Numerous Graph Neural Networks (GNNs) have been developed to tackle the challenge of Knowledge Graph Embedding (KGE). However, many of these approaches overlook the crucial role of relation information and inadequately integrate it with entity information, resulting in diminished expressive power. In this paper, we propose a novel knowledge graph encoder that incorporates tensor decomposition within the aggregation function of Relational Graph Convolutional Network (R-GCN). Our model enhances the representation of neighboring entities by employing projection matrices of a low-rank tensor defined by relation types. This approach facilitates multi-task learning, thereby generating relation-aware representations. Furthermore, we introduce a low-rank estimation technique for the core tensor through CP decomposition, which effectively compresses and regularizes our model. We adopt a training strategy inspired by contrastive learning, which relieves the training limitation of the 1-N method inherent in handling vast graphs. We outperformed all our competitors on two common benchmark datasets, FB15k-237 and WN18RR, while using low-dimensional embeddings for entities and relations.
Few-shot domain adaptation to multiple domains aims to learn a complex image distribution across multiple domains from a few training images. A na\"ive solution here is to train a separate model for each domain using few-shot domain adaptation methods. Unfortunately, this approach mandates linearly-scaled computational resources both in memory and computation time and, more importantly, such separate models cannot exploit the shared knowledge between target domains. In this paper, we propose DynaGAN, a novel few-shot domain-adaptation method for multiple target domains. DynaGAN has an adaptation module, which is a hyper-network that dynamically adapts a pretrained GAN model into the multiple target domains. Hence, we can fully exploit the shared knowledge across target domains and avoid the linearly-scaled computational requirements. As it is still computationally challenging to adapt a large-size GAN model, we design our adaptation module light-weight using the rank-1 tensor decomposition. Lastly, we propose a contrastive-adaptation loss suitable for multi-domain few-shot adaptation. We validate the effectiveness of our method through extensive qualitative and quantitative evaluations.
Identifying abnormal patterns in electroencephalography (EEG) remains the cornerstone of diagnosing several neurological diseases. The current clinical EEG review process relies heavily on expert visual review, which is unscalable and error-prone. In an effort to augment the expert review process, there is a significant interest in mining population-level EEG patterns using unsupervised approaches. Current approaches rely either on two-dimensional decompositions (e.g., principal and independent component analyses) or deep representation learning (e.g., auto-encoders, self-supervision). However, most approaches do not leverage the natural multi-dimensional structure of EEGs and lack interpretability. In this study, we propose a tensor decomposition approach using the canonical polyadic decomposition to discover a parsimonious set of population-level EEG patterns, retaining the natural multi-dimensional structure of EEGs (time x space x frequency). We then validate their clinical value using a cohort of patients including varying stages of cognitive impairment. Our results show that the discovered patterns reflect physiologically meaningful features and accurately classify the stages of cognitive impairment (healthy vs mild cognitive impairment vs Alzheimer's dementia) with substantially fewer features compared to classical and deep learning-based baselines. We conclude that the decomposition of population-level EEG tensors recovers expert-interpretable EEG patterns that can aid in the study of smaller specialized clinical cohorts.
We present Tensor4D, an efficient yet effective approach to dynamic scene modeling. The key of our solution is an efficient 4D tensor decomposition method so that the dynamic scene can be directly represented as a 4D spatio-temporal tensor. To tackle the accompanying memory issue, we decompose the 4D tensor hierarchically by projecting it first into three time-aware volumes and then nine compact feature planes. In this way, spatial information over time can be simultaneously captured in a compact and memory-efficient manner. When applying Tensor4D for dynamic scene reconstruction and rendering, we further factorize the 4D fields to different scales in the sense that structural motions and dynamic detailed changes can be learned from coarse to fine. The effectiveness of our method is validated on both synthetic and real-world scenes. Extensive experiments show that our method is able to achieve high-quality dynamic reconstruction and rendering from sparse-view camera rigs or even a monocular camera. The code and dataset will be released at https://liuyebin.com/tensor4d/tensor4d.html.
We present an alternating least squares type numerical optimization scheme to estimate conditionally-independent mixture models in $\mathbb{R}^n$, without parameterizing the distributions. Following the method of moments, we tackle an incomplete tensor decomposition problem to learn the mixing weights and componentwise means. Then we compute the cumulative distribution functions, higher moments and other statistics of the component distributions through linear solves. Crucially for computations in high dimensions, the steep costs associated with high-order tensors are evaded, via the development of efficient tensor-free operations. Numerical experiments demonstrate the competitive performance of the algorithm, and its applicability to many models and applications. Furthermore we provide theoretical analyses, establishing identifiability from low-order moments of the mixture and guaranteeing local linear convergence of the ALS algorithm.
