Wei Hu

Qihao Zhao, Chen Jiang, Wei Hu et al.

Recently, multi-expert methods have led to significant improvements in long-tail recognition (LTR). We summarize two aspects that need further enhancement to contribute to LTR boosting: (1) More diverse experts; (2) Lower model variance. However, the previous methods didn't handle them well. To this end, we propose More Diverse experts with Consistency Self-distillation (MDCS) to bridge the gap left by earlier methods. Our MDCS approach consists of two core components: Diversity Loss (DL) and Consistency Self-distillation (CS). In detail, DL promotes diversity among experts by controlling their focus on different categories. To reduce the model variance, we employ KL divergence to distill the richer knowledge of weakly augmented instances for the experts' self-distillation. In particular, we design Confident Instance Sampling (CIS) to select the correctly classified instances for CS to avoid biased/noisy knowledge. In the analysis and ablation study, we demonstrate that our method compared with previous work can effectively increase the diversity of experts, significantly reduce the variance of the model, and improve recognition accuracy. Moreover, the roles of our DL and CS are mutually reinforcing and coupled: the diversity of experts benefits from the CS, and the CS cannot achieve remarkable results without the DL. Experiments show our MDCS outperforms the state-of-the-art by 1% $\sim$ 2% on five popular long-tailed benchmarks, including CIFAR10-LT, CIFAR100-LT, ImageNet-LT, Places-LT, and iNaturalist 2018. The code is available at https://github.com/fistyee/MDCS.

Wei Hu, Yangyu Huang, Fan Zhang et al.

Benefit from large-scale training datasets, deep Convolutional Neural Networks(CNNs) have achieved impressive results in face recognition(FR). However, tremendous scale of datasets inevitably lead to noisy data, which obviously reduce the performance of the trained CNN models. Kicking out wrong labels from large-scale FR datasets is still very expensive, although some cleaning approaches are proposed. According to the analysis of the whole process of training CNN models supervised by angular margin based loss(AM-Loss) functions, we find that the $θ$ distribution of training samples implicitly reflects their probability of being clean. Thus, we propose a novel training paradigm that employs the idea of weighting samples based on the above probability. Without any prior knowledge of noise, we can train high performance CNN models with large-scale FR datasets. Experiments demonstrate the effectiveness of our training paradigm. The codes are available at https://github.com/huangyangyu/NoiseFace.

Wei Hu, QiHao Zhao, Yangyu Huang et al.

Learning deep neural network (DNN) classifier with noisy labels is a challenging task because the DNN can easily over-fit on these noisy labels due to its high capability. In this paper, we present a very simple but effective training paradigm called P-DIFF, which can train DNN classifiers but obviously alleviate the adverse impact of noisy labels. Our proposed probability difference distribution implicitly reflects the probability of a training sample to be clean, then this probability is employed to re-weight the corresponding sample during the training process. P-DIFF can also achieve good performance even without prior knowledge on the noise rate of training samples. Experiments on benchmark datasets also demonstrate that P-DIFF is superior to the state-of-the-art sample selection methods.

Rohith Kuditipudi, Xiang Wang, Holden Lee et al.

Mode connectivity is a surprising phenomenon in the loss landscape of deep nets. Optima -- at least those discovered by gradient-based optimization -- turn out to be connected by simple paths on which the loss function is almost constant. Often, these paths can be chosen to be piece-wise linear, with as few as two segments. We give mathematical explanations for this phenomenon, assuming generic properties (such as dropout stability and noise stability) of well-trained deep nets, which have previously been identified as part of understanding the generalization properties of deep nets. Our explanation holds for realistic multilayer nets, and experiments are presented to verify the theory.

Wei Hu, Zhiyuan Li, Dingli Yu

Over-parameterized deep neural networks trained by simple first-order methods are known to be able to fit any labeling of data. Such over-fitting ability hinders generalization when mislabeled training examples are present. On the other hand, simple regularization methods like early-stopping can often achieve highly nontrivial performance on clean test data in these scenarios, a phenomenon not theoretically understood. This paper proposes and analyzes two simple and intuitive regularization methods: (i) regularization by the distance between the network parameters to initialization, and (ii) adding a trainable auxiliary variable to the network output for each training example. Theoretically, we prove that gradient descent training with either of these two methods leads to a generalization guarantee on the clean data distribution despite being trained using noisy labels. Our generalization analysis relies on the connection between wide neural network and neural tangent kernel (NTK). The generalization bound is independent of the network size, and is comparable to the bound one can get when there is no label noise. Experimental results verify the effectiveness of these methods on noisily labeled datasets.

Simon S. Du, Wei Hu

We prove that for an $L$-layer fully-connected linear neural network, if the width of every hidden layer is $\tildeΩ(L \cdot r \cdot d_{\mathrm{out}} \cdot κ^3 )$, where $r$ and $κ$ are the rank and the condition number of the input data, and $d_{\mathrm{out}}$ is the output dimension, then gradient descent with Gaussian random initialization converges to a global minimum at a linear rate. The number of iterations to find an $ε$-suboptimal solution is $O(κ\log(\frac{1}ε))$. Our polynomial upper bound on the total running time for wide deep linear networks and the $\exp\left(Ω\left(L\right)\right)$ lower bound for narrow deep linear neural networks [Shamir, 2018] together demonstrate that wide layers are necessary for optimizing deep models.

Sanjeev Arora, Nadav Cohen, Noah Golowich et al.

We analyze speed of convergence to global optimum for gradient descent training a deep linear neural network (parameterized as $x \mapsto W_N W_{N-1} \cdots W_1 x$) by minimizing the $\ell_2$ loss over whitened data. Convergence at a linear rate is guaranteed when the following hold: (i) dimensions of hidden layers are at least the minimum of the input and output dimensions; (ii) weight matrices at initialization are approximately balanced; and (iii) the initial loss is smaller than the loss of any rank-deficient solution. The assumptions on initialization (conditions (ii) and (iii)) are necessary, in the sense that violating any one of them may lead to convergence failure. Moreover, in the important case of output dimension 1, i.e. scalar regression, they are met, and thus convergence to global optimum holds, with constant probability under a random initialization scheme. Our results significantly extend previous analyses, e.g., of deep linear residual networks (Bartlett et al., 2018).

Simon S. Du, Wei Hu

We consider the convex-concave saddle point problem $\min_{x}\max_{y} f(x)+y^\top A x-g(y)$ where $f$ is smooth and convex and $g$ is smooth and strongly convex. We prove that if the coupling matrix $A$ has full column rank, the vanilla primal-dual gradient method can achieve linear convergence even if $f$ is not strongly convex. Our result generalizes previous work which either requires $f$ and $g$ to be quadratic functions or requires proximal mappings for both $f$ and $g$. We adopt a novel analysis technique that in each iteration uses a "ghost" update as a reference, and show that the iterates in the primal-dual gradient method converge to this "ghost" sequence. Using the same technique we further give an analysis for the primal-dual stochastic variance reduced gradient (SVRG) method for convex-concave saddle point problems with a finite-sum structure.