arXiv subjects

Yichao Wu

Publications and source records attributed to Yichao Wu.

At least 55 records · Page 3Linked to original sources

arXiv2021

One to Transfer All: A Universal Transfer Framework for Vision Foundation Model with Few Data

The foundation model is not the last chapter of the model production pipeline. Transferring with few data in a general way to thousands of downstream tasks is becoming a trend of the foundation model's application. In this paper, we proposed a universal transfer framework: One to Transfer All (OTA) to transfer any Vision Foundation Model (VFM) to any downstream tasks with few downstream data. We first transfer a VFM to a task-specific model by Image Re-representation Fine-tuning (IRF) then distilling knowledge from a task-specific model to a deployed model with data produced by Downstream Image-Guided Generation (DIGG). OTA has no dependency on upstream data, VFM, and downstream tasks when transferring. It also provides a way for VFM researchers to release their upstream information for better transferring but not leaking data due to privacy requirements. Massive experiments validate the effectiveness and superiority of our methods in few data setting. Our code will be released.

cs.CV↗

arXiv2021

Inter-class Discrepancy Alignment for Face Recognition

The field of face recognition (FR) has witnessed great progress with the surge of deep learning. Existing methods mainly focus on extracting discriminative features, and directly compute the cosine or L2 distance by the point-to-point way without considering the context information. In this study, we make a key observation that the local con-text represented by the similarities between the instance and its inter-class neighbors1plays an important role forFR. Specifically, we attempt to incorporate the local in-formation in the feature space into the metric, and pro-pose a unified framework calledInter-class DiscrepancyAlignment(IDA), with two dedicated modules, Discrepancy Alignment Operator(IDA-DAO) andSupport Set Estimation(IDA-SSE). IDA-DAO is used to align the similarity scores considering the discrepancy between the images and its neighbors, which is defined by adaptive support sets on the hypersphere. For practical inference, it is difficult to acquire support set during online inference. IDA-SSE can provide convincing inter-class neighbors by introducing virtual candidate images generated with GAN. Further-more, we propose the learnable IDA-SSE, which can implicitly give estimation without the need of any other images in the evaluation process. The proposed IDA can be incorporated into existing FR systems seamlessly and efficiently. Extensive experiments demonstrate that this frame-work can 1) significantly improve the accuracy, and 2) make the model robust to the face images of various distributions.Without bells and whistles, our method achieves state-of-the-art performance on multiple standard FR benchmarks.

cs.CV↗

arXiv2019

Learning to Auto Weight: Entirely Data-driven and Highly Efficient Weighting Framework

Example weighting algorithm is an effective solution to the training bias problem, however, most previous typical methods are usually limited to human knowledge and require laborious tuning of hyperparameters. In this paper, we propose a novel example weighting framework called Learning to Auto Weight (LAW). The proposed framework finds step-dependent weighting policies adaptively, and can be jointly trained with target networks without any assumptions or prior knowledge about the dataset. It consists of three key components: Stage-based Searching Strategy (3SM) is adopted to shrink the huge searching space in a complete training process; Duplicate Network Reward (DNR) gives more accurate supervision by removing randomness during the searching process; Full Data Update (FDU) further improves the updating efficiency. Experimental results demonstrate the superiority of weighting policy explored by LAW over standard training pipeline. Compared with baselines, LAW can find a better weighting schedule which achieves much more superior accuracy on both biased CIFAR and ImageNet.

cs.LG↗

arXiv2019

A backward procedure for change-point detection with applications to copy number variation detection

Change-point detection regains much attention recently for analyzing array or sequencing data for copy number variation (CNV) detection. In such applications, the true signals are typically very short and buried in the long data sequence, which makes it challenging to identify the variations efficiently and accurately. In this article, we propose a new change-point detection method, a backward procedure, which is not only fast and simple enough to exploit high-dimensional data but also performs very well for detecting short signals. Although motivated by CNV detection, the backward procedure is generally applicable to assorted change-point problems that arise in a variety of scientific applications. It is illustrated by both simulated and real CNV data that the backward detection has clear advantages over other competing methods especially when the true signal is short.

