Time-consuming performance evaluation is the bottleneck of traditional Neural Architecture Search (NAS) methods. Predictor-based NAS can speed up performance evaluation by directly predicting performance, rather than training a large number of sub-models and then validating their performance. Most predictor-based NAS approaches use a proxy dataset to train model-based predictors efficiently but suffer from performance degradation and generalization problems. We attribute these problems to the poor abilities of existing predictors to character the sub-models' structure, specifically the topology information extraction and the node feature representation of the input graph data. To address these problems, we propose a Transformer-like NAS predictor PINAT, consisting of a Permutation INvariance Augmentation module serving as both token embedding layer and self-attention head, as well as a Laplacian matrix to be the positional encoding. Our design produces more representative features of the encoded architecture and outperforms state-of-the-art NAS predictors on six search spaces: NAS-Bench-101, NAS-Bench-201, DARTS, ProxylessNAS, PPI, and ModelNet. The code is available at https://github.com/ShunLu91/PINAT.

Designing neural architectures requires immense manual efforts. This has promoted the development of neural architecture search (NAS) to automate the design. While previous NAS methods achieve promising results but run slowly, zero-cost proxies run extremely fast but are less promising. Therefore, it’s of great potential to accelerate NAS via those zero-cost proxies. The existing method has two limitations, which are unforeseeable reliability and one-shot usage. To address the limitations, we present ProxyBO, an efficient Bayesian optimization (BO) framework that utilizes the zero-cost proxies to accelerate neural architecture search. We apply the generalization ability measurement to estimate the fitness of proxies on the task during each iteration and design a novel acquisition function to combine BO with zero-cost proxies based on their dynamic influence. Extensive empirical studies show that ProxyBO consistently outperforms competitive baselines on five tasks from three public benchmarks. Concretely, ProxyBO achieves up to 5.41× and 3.86× speedups over the state-of-the-art approaches REA and BRP-NAS.

Recent years have witnessed an exponential growth of model scale in deep learning-based recommender systems---from Google's 2016 model with 1 billion parameters to the latest Facebook's model with 12 trillion parameters. Significant quality boost has come with each jump of the model capacity, which makes us believe the era of 100 trillion parameters is around the corner. However, the training of such models is challenging even within industrial scale data centers. We resolve this challenge by careful co-design of both optimization algorithm and distributed system architecture. Specifically, to ensure both the training efficiency and the training accuracy, we design a novel hybrid training algorithm, where the embedding layer and the dense neural network are handled by different synchronization mechanisms; then we build a system called Persia (short for parallel recommendation training system with hybrid acceleration) to support this hybrid training algorithm. Both theoretical demonstrations and empirical studies with up to 100 trillion parameters have been conducted to justify the system design and implementation of Persia. We make Persia publicly available (at github.com/PersiaML/Persia) so that anyone can easily train a recommender model at the scale of 100 trillion parameters.

Multi-task learning (MTL) has been widely utilized in various industrial scenarios, such as recommender systems and search engines. MTL can improve learning efficiency and prediction accuracy by exploiting commonalities and differences across tasks. However, MTL is sensitive to relationships among tasks and may have performance degradation in real-world applications, because existing neural-based MTL models often share the same network structures and original input features. To address this issue, we propose a novel multi-task learning model based on Prototype Feature Extraction (PFE) to balance task-specific objectives and inter-task relationships. PFE is a novel component to disentangle features for multiple tasks. To better extract features from original inputs before gating networks, we introduce a new concept, namely prototype feature center, to disentangle features for multiple tasks. The extracted prototype features fuse various features from different tasks to better learn inter-task relationships. PFE updates prototype feature centers and prototype features iteratively. Our model utilizes the learned prototype features and task-specific experts for MTL. We implement PFE on two public datasets. Empirical results show that PFE outperforms state-of-the-art MTL models by extracting prototype features. Furthermore, we deploy PFE in a real-world recommender system (one of the world’s top-tier short video sharing platforms) to showcase that PFE can be widely applied in industrial scenarios.

Recently, x-vector [1] has been a successful and popular approach for speaker verification, which employs a time delay neural network (TDNN) and statistics pooling to extract speaker characterizing embedding from variable-length utterances. Improvement upon the x-vector has been an active research area, and enormous neural networks have been elaborately designed based on the x-vector, e.g., extended TDNN (E-TDNN) [2], factorized TDNN (F-TDNN) [3], and densely connected TDNN (D-TDNN) [4]. In this work, we try to identify the optimal architectures from a TDNN based search space employing neural architecture search (NAS), named SpeechNAS. Leveraging the recent advances in the speaker recognition, such as high-order statistics pooling, multi-branch mechanism, D-TDNN and angular additive margin softmax (AAM) loss with a minimum hyper-spherical energy (MHE), SpeechNAS automatically discovers five network architectures, from SpeechNAS-1 to SpeechNAS-5, of various numbers of parameters and GFLOPs on the large-scale text-independent speaker recognition dataset VoxCelebl. Our derived best neural network achieves an equal error rate (EER) of 1.02% on the standard test set of VoxCelebl, which surpasses previous TDNN based state-of-the-art approaches by a large margin.

