Published in Computer Physics Communications, 2016
This work presents an adaptive multi-GPU Exchange Monte Carlo approach for the simulation of the 3D Random Field Ising Model (RFIM).
Download here
Published in Physical Review E, 2018
We study critical bond percolation on a seven-dimensional (7D) hypercubic lattice with periodic boundary conditions and on the complete graph (CG) of finite volume (number of vertices).
Download here
Published in ECAI, 2020
We perform theoretical computation on linear dropout networks and a series of experiments on dropout networks with different activation functions.
Download here
Published in IJCAI, 2021
In this work, we study the dynamics of ultra-wide networks across a range of architectures, including Fully Connected Networks (FCNs) and Convolutional Neural Networks (CNNs) with orthogonal initialization via neural tangent kernel (NTK).
Download here
Published in Pattern Recognition Letters, 2021
In order to address such shortcomings, we propose a Gaussian Process Latent Variable Model Factorization (GPLVMF) method, where we apply an appropriate prior to the original GP model.
Download here
Published in NeurIPS, 2021
We propose to establish the equivalence between NN and SVM, and specifically, the infinitely wide NN trained by soft margin loss and the standard soft margin SVM with NTK trained by subgradient descent.
Download here
Published in ICLR, 2022
This work exploits the Graph Neural Tangent Kernel (GNTK), which governs the optimization trajectory under gradient descent for wide GCNs. We formulate the asymptotic behaviors of GNTK in the large depth, which enables us to reveal the dropping trainability of wide and deep GCNs at an exponential rate in the optimization process.
Download here
Published in ICLR, 2022
This work targets automated designing and scaling of Vision Transformers (ViTs). We propose As-ViT, an auto-scaling framework for ViTs without training, which automatically discovers and scales up ViTs in an efficient and principled manner.
Download here
Published in Knowledge-Based Systems, 2022
We formulate the performance of GNNs mathematically with respect to the properties of their edges, elucidating how the performance drop can be avoided by pruning negative edges and nonbridges. This leads to our simple but effective two-step method for GNN pruning, leveraging the saliency metrics for the network pruning while sparsifying the graph with preservation of the loss performance.
Download here
Published in NeurIPS, 2022
We theoretically characterize the impact of connectivity patterns on the convergence of DNNs under gradient descent training in fine granularity. By analyzing a wide network’s Neural Network Gaussian Process (NNGP), we are able to depict how the spectrum of an NNGP kernel propagates through a particular connectivity pattern, and how that affects the bound of convergence rates.
Download here
Published in NeurIPS, 2022
In this work, we leverage influence functions, the functional derivatives of the loss function, to theoretically reveal the operation selection part in DARTS and estimate the candidate operation importance by approximating its influence on the supernet with Taylor expansions. We show the operation strength is not only related to the magnitude but also secondorder information, leading to a fundamentally new criterion for operation selection in DARTS, named Influential Magnitude.
Download here
Published in NeurIPS, 2022
In this paper, by exploring the connection between the generalization performance and the training dynamics, we propose a theory-driven deep active learning method (dynamicAL) which selects samples to maximize training dynamics. In particular, we prove that the convergence speed of training and the generalization performance are positively correlated under the ultra-wide condition and show that maximizing the training dynamics leads to better generalization performance.
Download here
Published in NeurIPS, 2022
This paper, for the first time, leverages a bi-level formulation to estimate the relative importance of peers with a close-form, to further boost the effectiveness of the distillation from each other. Extensive experiments show the generalization of the proposed framework, which outperforms existing online distillation methods on a variety of deep neural networks.
Download here
Published in TMLR, 2023
This paper proposes a theoretical convergence and generalization analysis for Deep PAC-Bayesian learning. For a deep and wide probabilistic neural network, our analysis shows that PAC-Bayesian learning corresponds to solving a kernel ridge regression when the probabilistic neural tangent kernel (PNTK) is used as the kernel.
Download here
Published in Auto-ML, 2023
To facilitate the interpretation and understanding of the architecture design by AutoML, we target connecting a bigger picture: how does the architecture jointly impact its expressivity, convergence, and generalization?
Download here
Published in TMLR, 2023
We theoretically study the concentration property of features obtained by neighborhood aggregation on homophilic and heterophilic graphs, introduce the single-pass graph contrastive learning loss based on the property, and provide performance guarantees for the minimizer of the loss on downstream tasks.
Download here
Published in NeurIPS, 2023
We propose Feature Augmented Training (FeAT), to enforce the model to learn richer features ready for OOD generalization
Download here
Published in NeurIPS, 2023
We propose Fed-CO2, a universal FL framework that handles both label distribution skew and feature skew within a Cooperation mechanism between the Online and Offline models
Download here
Published in NeurIPS, 2023
We establish a new connection between the loss dynamics of gradient flow and general kernel machines by proposing a new kernel, called loss path kernel.