Performance tuning, software/hardware co-design, and job scheduling are among the many tasks that rely on models to predict application performance. We propose and evaluate low-rank tensor decomposition for modeling application performance. We discretize the input and configuration domains of an application using regular grids. Application execution times mapped within grid-cells are averaged and represented by tensor elements. We show that low-rank canonical-polyadic (CP) tensor decomposition is effective in approximating these tensors. We further show that this decomposition enables accurate extrapolation of unobserved regions of an application's parameter space. We then employ tensor completion to optimize a CP decomposition given a sparse set of observed execution times. We consider alternative piecewise/grid-based models and supervised learning models for six applications and demonstrate that CP decomposition optimized using tensor completion offers higher prediction accuracy and memory-efficiency for high-dimensional performance modeling.
We show how to develop sampling-based alternating least squares (ALS) algorithms for decomposition of tensors into any tensor network (TN) format. Provided the TN format satisfies certain mild assumptions, resulting algorithms will have input sublinear per-iteration cost. Unlike most previous works on sampling-based ALS methods for tensor decomposition, the sampling in our framework is done according to the exact leverage score distribution of the design matrices in the ALS subproblems. We implement and test two tensor decomposition algorithms that use our sampling framework in a feature extraction experiment where we compare them against a number of other decomposition algorithms.
This work analyzes the solution trajectory of gradient-based algorithms via a novel basis function decomposition. We show that, although solution trajectories of gradient-based algorithms may vary depending on the learning task, they behave almost monotonically when projected onto an appropriate orthonormal function basis. Such projection gives rise to a basis function decomposition of the solution trajectory. Theoretically, we use our proposed basis function decomposition to establish the convergence of gradient descent (GD) on several representative learning tasks. In particular, we improve the convergence of GD on symmetric matrix factorization and provide a completely new convergence result for the orthogonal symmetric tensor decomposition. Empirically, we illustrate the promise of our proposed framework on realistic deep neural networks (DNNs) across different architectures, gradient-based solvers, and datasets. Our key finding is that gradient-based algorithms monotonically learn the coefficients of a particular orthonormal function basis of DNNs defined as the eigenvectors of the conjugate kernel after training. Our code is available at https://github.com/jianhaoma/function-basis-decomposition.
Knowledge graph embedding aims to predict the missing relations between entities in knowledge graphs. Tensor-decomposition-based models, such as ComplEx, provide a good trade-off between efficiency and expressiveness, that is crucial because of the large size of real world knowledge graphs. The recent multi-partition embedding interaction (MEI) model subsumes these models by using the block term tensor format and provides a systematic solution for the trade-off. However, MEI has several drawbacks, some of which carried from its subsumed tensor-decomposition-based models. In this paper, we address these drawbacks and introduce the Multi-partition Embedding Interaction iMproved beyond block term format (MEIM) model, with independent core tensor for ensemble effects and soft orthogonality for max-rank mapping, in addition to multi-partition embedding. MEIM improves expressiveness while still being highly efficient, helping it to outperform strong baselines and achieve state-of-the-art results on difficult link prediction benchmarks using fairly small embedding sizes. The source code is released at https://github.com/tranhungnghiep/MEIM-KGE.
Last years have seen a regain of interest for the use of stochastic block modeling (SBM) in recommender systems. These models are seen as a flexible alternative to tensor decomposition techniques that are able to handle labeled data. Recent works proposed to tackle discrete recommendation problems via SBMs by considering larger contexts as input data and by adding second order interactions between contexts' related elements. In this work, we show that these models are all special cases of a single global framework: the Serialized Interacting Mixed membership Stochastic Block Model (SIMSBM). It allows to model an arbitrarily large context as well as an arbitrarily high order of interactions. We demonstrate that SIMSBM generalizes several recent SBM-based baselines. Besides, we demonstrate that our formulation allows for an increased predictive power on six real-world datasets.
Kronecker regression is a highly-structured least squares problem $\min_{\mathbf{x}} \lVert \mathbf{K}\mathbf{x} - \mathbf{b} \rVert_{2}^2$, where the design matrix $\mathbf{K} = \mathbf{A}^{(1)} \otimes \cdots \otimes \mathbf{A}^{(N)}$ is a Kronecker product of factor matrices. This regression problem arises in each step of the widely-used alternating least squares (ALS) algorithm for computing the Tucker decomposition of a tensor. We present the first subquadratic-time algorithm for solving Kronecker regression to a $(1+\varepsilon)$-approximation that avoids the exponential term $O(\varepsilon^{-N})$ in the running time. Our techniques combine leverage score sampling and iterative methods. By extending our approach to block-design matrices where one block is a Kronecker product, we also achieve subquadratic-time algorithms for (1) Kronecker ridge regression and (2) updating the factor matrices of a Tucker decomposition in ALS, which is not a pure Kronecker regression problem, thereby improving the running time of all steps of Tucker ALS. We demonstrate the speed and accuracy of this Kronecker regression algorithm on synthetic data and real-world image tensors.