stat.AP↗

arXiv2019

Knowledge Distillation via Route Constrained Optimization

Distillation-based learning boosts the performance of the miniaturized neural network based on the hypothesis that the representation of a teacher model can be used as structured and relatively weak supervision, and thus would be easily learned by a miniaturized model. However, we find that the representation of a converged heavy model is still a strong constraint for training a small student model, which leads to a high lower bound of congruence loss. In this work, inspired by curriculum learning we consider the knowledge distillation from the perspective of curriculum learning by routing. Instead of supervising the student model with a converged teacher model, we supervised it with some anchor points selected from the route in parameter space that the teacher model passed by, as we called route constrained optimization (RCO). We experimentally demonstrate this simple operation greatly reduces the lower bound of congruence loss for knowledge distillation, hint and mimicking learning. On close-set classification tasks like CIFAR100 and ImageNet, RCO improves knowledge distillation by 2.14% and 1.5% respectively. For the sake of evaluating the generalization, we also test RCO on the open-set face recognition task MegaFace.

cs.LG↗

arXiv2019

Dynamic Multi-path Neural Network

Although deeper and larger neural networks have achieved better performance, the complex network structure and increasing computational cost cannot meet the demands of many resource-constrained applications. Existing methods usually choose to execute or skip an entire specific layer, which can only alter the depth of the network. In this paper, we propose a novel method called Dynamic Multi-path Neural Network (DMNN), which provides more path selection choices in terms of network width and depth during inference. The inference path of the network is determined by a controller, which takes into account both previous state and object category information. The proposed method can be easily incorporated into most modern network architectures. Experimental results on ImageNet and CIFAR-100 demonstrate the superiority of our method on both efficiency and overall classification accuracy. To be specific, DMNN-101 significantly outperforms ResNet-101 with an encouraging 45.1% FLOPs reduction, and DMNN-50 performs comparably to ResNet-101 while saving 42.1% parameters.

cs.CV↗

arXiv2019

Correlation Congruence for Knowledge Distillation

Most teacher-student frameworks based on knowledge distillation (KD) depend on a strong congruent constraint on instance level. However, they usually ignore the correlation between multiple instances, which is also valuable for knowledge transfer. In this work, we propose a new framework named correlation congruence for knowledge distillation (CCKD), which transfers not only the instance-level information, but also the correlation between instances. Furthermore, a generalized kernel method based on Taylor series expansion is proposed to better capture the correlation between instances. Empirical experiments and ablation studies on image classification tasks (including CIFAR-100, ImageNet-1K) and metric learning tasks (including ReID and Face Recognition) show that the proposed CCKD substantially outperforms the original KD and achieves state-of-the-art accuracy compared with other SOTA KD-based methods. The CCKD can be easily deployed in the majority of the teacher-student framework such as KD and hint-based learning methods.

cs.CV↗

arXiv2017

Nonparametric Independence Screening via Favored Smoothing Bandwidth

We propose a flexible nonparametric regression method for ultrahigh-dimensional data. As a first step, we propose a fast screening method based on the favored smoothing bandwidth of the marginal local constant regression. Then, an iterative procedure is developed to recover both the important covariates and the regression function. Theoretically, we prove that the favored smoothing bandwidth based screening possesses the model selection consistency property. Simulation studies as well as real data analysis show the competitive performance of the new procedure.

stat.ME↗

arXiv2016

Local independence feature screening for nonparametric and semiparametric models by marginal empirical likelihood

We consider an independence feature screening technique for identifying explanatory variables that locally contribute to the response variable in high-dimensional regression analysis. Without requiring a specific parametric form of the underlying data model, our approach accommodates a wide spectrum of nonparametric and semiparametric model families. To detect the local contributions of explanatory variables, our approach constructs empirical likelihood locally in conjunction with marginal nonparametric regressions. Since our approach actually requires no estimation, it is advantageous in scenarios such as the single-index models where even specification and identification of a marginal model is an issue. By automatically incorporating the level of variation of the nonparametric regression and directly assessing the strength of data evidence supporting local contribution from each explanatory variable, our approach provides a unique perspective for solving feature screening problems. Theoretical analysis shows that our approach can handle data dimensionality growing exponentially with the sample size. With extensive theoretical illustrations and numerical examples, we show that the local independence screening approach performs promisingly.

math.ST↗

arXiv2014

IsoDOT Detects Differential RNA-isoform Expression/Usage with respect to a Categorical or Continuous Covariate with High Sensitivity and Specificity