Download here
Published in ICLR, 2024
This work aims to establish a unified theoretical foundation for understanding FL through feature learning theory.
Download here
Published in ICLR, 2024
This paper introduces an Adaptive, Dynamic, and Automated framework for identifying Graph Lottery Tickets (AdaGLT).
Download here
Published in CVPR, 2024
We present Federated Prompts Cooperation via Optimal Transport (FedOTP), which introduces efficient collaborative prompt learning strategies to capture diverse category traits on a per-client basis.
Download here
Published in KDD, 2024
In this paper, we transfer the prevailing concept of ``one node one receptive field” to the heterophilic graph.
Download here
Published in ICML, 2024
We provably show that both uncertainty-based and diversity-based NAL are inherently amenable to one and the same principle, i.e., striving to prioritize samples that contain yet-to-be-learned features.
Download here
Published in ICML, 2024
In this work, we propose to guide diffusion models for adversarial purification using contrastive guidance.
Download here
Published in NeurIPS, 2024
We investigate the non-convex dynamics of a one-layer linear causal self-attention model autoregressively trained by gradient flow, where the sequences are generated by an AR process.
Download here
Published in NeurIPS, 2024
In this work, we propose to parameterize the weights as a sum of low-rank and sparse matrices for pretraining, which we call SLTrain.
Download here
Published in NeurIPS, 2024
In this paper, by focusing on dataset distillation for kernel ridge regression (KRR), we show that one data point per class is already necessary and sufficient to recover the original models performance in many settings.
Download here
Published in NeurIPS, 2024
In this work, we construct a theoretical analysis framework for prompt-based federated learning via feature learning theory.
Download here
Published in NeurIPS, 2024
This work provides a fine-grained mathematical analysis to show how transformers leverage the multi-concept semantics of words to enable powerful ICL and excellent out-of-distribution ICL abilities, offering insights into how transformers innovate solutions for certain unseen tasks encoded with multiple cross-concept semantics
Download here
Published in NeurIPS, 2024
This work delves deeply into the benign overfitting perspective of transformers in vision
Download here
Published in NeurIPS, 2024
In this work, we introduce a feature learning theory framework that provides a theoretical foundation for understanding the differences between multi-modal and single-modal contrastive learning.
Download here
Published:
Recently researcher find that the dynamics of infinitely-wide neural networks under gradient desent training are captured by neural tangent kernel. With the help of neural tangent kernel, researcher can prove that over-paramterized neural network can find global minimum, which is a milestone in the area of deep learning theory. This talk will present the basic properties of neural tangent kernel and my research output regarding neural tangent kernel.
Published:
The learning dynamics of neural networks trained by gradient descent are captured by the so-called neural tangent kernel (NTK) in the infinite-width limit. The NTK has been a powerful tool for researchers to understand the optimization and generalization of over-parameterized networks. In this talk, the foundation of the NTK in addition to its application to orthogonally-initialized networks and ultra-wide graph networks will be introduced.
Published:
Deep learning has been responsible for a step-change in performance across machine learning, setting new benchmarks in a large number of applications. During my Ph.D. study, I seek to understand the theoretical properties of deep neural networks and close the gap between the theory and application sides. This presentation will introduce three concrete works with respect to the neural tangent kernel (NTK), one of the seminal advances in deep learning theory recently.
Published:
We formulate the asymptotic behaviors of GNTK in the large depth, which enables us to reveal the dropping trainability of wide and deep GCNs at an exponential rate in the optimization process. Additionally, we extend our theoretical framework to analyze residual connection-based techniques, which are found to be merely able to mitigate the exponential decay of trainability mildly. Inspired by our theoretical insights on trainability, we propose Critical DropEdge, a connectivity-aware and graph-adaptive sampling method, to alleviate the exponential decay problem more fundamentally.
Published:
Understanding the learning dynamics of neural networks with (stochastic) gradient descent is a long-term goal for deep learning theory research. In this talk, the trend from the neural tangent kernel (NTK) regime to feature learning dynamics will be introduced. The NTK has been a powerful tool for researchers to understand the optimization and generalization of over-parameterized networks. We first introduce the foundation of the NTK in addition to its application to neural architecture search and active learning. Furthermore, more recent works found that the neural networks are performing feature learning during gradient descent training. We will then introduce how feature learning emerges and its application in understanding the role of graph convolution in graph neural networks.
Published:
Published:
Published:
Published:
Undergraduate course, University of Science and Technology of China, 2015
Teaching assistant of the Thermodynamics and Statistical Physics course for undergraduate students.
Undergraduate course, University of Science and Technology of China, 2016
Teaching assistant of the Statistical Physics course for undergraduate students. Achieved outstanding teaching assisant award.