Spatiotemporal activity prediction, aiming to predict user activities at a specific location and time, is crucial for applications like urban planning and mobile advertising. Existing solutions based on tensor decomposition or graph embedding suffer from the following two major limitations: 1) ignoring the fine-grained similarities of user preferences; 2) user's modeling is entangled. In this work, we propose a hypergraph neural network model called DisenHCN to bridge the above gaps. In particular, we first unify the fine-grained user similarity and the complex matching between user preferences and spatiotemporal activity into a heterogeneous hypergraph. We then disentangle the user representations into different aspects (location-aware, time-aware, and activity-aware) and aggregate corresponding aspect's features on the constructed hypergraph, capturing high-order relations from different aspects and disentangles the impact of each aspect for final prediction. Extensive experiments show that our DisenHCN outperforms the state-of-the-art methods by 14.23% to 18.10% on four real-world datasets. Further studies also convincingly verify the rationality of each component in our DisenHCN.
Recent studies shows that the majority of existing deep steganalysis models have a large amount of redundancy, which leads to a huge waste of storage and computing resources. The existing model compression method cannot flexibly compress the convolutional layer in residual shortcut block so that a satisfactory shrinking rate cannot be obtained. In this paper, we propose STD-NET, an unsupervised deep-learning architecture search approach via hierarchical tensor decomposition for image steganalysis. Our proposed strategy will not be restricted by various residual connections, since this strategy does not change the number of input and output channels of the convolution block. We propose a normalized distortion threshold to evaluate the sensitivity of each involved convolutional layer of the base model to guide STD-NET to compress target network in an efficient and unsupervised approach, and obtain two network structures of different shapes with low computation cost and similar performance compared with the original one. Extensive experiments have confirmed that, on one hand, our model can achieve comparable or even better detection performance in various steganalytic scenarios due to the great adaptivity of the obtained network architecture. On the other hand, the experimental results also demonstrate that our proposed strategy is more efficient and can remove more redundancy compared with previous steganalytic network compression methods.
We introduce canonical weight normalization for convolutional neural networks. Inspired by the canonical tensor decomposition, we express the weight tensors in so-called canonical networks as scaled sums of outer vector products. In particular, we train network weights in the decomposed form, where scale weights are optimized separately for each mode. Additionally, similarly to weight normalization, we include a global scaling parameter. We study the initialization of the canonical form by running the power method and by drawing randomly from Gaussian or uniform distributions. Our results indicate that we can replace the power method with cheaper initializations drawn from standard distributions. The canonical re-parametrization leads to competitive normalization performance on the MNIST, CIFAR10, and SVHN data sets. Moreover, the formulation simplifies network compression. Once training has converged, the canonical form allows convenient model-compression by truncating the parameter sums.
Tensorial Convolutional Neural Networks (TCNNs) have attracted much research attention for their power in reducing model parameters or enhancing the generalization ability. However, exploration of TCNNs is hindered even from weight initialization methods. To be specific, general initialization methods, such as Xavier or Kaiming initialization, usually fail to generate appropriate weights for TCNNs. Meanwhile, although there are ad-hoc approaches for specific architectures (e.g., Tensor Ring Nets), they are not applicable to TCNNs with other tensor decomposition methods (e.g., CP or Tucker decomposition). To address this problem, we propose a universal weight initialization paradigm, which generalizes Xavier and Kaiming methods and can be widely applicable to arbitrary TCNNs. Specifically, we first present the Reproducing Transformation to convert the backward process in TCNNs to an equivalent convolution process. Then, based on the convolution operators in the forward and backward processes, we build a unified paradigm to control the variance of features and gradients in TCNNs. Thus, we can derive fan-in and fan-out initialization for various TCNNs. We demonstrate that our paradigm can stabilize the training of TCNNs, leading to faster convergence and better results.