We have developed a statistical method named IsoDOT to assess differential isoform expression (DIE) and differential isoform usage (DIU) using RNA-seq data. Here isoform usage refers to relative isoform expression given the total expression of the corresponding gene. IsoDOT performs two tasks that cannot be accomplished by existing methods: to test DIE/DIU with respect to a continuous covariate, and to test DIE/DIU for one case versus one control. The latter task is not an uncommon situation in practice, e.g., comparing paternal and maternal allele of one individual or comparing tumor and normal sample of one cancer patient. Simulation studies demonstrate the high sensitivity and specificity of IsoDOT. We apply IsoDOT to study the effects of haloperidol treatment on mouse transcriptome and identify a group of genes whose isoform usages respond to haloperidol treatment.

stat.AP↗

arXiv2013

Marginal empirical likelihood and sure independence feature screening

We study a marginal empirical likelihood approach in scenarios when the number of variables grows exponentially with the sample size. The marginal empirical likelihood ratios as functions of the parameters of interest are systematically examined, and we find that the marginal empirical likelihood ratio evaluated at zero can be used to differentiate whether an explanatory variable is contributing to a response variable or not. Based on this finding, we propose a unified feature screening procedure for linear models and the generalized linear models. Different from most existing feature screening approaches that rely on the magnitudes of some marginal estimators to identify true signals, the proposed screening approach is capable of further incorporating the level of uncertainties of such estimators. Such a merit inherits the self-studentization property of the empirical likelihood approach, and extends the insights of existing feature screening methods. Moreover, we show that our screening approach is less restrictive to distributional assumptions, and can be conveniently adapted to be applied in a broad range of scenarios such as models specified using general moment conditions. Our theoretical results and extensive numerical examples by simulations and data analysis demonstrate the merits of the marginal empirical likelihood approach.

math.ST↗

arXiv2013

Homogeneity in Regression

This paper explores the homogeneity of coefficients in high-dimensional regression, which extends the sparsity concept and is more general and suitable for many applications. Homogeneity arises when one expects regression coefficients corresponding to neighboring geographical regions or a similar cluster of covariates to be approximately the same. Sparsity corresponds to a special case of homogeneity with a known atom zero. In this article, we propose a new method called clustering algorithm in regression via data-driven segmentation (CARDS) to explore homogeneity. New mathematics are provided on the gain that can be achieved by exploring homogeneity. Statistical properties of two versions of CARDS are analyzed. In particular, the asymptotic normality of our proposed CARDS estimator is established, which reveals better estimation accuracy for homogeneous parameters than that without homogeneity exploration. When our methods are combined with sparsity exploration, further efficiency can be achieved beyond the exploration of sparsity alone. This provides additional insights into the power of exploring low-dimensional strucuture in high-dimensional regression: homogeneity and sparsity. The newly developed method is further illustrated by simulation studies and applications to real data.

stat.ME↗

arXiv2012

A Generic Path Algorithm for Regularized Statistical Estimation

Regularization is widely used in statistics and machine learning to prevent overfitting and gear solution towards prior information. In general, a regularized estimation problem minimizes the sum of a loss function and a penalty term. The penalty term is usually weighted by a tuning parameter and encourages certain constraints on the parameters to be estimated. Particular choices of constraints lead to the popular lasso, fused-lasso, and other generalized $l_1$ penalized regression methods. Although there has been a lot of research in this area, developing efficient optimization methods for many nonseparable penalties remains a challenge. In this article we propose an exact path solver based on ordinary differential equations (EPSODE) that works for any convex loss function and can deal with generalized $l_1$ penalties as well as more complicated regularization such as inequality constraints encountered in shape-restricted regressions and nonparametric density estimation. In the path following process, the solution path hits, exits, and slides along the various constraints and vividly illustrates the tradeoffs between goodness of fit and model parsimony. In practice, the EPSODE can be coupled with AIC, BIC, $C_p$ or cross-validation to select an optimal tuning parameter. Our applications to generalized $l_1$ regularized generalized linear models, shape-restricted regressions, Gaussian graphical models, and nonparametric density estimation showcase the potential of the EPSODE algorithm.

