Spectral computed tomography (CT) can reconstruct spectral images from different energy bins using photon counting detectors (PCDs). However, due to the limited photons and counting rate in the corresponding spectral fraction, the reconstructed spectral images usually suffer from severe noise. In this paper, a fourth-order nonlocal tensor decomposition model for spectral CT image reconstruction (FONT-SIR) method is proposed. Similar patches are collected in both spatial and spectral dimensions simultaneously to form the basic tensor unit. Additionally, principal component analysis (PCA) is applied to extract latent features from the patches for a robust and efficient similarity measure. Then, low-rank and sparsity decomposition is performed on the produced fourth-order tensor unit, and the weighted nuclear norm and total variation (TV) norm are used to enforce the low-rank and sparsity constraints, respectively. The alternating direction method of multipliers (ADMM) is adopted to optimize the objective function. The experimental results with our proposed FONT-SIR demonstrates a superior qualitative and quantitative performance for both simulated and real data sets relative to several state-of-the-art methods, in terms of noise suppression and detail preservation.
Over-parametrization is an important technique in training neural networks. In both theory and practice, training a larger network allows the optimization algorithm to avoid bad local optimal solutions. In this paper we study a closely related tensor decomposition problem: given an $l$-th order tensor in $(R^d)^{\otimes l}$ of rank $r$ (where $r\ll d$), can variants of gradient descent find a rank $m$ decomposition where $m > r$? We show that in a lazy training regime (similar to the NTK regime for neural networks) one needs at least $m = \Omega(d^{l-1})$, while a variant of gradient descent can find an approximate tensor when $m = O^*(r^{2.5l}\log d)$. Our results show that gradient descent on over-parametrized objective could go beyond the lazy training regime and utilize certain low-rank structure in the data.
While post-training model compression can greatly reduce the inference cost of a deep neural network, uncompressed training still consumes a huge amount of hardware resources, run-time and energy. It is highly desirable to directly train a compact neural network from scratch with low memory and low computational cost. Low-rank tensor decomposition is one of the most effective approaches to reduce the memory and computing requirements of large-size neural networks. However, directly training a low-rank tensorized neural network is a very challenging task because it is hard to determine a proper tensor rank {\it a priori}, which controls the model complexity and compression ratio in the training process. This paper presents a novel end-to-end framework for low-rank tensorized training of neural networks. We first develop a flexible Bayesian model that can handle various low-rank tensor formats (e.g., CP, Tucker, tensor train and tensor-train matrix) that compress neural network parameters in training. This model can automatically determine the tensor ranks inside a nonlinear forward model, which is beyond the capability of existing Bayesian tensor methods. We further develop a scalable stochastic variational inference solver to estimate the posterior density of large-scale problems in training. Our work provides the first general-purpose rank-adaptive framework for end-to-end tensorized training. Our numerical results on various neural network architectures show orders-of-magnitude parameter reduction and little accuracy loss (or even better accuracy) in the training process. Specifically, on a very large deep learning recommendation system with over $4.2\times 10^9$ model parameters, our method can reduce the variables to only $1.6\times 10^5$ automatically in the training process (i.e., by $2.6\times 10^4$ times) while achieving almost the same accuracy.
Extracting entity pairs along with relation types from unstructured texts is a fundamental subtask of information extraction. Most existing joint models rely on fine-grained labeling scheme or focus on shared embedding parameters. These methods directly model the joint probability of multi-labeled triplets, which suffer from extracting redundant triplets with all relation types. However, each sentence may contain very few relation types. In this paper, we first model the final triplet extraction result as a three-order tensor of word-to-word pairs enriched with each relation type. And in order to obtain the sentence contained relations, we introduce an independent but joint training relation classification module. The tensor decomposition strategy is finally utilized to decompose the triplet tensor with predicted relational components which omits the calculations for unpredicted relation types. According to effective decomposition methods, we propose the Tensor Decomposition based Relation Extraction (TDRE) approach which is able to extract overlapping triplets and avoid detecting unnecessary entity pairs. Experiments on benchmark datasets NYT, CoNLL04 and ADE datasets demonstrate that the proposed method outperforms existing strong baselines.