Recent work explored the potential of large-scale Transformer-based pre-trained models, especially Pre-trained Language Models (PLMs) in natural language processing. This raises many concerns from various perspectives, e.g., financial costs and carbon emissions. Compressing PLMs like BERT with negligible performance loss for faster inference and cheaper deployment has attracted much attention. In this work, we aim to explore larger compression ratios for PLMs, among which tensor decomposition is a potential but under-investigated one. Two decomposition and reconstruction protocols are further proposed to improve the effectiveness and efficiency during compression. Our compressed BERT with ${1}/{7}$ parameters in Transformer layers performs on-par with, sometimes slightly better than the original BERT in GLUE benchmark. A tiny version achieves $96.7\%$ performance of BERT-base with $ {1}/{48} $ encoder parameters (i.e., less than 2M parameters excluding the embedding layer) and $2.7 \times$ faster on inference. To show that the proposed method is orthogonal to existing compression methods like knowledge distillation, we also explore the benefit of the proposed method on a distilled BERT.
Knowledge graph (KG) embedding methods which map entities and relations to unique embeddings in the KG have shown promising results on many reasoning tasks. However, the same embedding dimension for both dense entities and sparse entities will cause either over parameterization (sparse entities) or under fitting (dense entities). Normally, a large dimension is set to get better performance. Meanwhile, the inference time grows log-linearly with the number of entities for all entities are traversed and compared. Both the parameter and inference become challenges when working with huge amounts of entities. Thus, we propose PIE, a \textbf{p}arameter and \textbf{i}nference \textbf{e}fficient solution. Inspired from tensor decomposition methods, we find that decompose entity embedding matrix into low rank matrices can reduce more than half of the parameters while maintaining comparable performance. To accelerate model inference, we propose a self-supervised auxiliary task, which can be seen as fine-grained entity typing. By randomly masking and recovering entities' connected relations, the task learns the co-occurrence of entity and relations. Utilizing the fine grained typing, we can filter unrelated entities during inference and get targets with possibly sub-linear time requirement. Experiments on link prediction benchmarks demonstrate the proposed key capabilities. Moreover, we prove effectiveness of the proposed solution on the Open Graph Benchmark large scale challenge dataset WikiKG90Mv2 and achieve the state of the art performance.
Modern recommender systems operate in a fully server-based fashion. To cater to millions of users, the frequent model maintaining and the high-speed processing for concurrent user requests are required, which comes at the cost of a huge carbon footprint. Meanwhile, users need to upload their behavior data even including the immediate environmental context to the server, raising the public concern about privacy. On-device recommender systems circumvent these two issues with cost-conscious settings and local inference. However, due to the limited memory and computing resources, on-device recommender systems are confronted with two fundamental challenges: (1) how to reduce the size of regular models to fit edge devices? (2) how to retain the original capacity? Previous research mostly adopts tensor decomposition techniques to compress the regular recommendation model with limited compression ratio so as to avoid drastic performance degradation. In this paper, we explore ultra-compact models for next-item recommendation, by loosing the constraint of dimensionality consistency in tensor decomposition. Meanwhile, to compensate for the capacity loss caused by compression, we develop a self-supervised knowledge distillation framework which enables the compressed model (student) to distill the essential information lying in the raw data, and improves the long-tail item recommendation through an embedding-recombination strategy with the original model (teacher). The extensive experiments on two benchmarks demonstrate that, with 30x model size reduction, the compressed model almost comes with no accuracy loss, and even outperforms its uncompressed counterpart in most cases.
Mean field theory has provided theoretical insights into various algorithms by letting the problem size tend to infinity. We argue that the applications of mean-field theory go beyond theoretical insights as it can inspire the design of practical algorithms. Leveraging mean-field analyses in physics, we propose a novel algorithm for sparse measure recovery. For sparse measures over $\mathbb{R}$, we propose a polynomial-time recovery method from Fourier moments that improves upon convex relaxation methods in a specific parameter regime; then, we demonstrate the application of our results for the optimization of particular two-dimensional, single-layer neural networks in realizable settings.
Temporal knowledge graph completion (TKGC) has become a popular approach for reasoning over the event and temporal knowledge graphs, targeting the completion of knowledge with accurate but missing information. In this context, tensor decomposition has successfully modeled interactions between entities and relations. Their effectiveness in static knowledge graph completion motivates us to introduce Time-LowFER, a family of parameter-efficient and time-aware extensions of the low-rank tensor factorization model LowFER. Noting several limitations in current approaches to represent time, we propose a cycle-aware time-encoding scheme for time features, which is model-agnostic and offers a more generalized representation of time. We implement our methods in a unified temporal knowledge graph embedding framework, focusing on time-sensitive data processing. The experiments show that our proposed methods perform on par or better than the state-of-the-art semantic matching models on two benchmarks.