For vector fields, we study regular ergodic measures whose supports admit singular dominated splittings with one of the bundles having dimension . For such a measure ,…
Panoptic-FlashOcc: An Efficient Baseline to Marry Semantic Occupancy with Panoptic via Instance Center
Zichen Yu, Changyong Shu, Qianpu Sun +7
Panoptic occupancy poses a novel challenge by aiming to integrate instance occupancy and semantic occupancy within a unified framework. However, there is still a lack of efficient…
Reference-based OCT Angiogram Super-resolution with Learnable Texture Generation
Yuyan Ruan, Dawei Yang, Ziqi Tang +3
Optical coherence tomography angiography (OCTA) is a new imaging modality to visualize retinal microvasculature and has been readily adopted in clinics. High-resolution OCT angiogr…
Single-Image Depth Perception in the Wild
Weifeng Chen, Zhao Fu, Dawei Yang +1
This paper studies single-image depth perception in the wild, i.e., recovering depth from a single image taken in unconstrained settings. We introduce a new dataset "Depth in the W…
Partial Hyperbolicity and Homoclinic Tangencies
Sylvain Crovisier, Martin Sambarino, Dawei Yang
We show that any diffeomorphism of a compact manifold can be C1 approximated by diffeomorphisms exhibiting a homoclinic tangency or by diffeomorphisms having a partial hyperbolic s…
A Systematic Review on Affective Computing: Emotion Models, Databases, and Recent Advances
Yan Wang, Wei Song, Wei Tao +8
Affective computing plays a key role in human-computer interactions, entertainment, teaching, safe driving, and multimedia integration. Major breakthroughs have been made recently…
Symbolic extension of maps in arbitrary dimensions
Chiyi Luo, Dawei Yang
We prove that every self-map of a compact Riemannian manifold, in arbitrary dimension, admits a symbolic extension. This gives a positive answer to the conjectu…
OTARo: Once Tuning for All Precisions toward Robust On-Device LLMs
Shaoyuan Chen, Zhixuan Chen, Dawei Yang +2
Large Language Models (LLMs) fine-tuning techniques not only improve the adaptability to diverse downstream tasks, but also mitigate adverse effects of model quantization. Despite…
Unpaired Optical Coherence Tomography Angiography Image Super-Resolution via Frequency-Aware Inverse-Consistency GAN
Weiwen Zhang, Dawei Yang, Haoxuan Che +3
For optical coherence tomography angiography (OCTA) images, a limited scanning rate leads to a trade-off between field-of-view (FOV) and imaging resolution. Although larger FOV ima…
GSRender: Deduplicated Occupancy Prediction via Weakly Supervised 3D Gaussian Splatting
Qianpu Sun, Changyong Shu, Sifan Zhou +6
Weakly-supervised 3D occupancy perception is crucial for vision-based autonomous driving in outdoor environments. Previous methods based on NeRF often face a challenge in balancing…
Post-Training Quantization for Re-parameterization via Coarse & Fine Weight Splitting
Dawei Yang, Ning He, Xing Hu +4
Although neural networks have made remarkable advancements in various applications, they require substantial computational and memory resources. Network quantization is a powerful…
Ergodic measures with large entropy have long unstable manifolds for surface diffeomorphisms
Chiyi Luo, Dawei Yang
We prove that for ergodic measures with large entropy have long unstable manifolds for surface diffeomorphisms. Specifically, for any , there exist constants …
On Pesin's entropy formula for dominated splittings without mixed behavior
Dawei Yang, Yongluo Cao
For diffeomorphisms, we prove that the Pesin's entropy formula holds for some invariant measure supported on any topological attractor that admits a dominated splitting witho…
MGVQ: Synergizing Multi-dimensional Sensitivity-Aware and Gradient-Hessian Fusion for Vector Quantization
Zhong Wang, Zukang Xu, Xing Hu +1
Vision-Language Models (VLMs) achieve outstanding performance, yet their huge model size severely hinders deployment on edge devices with limited resources. As an efficient model c…
Randomization Inference for Peer Effects
Xinran Li, Peng Ding, Qian Lin +2
Many previous causal inference studies require no interference, that is, the potential outcomes of a unit do not depend on the treatments of other units. However, this no-interfere…
On the notions of singular domination and (multi-)singular hyperbolicity
Sylvain Crovisier, Adriana da Luz, Dawei Yang +1
The properties of uniform hyperbolicity and dominated splitting have been introduced to study the stability of the dynamics of diffeomorphisms. One meets difficulties when one trie…
Upper semi-continuity of metric entropy for diffeomorphisms
Chiyi Luo, Dawei Yang
We prove that for diffeomorphisms on a compact manifold with , if an invariant measure is a continuity point of the sum of positive…
Stably weakly shadowing transitive sets and dominated splittings
Dawei Yang
We prove that for any -stably weakly shadowing transitive set , either is a sink or a source, or admits a dominated splitting.