Temporal data (such as news articles or Twitter feeds) often consists of a mixture of long-lasting trends and popular but short-lasting topics of interest. A truly successful topic modeling strategy should be able to detect both types of topics and clearly locate them in time. In this paper, we first show that nonnegative CANDECOMP/PARAFAC decomposition (NCPD) is able to discover topics of variable persistence automatically. Then, we propose sparseness-constrained NCPD (S-NCPD) and its online variant in order to actively control the length of the learned topics effectively and efficiently. Further, we propose quantitative ways to measure the topic length and demonstrate the ability of S-NCPD (as well as its online variant) to discover short and long-lasting temporal topics in a controlled manner in semi-synthetic and real-world data including news headlines. We also demonstrate that the online variant of S-NCPD reduces the reconstruction error more rapidly than S-NCPD.
Effective non-parametric density estimation is a key challenge in high-dimensional multivariate data analysis. In this paper,we propose a novel approach that builds upon tensor factorization tools. Any multivariate density can be represented by its characteristic function, via the Fourier transform. If the sought density is compactly supported, then its characteristic function can be approximated, within controllable error, by a finite tensor of leading Fourier coefficients, whose size de-pends on the smoothness of the underlying density. This tensor can be naturally estimated from observed realizations of the random vector of interest, via sample averaging. In order to circumvent the curse of dimensionality, we introduce a low-rank model of this characteristic tensor, which significantly improves the density estimate especially for high-dimensional data and/or in the sample-starved regime. By virtue of uniqueness of low-rank tensor decomposition, under certain conditions, our method enables learning the true data-generating distribution. We demonstrate the very promising performance of the proposed method using several measured datasets.
Recurrent neural networks (RNNs) are powerful in the tasks oriented to sequential data, such as natural language processing and video recognition. However, since the modern RNNs, including long-short term memory (LSTM) and gated recurrent unit (GRU) networks, have complex topologies and expensive space/computation complexity, compressing them becomes a hot and promising topic in recent years. Among plenty of compression methods, tensor decomposition, e.g., tensor train (TT), block term (BT), tensor ring (TR) and hierarchical Tucker (HT), appears to be the most amazing approach since a very high compression ratio might be obtained. Nevertheless, none of these tensor decomposition formats can provide both the space and computation efficiency. In this paper, we consider to compress RNNs based on a novel Kronecker CANDECOMP/PARAFAC (KCP) decomposition, which is derived from Kronecker tensor (KT) decomposition, by proposing two fast algorithms of multiplication between the input and the tensor-decomposed weight. According to our experiments based on UCF11, Youtube Celebrities Face and UCF50 datasets, it can be verified that the proposed KCP-RNNs have comparable performance of accuracy with those in other tensor-decomposed formats, and even 278,219x compression ratio could be obtained by the low rank KCP. More importantly, KCP-RNNs are efficient in both space and computation complexity compared with other tensor-decomposed ones under similar ranks. Besides, we find KCP has the best potential for parallel computing to accelerate the calculations in neural networks.
Node embeddings have been attracting increasing attention during the past years. In this context, we propose a new ensemble node embedding approach, called TenSemble2Vec, by first generating multiple embeddings using the existing techniques and taking them as multiview data input of the state-of-art tensor decomposition model namely PARAFAC2 to learn the shared lower-dimensional representations of the nodes. Contrary to other embedding methods, our TenSemble2Vec takes advantage of the complementary information from different methods or the same method with different hyper-parameters, which bypasses the challenge of choosing models. Extensive tests using real-world data validates the efficiency of the proposed method.