Model compression techniques are recently gaining explosive attention for obtaining efficient AI models for various real time applications. Channel pruning is one important compression strategy, and widely used in slimming various DNNs. Previous gate-based or importance-based pruning methods aim to remove channels whose "importance" are smallest. However, it remains unclear what criteria the channel importance should be measured on, leading to various channel selection heuristics. Some other sampling-based pruning methods deploy sampling strategy to train sub-nets, which often causes the training instability and the compressed model’s degraded performance. In view of the research gaps, we present a new module named Gates with Differentiable Polarization (GDP), inspired by principled optimization ideas. GDP can be plugged before convolutional layers without bells and whistles, to control the on-and-off of each channel or whole layer block. During the training process, the polarization effect will drive a subset of gates to smoothly decrease to exact zero, while other gates gradually stay away from zero by a large margin. When training terminates, those zero-gated channels can be painlessly removed, while other non-zero gates can be absorbed into the succeeding convolution kernel, causing completely no interruption to training nor damage to the trained model. Experiments conducted over CIFAR-10 and ImageNet datasets show that the proposed GDP algorithm achieves the state-of-the-art performance on various benchmark DNNs at a broad range of pruning ratios. We also apply GDP to DeepLabV3Plus-ResNet50 on the challenging Pascal VOC segmentation task, whose test performance sees no drop (even slightly improved) with over 60% FLOPs saving.

Predictor-based Neural Architecture Search (NAS) continues to be an important topic because it aims to mitigate the time-consuming search procedure of traditional NAS methods. A promising performance predictor determines the quality of final searched models in predictor-based NAS methods. Most existing predictor-based methodologies train model-based predictors under a proxy dataset setting, which may suffer from the accuracy decline and the generalization problem, mainly due to their poor abilities to represent spatial topology information of the graph structure data. Besides the poor encoding for spatial topology information, these works did not take advantage of the temporal information such as historical evaluations during training. Thus, we propose a Transformer-based NAS performance predictor, associated with a Laplacian matrix based positional encoding strategy, which better represents topology information and achieves better performance than previous state-of-the-art methods on NAS-Bench-101, NAS-Bench-201, and DARTS search space. Furthermore, we also propose a self-evolution framework that can fully utilize temporal information as guidance. This framework iteratively involves the evaluations of previously predicted results as constraints into current optimization iteration, thus further improving the performance of our predictor. Such framework is model-agnostic, thus can enhance performance on various backbone structures for the prediction task. Our proposed method helped us rank 2nd among all teams in CVPR 2021 NAS Competition Track 2: Performance Prediction Track.

Spatio-temporal representational learning has been widely adopted in various fields such as action recognition, video object segmentation, and action anticipation.Previous spatio-temporal representational learning approaches primarily employ ConvNets or sequential models, e.g., LSTM, to learn the intra-frame and inter-frame features. Recently, Transformer models have successfully dominated the study of natural language processing (NLP), image classification, etc. However, the pure-Transformer based spatio-temporal learning can be prohibitively costly on memory and computation to extract fine-grained features from a tiny patch. To tackle the training difficulty and enhance the spatio-temporal learning, we construct a shifted chunk Transformer with pure self-attention blocks. Leveraging the recent efficient Transformer design in NLP, this shifted chunk Transformer can learn hierarchical spatio-temporal features from a local tiny patch to a global videoclip. Our shifted self-attention can also effectively model complicated inter-frame variances. Furthermore, we build a clip encoder based on Transformer to model long-term temporal dependencies. We conduct thorough ablation studies to validate each component and hyper-parameters in our shifted chunk Transformer, and it outperforms previous state-of-the-art approaches on Kinetics-400, Kinetics-600,UCF101, and HMDB51.

Recent years have witnessed a growing list of systems for distributed data-parallel training. Existing systems largely fit into two paradigms, i.e., parameter server and MPI-style collective operations. On the algorithmic side, researchers have proposed a wide range of techniques to lower the communication via "system relaxations": quantization, decentralization, and communication delay. However, most, if not all, existing systems only rely on standard synchronous and asynchronous stochastic gradient (SG) based optimization, therefore, cannot take advantage of all possible optimizations that the machine learning community has been developing recently. Given this emerging gap between the current landscapes of systems and theory, we build Bagua, a MPI-style communication library, providing a collection of primitives, that is both flexible and modular to support state-of-the-art system relaxation techniques of distributed training. Powered by this design, Bagua has a great ability to implement and extend various state-of-the-art distributed learning algorithms. In a production cluster with up to 16 machines (128 GPUs), Bagua can outperform PyTorch-DDP, Horovod and BytePS in the end-to-end training time by a significant margin (up to 2X) across a diverse range of tasks. Moreover, we conduct a rigorous tradeoff exploration showing that different algorithms and system relaxations achieve the best performance over different network conditions.