Volume growth and topological entropy of certain partially hyperbolic systems
Dawei Yang, Yuntao Zang
Let be a diffeomorphism on a compact manifold admitting a partially hyperbolic splitting $TM=E^{s}\oplus_{\prec} E^{1}\oplus_{\prec} E^{2}\cdots \oplus_{\prec}E^{l}…
ASVD: Activation-aware Singular Value Decomposition for Compressing Large Language Models
Zhihang Yuan, Yuzhang Shang, Yue Song +4
In this paper, we introduce a new post-training compression paradigm for Large Language Models (LLMs) to facilitate their wider adoption. We delve into LLM weight low-rank decompos…
Shape from Shading through Shape Evolution
Dawei Yang, Jia Deng
In this paper, we address the shape-from-shading problem by training deep networks with synthetic images. Unlike conventional approaches that combine deep learning and synthetic im…
PD-Quant: Post-Training Quantization based on Prediction Difference Metric
Jiawei Liu, Lin Niu, Zhihang Yuan +3
Post-training quantization (PTQ) is a neural network compression technique that converts a full-precision model into a quantized model using lower-precision data types. Although it…
On the partial hyperbolicity of robustly transitive sets with singularities
Xiao Wen, Dawei Yang
Homoclinic tangencies and singular hyperbolicity are involved in the Palis conjecture for vector fields. Typical three dimensional vector fields are well understood by recent works…
M&M VTO: Multi-Garment Virtual Try-On and Editing
Luyang Zhu, Yingwei Li, Nan Liu +3
We present M&M VTO, a mix and match virtual try-on method that takes as input multiple garment images, text description for garment layout and an image of a person. An example inpu…
SRB measures for diffeomorphisms with continuous invariant splittings
Zeya Mi, Yongluo Cao, Dawei Yang
We study the existence of SRB measures of C 2 diffeomorphisms for attractors whose bundles admit Holder continuous invariant (non-dominated) splittings. We prove the existence when…
A graph neural network-based multispectral-view learning model for diabetic macular ischemia detection from color fundus photographs
Qinghua He, Hongyang Jiang, Danqi Fang +7
Diabetic macular ischemia (DMI), marked by the loss of retinal capillaries in the macular area, contributes to vision impairment in patients with diabetes. Although color fundus ph…
Benchmarking the Reliability of Post-training Quantization: a Particular Focus on Worst-case Performance
Zhihang Yuan, Jiawei Liu, Jiaxiang Wu +6
Post-training quantization (PTQ) is a popular method for compressing deep neural networks (DNNs) without modifying their original architecture or training procedures. Despite its e…
Homoclinic tangencies and singular hyperbolicity for three-dimensional vector fields
Sylvain Crovisier, Dawei Yang
We prove that any vector field on a three-dimensional compact manifold can be approximated in the C1-topology by one which is singular hyperbolic or by one which exhibits a homocli…
NLI:Non-uniform Linear Interpolation Approximation of Nonlinear Operations for Efficient LLMs Inference
Jiangyong Yu, Xiaomeng Han, Xing Hu +3
Large Language Models (LLMs) have demonstrated remarkable performance across a wide range of tasks, but their deployment is often constrained by substantial memory footprints and c…
CAR-SAM: Cross-Attention Reconstruction for Post-Training Quantization of the Segment Anything Model
Houji Wen, Jiangyong Yu, Jun Li +1
Segment Anything Models (SAMs) are extensively used in computer vision for universal image segmentation, but deploying them on resource-constrained devices is challenging due to th…
TWLA: Achieving Ternary Weights and Low-Bit Activations for LLMs via Post-Training Quantization
Zhixiong Zhao, Zukang Xu, Zhixuan Chen +3
Large language models (LLMs) exhibit exceptional general language processing capabilities, but their memory and compute costs hinder deployment. Ternarization has emerged as a prom…
TryOnDiffusion: A Tale of Two UNets
Luyang Zhu, Dawei Yang, Tyler Zhu +5
Given two images depicting a person and a garment worn by another person, our goal is to generate a visualization of how the garment might look on the input person. A key challenge…
OstQuant: Refining Large Language Model Quantization with Orthogonal and Scaling Transformations for Better Distribution Fitting
Xing Hu, Yuan Cheng, Dawei Yang +6