Most state of the art deep neural networks are overparameterized and exhibit a high computational cost. A straightforward approach to this problem is to replace convolutional kernels with its low-rank tensor approximations, whereas the Canonical Polyadic tensor Decomposition is one of the most suited models. However, fitting the convolutional tensors by numerical optimization algorithms often encounters diverging components, i.e., extremely large rank-one tensors but canceling each other. Such degeneracy often causes the non-interpretable result and numerical instability for the neural network fine-tuning. This paper is the first study on degeneracy in the tensor decomposition of convolutional kernels. We present a novel method, which can stabilize the low-rank approximation of convolutional kernels and ensure efficient compression while preserving the high-quality performance of the neural networks. We evaluate our approach on popular CNN architectures for image classification and show that our method results in much lower accuracy degradation and provides consistent performance.
Tensor Networks (TN) offer a powerful framework to efficiently represent very high-dimensional objects. TN have recently shown their potential for machine learning applications and offer a unifying view of common tensor decomposition models such as Tucker, tensor train (TT) and tensor ring (TR). However, identifying the best tensor network structure from data for a given task is challenging. In this work, we leverage the TN formalism to develop a generic and efficient adaptive algorithm to jointly learn the structure and the parameters of a TN from data. Our method is based on a simple greedy approach starting from a rank one tensor and successively identifying the most promising tensor network edges for small rank increments. Our algorithm can adaptively identify TN structures with small number of parameters that effectively optimize any differentiable objective function. Experiments on tensor decomposition, tensor completion and model compression tasks demonstrate the effectiveness of the proposed algorithm. In particular, our method outperforms the state-of-the-art evolutionary topology search [Li and Sun, 2020] for tensor decomposition of images (while being orders of magnitude faster) and finds efficient tensor network structures to compress neural networks outperforming popular TT based approaches [Novikov et al., 2015].
In recent years, the fields of natural language processing (NLP) and information retrieval (IR) have made tremendous progress thanksto deep learning models like Recurrent Neural Networks (RNNs), Gated Recurrent Units (GRUs) and Long Short-Term Memory (LSTMs)networks, and Transformer [120] based models like Bidirectional Encoder Representations from Transformers (BERT) [24], GenerativePre-training Transformer (GPT-2) [94], Multi-task Deep Neural Network (MT-DNN) [73], Extra-Long Network (XLNet) [134], Text-to-text transfer transformer (T5) [95], T-NLG [98] and GShard [63]. But these models are humongous in size. On the other hand,real world applications demand small model size, low response times and low computational power wattage. In this survey, wediscuss six different types of methods (Pruning, Quantization, Knowledge Distillation, Parameter Sharing, Tensor Decomposition, andSub-quadratic Transformer based methods) for compression of such models to enable their deployment in real industry NLP projects.Given the critical need of building applications with efficient and small models, and the large amount of recently published work inthis area, we believe that this survey organizes the plethora of work done by the 'deep learning for NLP' community in the past fewyears and presents it as a coherent story.
In this work, we tackle model efficiency by exploiting redundancy in the \textit{implicit structure} of the building blocks of convolutional neural networks. We start our analysis by introducing a general definition of Composite Kernel structures that enable the execution of convolution operations in the form of efficient, scaled, sum-pooling components. As its special case, we propose \textit{Structured Convolutions} and show that these allow decomposition of the convolution operation into a sum-pooling operation followed by a convolution with significantly lower complexity and fewer weights. We show how this decomposition can be applied to 2D and 3D kernels as well as the fully-connected layers. Furthermore, we present a Structural Regularization loss that promotes neural network layers to leverage on this desired structure in a way that, after training, they can be decomposed with negligible performance loss. By applying our method to a wide range of CNN architectures, we demonstrate "structured" versions of the ResNets that are up to 2$\times$ smaller and a new Structured-MobileNetV2 that is more efficient while staying within an accuracy loss of 1% on ImageNet and CIFAR-10 datasets. We also show similar structured versions of EfficientNet on ImageNet and HRNet architecture for semantic segmentation on the Cityscapes dataset. Our method performs equally well or superior in terms of the complexity reduction in comparison to the existing tensor decomposition and channel pruning methods.
