Dimensionality reduction techniques aim at representing high-dimensional data in low-dimensional spaces to extract hidden and useful information or facilitate visual understanding and interpretation of the data. However, few of them take into consideration the potential cluster information contained implicitly in the high-dimensional data. In this paper, we propose LaptSNE, a new graph-layout nonlinear dimensionality reduction method based on t-SNE, one of the best techniques for visualizing high-dimensional data as 2D scatter plots. Specifically, LaptSNE leverages the eigenvalue information of the graph Laplacian to shrink the potential clusters in the low-dimensional embedding when learning to preserve the local and global structure from high-dimensional space to low-dimensional space. It is nontrivial to solve the proposed model because the eigenvalues of normalized symmetric Laplacian are functions of the decision variable. We provide a majorization-minimization algorithm with convergence guarantee to solve the optimization problem of LaptSNE and show how to calculate the gradient analytically, which may be of broad interest when considering optimization with Laplacian-composited objective. We evaluate our method by a formal comparison with state-of-the-art methods, both visually and via established quantitative measurements. The results demonstrate the superiority of our method over baselines such as t-SNE and UMAP. We also extend our method to spectral clustering and establish an accurate and parameter-free clustering algorithm, which provides us high reliability and convenience in real applications.
This work presents an unsupervised deep discriminant analysis for clustering. The method is based on deep neural networks and aims to minimize the intra-cluster discrepancy and maximize the inter-cluster discrepancy in an unsupervised manner. The method is able to project the data into a nonlinear low-dimensional latent space with compact and distinct distribution patterns such that the data clusters can be effectively identified. We further provide an extension of the method such that available graph information can be effectively exploited to improve the clustering performance. Extensive numerical results on image and non-image data with or without graph information demonstrate the effectiveness of the proposed methods.
This paper presents a simple yet effective method for anomaly detection. The main idea is to learn small perturbations to perturb normal data and learn a classifier to classify the normal data and the perturbed data into two different classes. The perturbator and classifier are jointly learned using deep neural networks. Importantly, the perturbations should be as small as possible but the classifier is still able to recognize the perturbed data from unperturbed data. Therefore, the perturbed data are regarded as abnormal data and the classifier provides a decision boundary between the normal data and abnormal data, although the training data do not include any abnormal data. Compared with the state-of-the-art of anomaly detection, our method does not require any assumption about the shape (e.g. hypersphere) of the decision boundary and has fewer hyper-parameters to determine. Empirical studies on benchmark datasets verify the effectiveness and superiority of our method.
Network quantization has emerged as a promising method for model compression and inference acceleration. However, tradtional quantization methods (such as quantization aware training and post training quantization) require original data for the fine-tuning or calibration of quantized model, which makes them inapplicable to the cases that original data are not accessed due to privacy or security. This gives birth to the data-free quantization with synthetic data generation. While current DFQ methods still suffer from severe performance degradation when quantizing a model into lower bit, caused by the low inter-class separability of semantic features. To this end, we propose a new and effective data-free quantization method termed ClusterQ, which utilizes the semantic feature distribution alignment for synthetic data generation. To obtain high inter-class separability of semantic features, we cluster and align the feature distribution statistics to imitate the distribution of real data, so that the performance degradation is alleviated. Moreover, we incorporate the intra-class variance to solve class-wise mode collapse. We also employ the exponential moving average to update the centroid of each cluster for further feature distribution improvement. Extensive experiments across various deep models (e.g., ResNet-18 and MobileNet-V2) over the ImageNet dataset demonstrate that our ClusterQ obtains state-of-the-art performance.
The performance of spectral clustering heavily relies on the quality of affinity matrix. A variety of affinity-matrix-construction methods have been proposed but they have hyper-parameters to determine beforehand, which requires strong experience and lead to difficulty in real applications especially when the inter-cluster similarity is high or/and the dataset is large. On the other hand, we often have to determine to use a linear model or a nonlinear model, which still depends on experience. To solve these two problems, in this paper, we present an eigen-gap guided search method for subspace clustering. The main idea is to find the most reliable affinity matrix among a set of candidates constructed by linear and kernel regressions, where the reliability is quantified by the \textit{relative-eigen-gap} of graph Laplacian defined in this paper. We show, theoretically and numerically, that the Laplacian matrix with a larger relative-eigen-gap often yields a higher clustering accuracy and stability. Our method is able to automatically search the best model and hyper-parameters in a pre-defined space. The search space is very easy to determine and can be arbitrarily large, though a relatively compact search space can reduce the highly unnecessary computation. Our method has high flexibility and convenience in real applications, and also has low computational cost because the affinity matrix is not computed by iterative optimization. We extend the method to large-scale datasets such as MNIST, on which the time cost is less than 90s and the clustering accuracy is state-of-the-art. Extensive experiments of natural image clustering show that our method is more stable, accurate, and efficient than baseline methods.