Post-training quantization (PTQ) has emerged as a widely adopted technique for compressing and accelerating Large Language Models (LLMs). The major challenge in LLM quantization is…
PCDVQ: Enhancing Vector Quantization for Large Language Models via Polar Coordinate Decoupling
Yuxuan Yue, Zukang Xu, Zhihang Yuan +3
Large Language Models (LLMs) face significant challenges in edge deployment due to their massive parameter scale. Vector Quantization (VQ), a clustering-based quantization method,…
RWKVQuant: Quantizing the RWKV Family with Proxy Guided Hybrid of Scalar and Vector Quantization
Chen Xu, Yuxuan Yue, Zukang Xu +6
RWKV is a modern RNN architecture with comparable performance to Transformer, but still faces challenges when deployed to resource-constrained devices. Post Training Quantization (…
NVR: Vector Runahead on NPUs for Sparse Memory Access
Hui Wang, Zhengpeng Zhao, Jing Wang +11
Deep Neural Networks are increasingly leveraging sparsity to reduce the scaling up of model parameter size. However, reducing wall-clock time through sparsity and pruning remains c…
MQuant: Unleashing the Inference Potential of Multimodal Large Language Models via Full Static Quantization
JiangYong Yu, Sifan Zhou, Dawei Yang +7
Multimodal large language models (MLLMs) have garnered widespread attention due to their ability to understand multimodal input. However, their large parameter sizes and substantia…
Adversarial Objects Against LiDAR-Based Autonomous Driving Systems
Yulong Cao, Chaowei Xiao, Dawei Yang +4
Deep neural networks (DNNs) are found to be vulnerable against adversarial examples, which are carefully crafted inputs with a small magnitude of perturbation aiming to induce arbi…
Morse-Smale systems and horseshoes for three dimensional singular flows
Shaobo Gan, Dawei Yang
We prove that for every three-dimensional vector field, either it can be accumulated by Morse-Smale ones, or it can be accumulated by ones with a transverse homoclinic intersection…
VAEVQ: Enhancing Discrete Visual Tokenization through Variational Modeling
Sicheng Yang, Xing Hu, Qiang Wu +1
Vector quantization (VQ) transforms continuous image features into discrete representations, providing compressed, tokenized inputs for generative models. However, VQ-based framewo…
On the finiteness of uniform sinks
Dawei Yang, Yong Zhang
We study the finiteness of uniform sinks for flow. Precisely, we prove that, for , if a vector field has only hyperbolic singularities or sectionally dissipative si…
EA-ViT: Efficient Adaptation for Elastic Vision Transformer
Chen Zhu, Wangbo Zhao, Huiwen Zhang +9
Vision Transformers (ViTs) have emerged as a foundational model in computer vision, excelling in generalization and adaptation to downstream tasks. However, deploying ViTs to suppo…
RSAVQ: Riemannian Sensitivity-Aware Vector Quantization for Large Language Models
Zukang Xu, Xing Hu, Qiang Wu +1
Large language models (LLMs) have demonstrated remarkable performance across a wide range of natural language processing tasks. However, their exponentially increasing parameters p…
Improving Post-Training Quantization on Object Detection with Task Loss-Guided Lp Metric
Lin Niu, Jiawei Liu, Zhihang Yuan +3
Efficient inference for object detection networks is a major challenge on edge devices. Post-Training Quantization (PTQ), which transforms a full-precision model into low bit-width…
KBVQ-MoE: KLT-guided SVD with Bias-Corrected Vector Quantization for MoE Large Language Models
Zukang Xu, Zhixiong Zhao, Xing Hu +2
Mixture of Experts (MoE) models have achieved great success by significantly improving performance while maintaining computational efficiency through sparse expert activation. Howe…
Decorrelated Batch Normalization
Lei Huang, Dawei Yang, Bo Lang +1
Batch Normalization (BN) is capable of accelerating the training of deep models by centering and scaling activations within mini-batches. In this work, we propose Decorrelated Batc…
MeshAdv: Adversarial Meshes for Visual Recognition
Chaowei Xiao, Dawei Yang, Bo Li +2
Highly expressive models such as deep neural networks (DNNs) have been widely applied to various applications. However, recent studies show that DNNs are vulnerable to adversarial…
MambaQuant: Quantizing the Mamba Family with Variance Aligned Rotation Methods