Tensor, also known as multi-dimensional array, arises from many applications in signal processing, manufacturing processes, healthcare, among others. As one of the most popular methods in tensor literature, Robust tensor principal component analysis (RTPCA) is a very effective tool to extract the low rank and sparse components in tensors. In this paper, a new method to analyze RTPCA is proposed based on the recently developed tensor-tensor product and tensor singular value decomposition (t-SVD). Specifically, it aims to solve a convex optimization problem whose objective function is a weighted combination of the tensor nuclear norm and the l1-norm. In most of literature of RTPCA, the exact recovery is built on the tensor incoherence conditions and the assumption of a uniform model on the sparse support. Unlike this conventional way, in this paper, without any assumption of randomness, the exact recovery can be achieved in a completely deterministic fashion by characterizing the tensor rank-sparsity incoherence, which is an uncertainty principle between the low-rank tensor spaces and the pattern of sparse tensor.
Boolean tensor has been broadly utilized in representing high dimensional logical data collected on spatial, temporal and/or other relational domains. Boolean Tensor Decomposition (BTD) factorizes a binary tensor into the Boolean sum of multiple rank-1 tensors, which is an NP-hard problem. Existing BTD methods have been limited by their high computational cost, in applications to large scale or higher order tensors. In this work, we presented a computationally efficient BTD algorithm, namely \textit{Geometric Expansion for all-order Tensor Factorization} (GETF), that sequentially identifies the rank-1 basis components for a tensor from a geometric perspective. We conducted rigorous theoretical analysis on the validity as well as algorithemic efficiency of GETF in decomposing all-order tensor. Experiments on both synthetic and real-world data demonstrated that GETF has significantly improved performance in reconstruction accuracy, extraction of latent structures and it is an order of magnitude faster than other state-of-the-art methods.
This chapter studies the problem of decomposing a tensor into a sum of constituent rank one tensors. While tensor decompositions are very useful in designing learning algorithms and data analysis, they are NP-hard in the worst-case. We will see how to design efficient algorithms with provable guarantees under mild assumptions, and using beyond worst-case frameworks like smoothed analysis.
Tensor data with rich structural information becomes increasingly important in process modeling, monitoring, and diagnosis. Here structural information is referred to structural properties such as sparsity, smoothness, low-rank, and piecewise constancy. To reveal useful information from tensor data, we propose to decompose the tensor into the summation of multiple components based on different structural information of them. In this paper, we provide a new definition of structural information in tensor data. Based on it, we propose an additive tensor decomposition (ATD) framework to extract useful information from tensor data. This framework specifies a high dimensional optimization problem to obtain the components with distinct structural information. An alternating direction method of multipliers (ADMM) algorithm is proposed to solve it, which is highly parallelable and thus suitable for the proposed optimization problem. Two simulation examples and a real case study in medical image analysis illustrate the versatility and effectiveness of the ATD framework.
We propose a new algorithm for tensor decomposition, based on Jennrich's algorithm, and apply our new algorithmic ideas to blind deconvolution and Gaussian mixture models. Our first contribution is a simple and efficient algorithm to decompose certain symmetric overcomplete order-3 tensors, that is, three dimensional arrays of the form $T = \sum_{i=1}^n a_i \otimes a_i \otimes a_i$ where the $a_i$s are not linearly independent.Our algorithm comes with a detailed robustness analysis. Our second contribution builds on top of our tensor decomposition algorithm to expand the family of Gaussian mixture models whose parameters can be estimated efficiently. These ideas are also presented in a more general framework of blind deconvolution that makes them applicable to mixture models of identical but very general distributions, including all centrally symmetric distributions with finite 6th moment.