Subspace clustering (SC) aims to cluster data lying in a union of low-dimensional subspaces. Usually, SC learns an affinity matrix and then performs spectral clustering. Both steps suffer from high time and space complexity, which leads to difficulty in clustering large datasets. This paper presents a method called k-Factorization Subspace Clustering (k-FSC) for large-scale subspace clustering. K-FSC directly factorizes the data into k groups via pursuing structured sparsity in the matrix factorization model. Thus, k-FSC avoids learning affinity matrix and performing eigenvalue decomposition, and hence has low time and space complexity on large datasets. An efficient algorithm is proposed to solve the optimization of k-FSC. In addition, k-FSC is able to handle noise, outliers, and missing data and applicable to arbitrarily large datasets and streaming data. Extensive experiments show that k-FSC outperforms state-of-the-art subspace clustering methods.
The nuclear norm and Schatten-$p$ quasi-norm of a matrix are popular rank proxies in low-rank matrix recovery. Unfortunately, computing the nuclear norm or Schatten-$p$ quasi-norm of a tensor is NP-hard, which is a pity for low-rank tensor completion (LRTC) and tensor robust principal component analysis (TRPCA). In this paper, we propose a new class of rank regularizers based on the Euclidean norms of the CP component vectors of a tensor and show that these regularizers are monotonic transformations of tensor Schatten-$p$ quasi-norm. This connection enables us to minimize the Schatten-$p$ quasi-norm in LRTC and TRPCA implicitly. The methods do not use the singular value decomposition and hence scale to big tensors. Moreover, the methods are not sensitive to the choice of initial rank and provide an arbitrarily sharper rank proxy for low-rank tensor recovery compared to nuclear norm. We provide theoretical guarantees in terms of recovery error for LRTC and TRPCA, which show relatively smaller $p$ of Schatten-$p$ quasi-norm leads to tighter error bounds. Experiments using LRTC and TRPCA on synthetic data and natural images verify the effectiveness and superiority of our methods compared to baseline methods.
This paper proposes a new variant of Frank-Wolfe (FW), called $k$FW. Standard FW suffers from slow convergence: iterates often zig-zag as update directions oscillate around extreme points of the constraint set. The new variant, $k$FW, overcomes this problem by using two stronger subproblem oracles in each iteration. The first is a $k$ linear optimization oracle ($k$LOO) that computes the $k$ best update directions (rather than just one). The second is a $k$ direction search ($k$DS) that minimizes the objective over a constraint set represented by the $k$ best update directions and the previous iterate. When the problem solution admits a sparse representation, both oracles are easy to compute, and $k$FW converges quickly for smooth convex objectives and several interesting constraint sets: $k$FW achieves finite $\frac{4L_f^3D^4}{\gamma\delta^2}$ convergence on polytopes and group norm balls, and linear convergence on spectrahedra and nuclear norm balls. Numerical experiments validate the effectiveness of $k$FW and demonstrate an order-of-magnitude speedup over existing approaches.
Data scientists seeking a good supervised learning model on a new dataset have many choices to make: they must preprocess the data, select features, possibly reduce the dimension, select an estimation algorithm, and choose hyperparameters for each of these pipeline components. With new pipeline components comes a combinatorial explosion in the number of choices! In this work, we design a new AutoML system to address this challenge: an automated system to design a supervised learning pipeline. Our system uses matrix and tensor factorization as surrogate models to model the combinatorial pipeline search space. Under these models, we develop greedy experiment design protocols to efficiently gather information about a new dataset. Experiments on large corpora of real-world classification problems demonstrate the effectiveness of our approach.
Low dimensional nonlinear structure abounds in datasets across computer vision and machine learning. Kernelized matrix factorization techniques have recently been proposed to learn these nonlinear structures from partially observed data, with impressive empirical performance, by observing that the image of the matrix in a sufficiently large feature space is low-rank. However, these nonlinear methods fail in the presence of noise or outliers. In this work, we propose a new robust nonlinear factorization method called Robust Non-Linear Matrix Factorization (RNLMF). RNLMF constructs a dictionary for the data space by factoring a kernelized feature space; a noisy matrix can then be decomposed as the sum of a sparse noise matrix and a clean data matrix that lies in a low dimensional nonlinear manifold. RNLMF is robust to noise and outliers and scales to matrices with thousands of rows and columns. Empirically, RNLMF achieves noticeable improvements over baseline methods in denoising and clustering.