stat.CO↗

arXiv2011

Varying-coefficient functional linear regression

Functional linear regression analysis aims to model regression relations which include a functional predictor. The analog of the regression parameter vector or matrix in conventional multivariate or multiple-response linear regression models is a regression parameter function in one or two arguments. If, in addition, one has scalar predictors, as is often the case in applications to longitudinal studies, the question arises how to incorporate these into a functional regression model. We study a varying-coefficient approach where the scalar covariates are modeled as additional arguments of the regression parameter function. This extension of the functional linear regression model is analogous to the extension of conventional linear regression models to varying-coefficient models and shares its advantages, such as increased flexibility; however, the details of this extension are more challenging in the functional case. Our methodology combines smoothing methods with regularization by truncation at a finite number of functional principal components. A practical version is developed and is shown to perform better than functional linear regression for longitudinal data. We investigate the asymptotic properties of varying-coefficient functional linear regression and establish consistency properties.

math.ST↗

arXiv2010

High-dimensional variable selection for Cox's proportional hazards model

Variable selection in high dimensional space has challenged many contemporary statistical problems from many frontiers of scientific disciplines. Recent technology advance has made it possible to collect a huge amount of covariate information such as microarray, proteomic and SNP data via bioimaging technology while observing survival information on patients in clinical studies. Thus, the same challenge applies to the survival analysis in order to understand the association between genomics information and clinical information about the survival time. In this work, we extend the sure screening procedure Fan and Lv (2008) to Cox's proportional hazards model with an iterative version available. Numerical simulation studies have shown encouraging performance of the proposed method in comparison with other techniques such as LASSO. This demonstrates the utility and versatility of the iterative sure independent screening scheme.

stat.ML↗

arXiv2009

Local quasi-likelihood with a parametric guide

Generalized linear models and the quasi-likelihood method extend the ordinary regression models to accommodate more general conditional distributions of the response. Nonparametric methods need no explicit parametric specification, and the resulting model is completely determined by the data themselves. However, nonparametric estimation schemes generally have a slower convergence rate such as the local polynomial smoothing estimation of nonparametric generalized linear models studied in Fan, Heckman and Wand [J. Amer. Statist. Assoc. 90 (1995) 141--150]. In this work, we propose a unified family of parametrically-guided nonparametric estimation schemes. This combines the merits of both parametric and nonparametric approaches and enables us to incorporate prior knowledge. Asymptotic results and numerical simulations demonstrate the improvement of our new estimation schemes over the original nonparametric counterpart.

math.ST↗

arXiv2009

Network exploration via the adaptive LASSO and SCAD penalties

Graphical models are frequently used to explore networks, such as genetic networks, among a set of variables. This is usually carried out via exploring the sparsity of the precision matrix of the variables under consideration. Penalized likelihood methods are often used in such explorations. Yet, positive-definiteness constraints of precision matrices make the optimization problem challenging. We introduce nonconcave penalties and the adaptive LASSO penalty to attenuate the bias problem in the network estimation. Through the local linear approximation to the nonconcave penalty functions, the problem of precision matrix estimation is recast as a sequence of penalized likelihood problems with a weighted $L_1$ penalty and solved using the efficient algorithm of Friedman et al. [Biostatistics 9 (2008) 432--441]. Our estimation schemes are applied to two real datasets. Simulation experiments and asymptotic theory are used to justify our proposed methods.

stat.AP↗

arXiv2008

Ultrahigh dimensional variable selection: beyond the linear model

Variable selection in high-dimensional space characterizes many contemporary problems in scientific discovery and decision making. Many frequently-used techniques are based on independence screening; examples include correlation ranking (Fan and Lv, 2008) or feature selection using a two-sample t-test in high-dimensional classification (Tibshirani et al., 2003). Within the context of the linear model, Fan and Lv (2008)showed that this simple correlation ranking possesses a sure independence screening property under certain conditions and that its revision, called iteratively sure independent screening (ISIS), is needed when the features are marginally unrelated but jointly related to the response variable. In this paper, we extend ISIS, without explicit definition of residuals, to a general pseudo-likelihood framework, which includes generalized linear models as a special case. Even in the least-squares setting, the new method improves ISIS by allowing variable deletion in the iterative process. Our technique allows us to select important features in high-dimensional classification where the popularly used two-sample t-method fails. A new technique is introduced to reduce the false discovery rate in the feature screening stage. Several simulated and two real data examples are presented to illustrate the methodology.

stat.ME↗