With the rapid development of knowledge bases (KBs), link prediction task, which completes KBs with missing facts, has been broadly studied in especially binary relational KBs (a.k.a knowledge graph) with powerful tensor decomposition related methods. However, the ubiquitous n-ary relational KBs with higher-arity relational facts are paid less attention, in which existing translation based and neural network based approaches have weak expressiveness and high complexity in modeling various relations. Tensor decomposition has not been considered for n-ary relational KBs, while directly extending tensor decomposition related methods of binary relational KBs to the n-ary case does not yield satisfactory results due to exponential model complexity and their strong assumptions on binary relations. To generalize tensor decomposition for n-ary relational KBs, in this work, we propose GETD, a generalized model based on Tucker decomposition and Tensor Ring decomposition. The existing negative sampling technique is also generalized to the n-ary case for GETD. In addition, we theoretically prove that GETD is fully expressive to completely represent any KBs. Extensive evaluations on two representative n-ary relational KB datasets demonstrate the superior performance of GETD, significantly improving the state-of-the-art methods by over 15\%. Moreover, GETD further obtains the state-of-the-art results on the benchmark binary relational KB datasets.
We consider nonlinear convergence acceleration methods for fixed-point iteration $x_{k+1}=q(x_k)$, including Anderson acceleration (AA), nonlinear GMRES (NGMRES), and Nesterov-type acceleration (corresponding to AA with window size one). We focus on fixed-point methods that converge asymptotically linearly with convergence factor $\rho<1$ and that solve an underlying fully smooth and non-convex optimization problem. It is often observed that AA and NGMRES substantially improve the asymptotic convergence behavior of the fixed-point iteration, but this improvement has not been quantified theoretically. We investigate this problem under simplified conditions. First, we consider stationary versions of AA and NGMRES, and determine coefficients that result in optimal asymptotic convergence factors, given knowledge of the spectrum of $q'(x)$ at the fixed point $x^*$. This allows us to understand and quantify the asymptotic convergence improvement that can be provided by nonlinear convergence acceleration, viewing $x_{k+1}=q(x_k)$ as a nonlinear preconditioner for AA and NGMRES. Second, for the case of infinite window size, we consider linear asymptotic convergence bounds for GMRES applied to the fixed-point iteration linearized about $x^*$. Since AA and NGMRES are equivalent to GMRES in the linear case, one may expect the GMRES convergence factors to be relevant for AA and NGMRES as $x_k \rightarrow x^*$. Our results are illustrated numerically for a class of test problems from canonical tensor decomposition, comparing steepest descent and alternating least squares (ALS) as the fixed-point iterations that are accelerated by AA and NGMRES. Our numerical tests show that both approaches allow us to estimate asymptotic convergence speed for nonstationary AA and NGMRES with finite window size.
Hyperspectral image (HSI) has some advantages over natural image for various applications due to the extra spectral information. During the acquisition, it is often contaminated by severe noises including Gaussian noise, impulse noise, deadlines, and stripes. The image quality degeneration would badly effect some applications. In this paper, we present a HSI restoration method named smooth and robust low rank tensor recovery. Specifically, we propose a structural tensor decomposition in accordance with the linear spectral mixture model of HSI. It decomposes a tensor into sums of outer matrix vector products, where the vectors are orthogonal due to the independence of endmember spectrums. Based on it, the global low rank tensor structure can be well exposited for HSI denoising. In addition, the 3D anisotropic total variation is used for spatial spectral piecewise smoothness of HSI. Meanwhile, the sparse noise including impulse noise, deadlines and stripes, is detected by the l1 norm regularization. The Frobenius norm is used for the heavy Gaussian noise in some real world scenarios. The alternating direction method of multipliers is adopted to solve the proposed optimization model, which simultaneously exploits the global low rank property and the spatial spectral smoothness of the HSI. Numerical experiments on both simulated and real data illustrate the superiority of the proposed method in comparison with the existing ones.