Zukang Xu, Yuxuan Yue, Xing Hu +7
Mamba is an efficient sequence model that rivals Transformers and demonstrates significant potential as a foundational architecture for various tasks. Quantization is commonly used…
On the density of singular hyperbolic three-dimensional vector fields: a conjecture of Palis
Sylvain Crovisier, Dawei Yang
In this note we announce a result for vector fields on three-dimensional manifolds: those who are singular hyperbolic or exhibit a homoclinic tangency form a dense subset of the sp…
Ergodic optimization for some dynamical systems beyond uniform hyperbolicity
Dawei Yang, Jinhua Zhang
In this paper, we show that for several interesting systems beyond uniform hyperbolicity, any generic continuous function has a unique maximizing measure with zero entropy. In some…
On the growth rate of periodic orbits for vector fields
Wanlou Wu, Dawei Yang, Yong Zhang
We establish the relationship between the growth rate of periodic orbits and the topological entropy for generic vector fields: this extends a classical result of Katok for $…
FQ-PETR: Fully Quantized Position Embedding Transformation for Multi-View 3D Object Detection
Jiangyong Yu, Changyong Shu, Sifan Zhou +4
Camera-based multi-view 3D detection is crucial for autonomous driving. PETR and its variants (PETRs) excel in benchmarks but face deployment challenges due to high computational c…
Is an Ultra Large Natural Image-Based Foundation Model Superior to a Retina-Specific Model for Detecting Ocular and Systemic Diseases?
Qingshan Hou, Yukun Zhou, Jocelyn Hui Lin Goh +19
The advent of foundation models (FMs) is transforming medical domain. In ophthalmology, RETFound, a retina-specific FM pre-trained sequentially on 1.4 million natural images and 1.…
Continuity properties of ergodic measures of maximal entropy for surface diffeomorphisms
Jérôme Buzzi, Chiyi Luo, Dawei Yang
Let be a surface diffeomorphism with large entropy (more precisely, ). Then the number of ergodic measures of maximal entropy is upper se…
SPENet: Self-guided Prototype Enhancement Network for Few-shot Medical Image Segmentation
Chao Fan, Xibin Jia, Anqi Xiao +6
Few-Shot Medical Image Segmentation (FSMIS) aims to segment novel classes of medical objects using only a few labeled images. Prototype-based methods have made significant progress…
On the nonlinear Poincaré flow
Sylvain Crovisier, Dawei Yang
We develop a tool in order to analyse the dynamics of differentiable flows with singularities. It provides an abstract model for the local dynamics that can be used in order to con…
SRB measures for mostly expanding partially hyperbolic diffeomorphisms via the variational approach
David Burguet, Dawei Yang
By using the variational approach, we prove the existence of Sinai-Ruelle-Bowen measures for partially hyperbolic diffeomorphisms with mostly expanding properties. T…
TORQ: Two-Level Orthogonal Rotation for MXFP4 Quantization
Zukang Xu, Xing Hu, Dawei Yang
As Large Language Models (LLMs) advance toward practical deployment, the Microscaling FP4 (MXFP4) format has emerged as a cornerstone for next-generation low-bit inference, owing t…
On physical measures of multi-singular hyperbolic vector fields
Sylvain Crovisier, Xiaodong Wang, Dawei Yang +1
Bonatti and da Luz have introduced the class of \emph{multi-singular hyperbolic} vector fields to characterize systems whose periodic orbits and singularities do not bifurcate unde…
SAES-SVD: Self-Adaptive Suppression of Accumulated and Local Errors for SVD-based LLM Compression
Xing Hu, Dawei Yang, Yuan Cheng +2
The rapid growth in the parameter scale of large language models (LLMs) has created a high demand for efficient compression techniques. As a hardware-agnostic and highly compatible…
Information Entropy Guided Height-aware Histogram for Quantization-friendly Pillar Feature Encoder
Sifan Zhou, Zhihang Yuan, Dawei Yang +5
Real-time and high-performance 3D object detection plays a critical role in autonomous driving and robotics. Recent pillar-based 3D object detectors have gained significant attenti…
The Bernoulli property of Sinai-Ruelle-Bowen measures for flows
Chiyi Luo, Dawei Yang
We prove that for flows whose generating vector fields may have singularities, every weakly mixing hyperbolic SRB measure is Bernoullian.