Investigations into brain connectivity aim to recover networks of brain regions connected by anatomical tracts or by functional associations. The inference of brain networks has recently attracted much interest due to the increasing availability of high-resolution brain imaging data. Sparse inverse covariance estimation with lasso and group lasso penalty has been demonstrated to be a powerful approach to discover brain networks. Motivated by the hierarchical structure of the brain networks, we consider the problem of estimating a graphical model with tree-structural regularization in this paper. The regularization encourages the graphical model to exhibit a brain-like structure. Specifically, in this hierarchical structure, hundreds of thousands of voxels serve as the leaf nodes of the tree. A node in the intermediate layer represents a region formed by voxels in the subtree rooted at that node. The whole brain is considered as the root of the tree. We propose to apply the tree-structural regularized graphical model to estimate the mouse brain network. However, the dimensionality of whole-brain data, usually on the order of hundreds of thousands, poses significant computational challenges. Efficient algorithms that are capable of estimating networks from high-dimensional data are highly desired. To address the computational challenge, we develop a screening rule which can quickly identify many zero blocks in the estimated graphical model, thereby dramatically reducing the computational cost of solving the proposed model. It is based on a novel insight on the relationship between screening and the so-called proximal operator that we first establish in this paper. We perform experiments on both synthetic data and real data from the Allen Developing Mouse Brain Atlas; results demonstrate the effectiveness and efficiency of the proposed approach.

Total variation (TV) models are among the most popular and successful tools in signal processing. However, due to the complex nature of the TV term, it is challenging to efficiently compute a solution for large-scale problems. State-of-the-art algorithms that are based on the alternating direction method of multipliers (ADMM) often involve solving large-size linear systems. In this paper, we propose a highly scalable parallel algorithm for TV models that is based on a novel decomposition strategy of the problem domain. As a result, the TV models can be decoupled into a set of small and independent subproblems, which admit closed form solutions. This makes our approach particularly suitable for parallel implementation. Our algorithm is guaranteed to converge to its global minimum. With N variables and n_p processes, the time complexity is O(N/(εn_p)) to reach an epsilon-optimal solution. Extensive experiments demonstrate that our approach outperforms existing state-of-the-art algorithms, especially in dealing with high-resolution, mega-size images.

Total variation (TV) regularization has important applications in signal processing including image denoising, image deblurring, and image reconstruction. A significant challenge in the practical use of TV regularization lies in the nondifferentiable convex optimization, which is difficult to solve especially for large-scale problems. In this paper, we propose an efficient alternating augmented Lagrangian method (ADMM) to solve total variation regularization problems. The proposed algorithm is applicable for tensors, thus it can solve multidimensional total variation regularization problems. One appealing feature of the proposed algorithm is that it does not need to solve a linear system of equations, which is often the most expensive part in previous ADMM-based methods. In addition, each step of the proposed algorithm involves a set of independent and smaller problems, which can be solved in parallel. Thus, the proposed algorithm scales to large size problems. Furthermore, the global convergence of the proposed algorithm is guaranteed, and the time complexity of the proposed algorithm is O(dN/ε) on a d-mode tensor with N entries for achieving an ε-optimal solution. Extensive experimental results demonstrate the superior performance of the proposed algorithm in comparison with current state-of-the-art methods.

High-dimensional regression/classification continues to be an important and challenging problem, especially when features are highly correlated. Feature selection, combined with additional structure information on the features has been considered to be promising in promoting regression/classification performance. Graph-guided fused lasso (GFlasso) has recently been proposed to facilitate feature selection and graph structure exploitation, when features exhibit certain graph structures. However, the formulation in GFlasso relies on pairwise sample correlations to perform feature grouping, which could introduce additional estimation bias. In this paper, we propose three new feature grouping and selection methods to resolve this issue. The first method employs a convex function to penalize the pairwise l∞ norm of connected regression/classification coefficients, achieving simultaneous feature grouping and selection. The second method improves the first one by utilizing a non-convex function to reduce the estimation bias. The third one is the extension of the second method using a truncated l1 regularization to further reduce the estimation bias. The proposed methods combine feature grouping and feature selection to enhance estimation accuracy. We employ the alternating direction method of multipliers (ADMM) and difference of convex functions (DC) programming to solve the proposed formulations. Our experimental results on synthetic data and two real datasets demonstrate the effectiveness of the proposed methods.