Neural Architecture Search (NAS) has been explosively studied to automate the discovery of top-performer neural networks. Current works require heavy training of supernet or intensive architecture evaluations, thus suffering from heavy resource consumption and often incurring search bias due to truncated training or approximations. Can we select the best neural architectures without involving any training and eliminate a drastic portion of the search cost? We provide an affirmative answer, by proposing a novel framework called training-free neural architecture search (TE-NAS). TE-NAS ranks architectures by analyzing the spectrum of the neural tangent kernel (NTK) and the number of linear regions in the input space. Both are motivated by recent theory advances in deep networks and can be computed without any training and any label. We show that: (1) these two measurements imply the trainability and expressivity of a neural network; (2) they strongly correlate with the network's test accuracy. Further on, we design a pruning-based NAS mechanism to achieve a more flexible and superior trade-off between the trainability and expressivity during the search. In NAS-Bench-201 and DARTS search spaces, TE-NAS completes high-quality search but only costs 0.5 and 4 GPU hours with one 1080Ti on CIFAR-10 and ImageNet, respectively. We hope our work inspires more attempts in bridging the theoretical findings of deep networks and practical impacts in real NAS applications. Code is available at: https://github.com/VITA-Group/TENAS.
Google Research（谷歌研究院）；University of California, Berkeley（加州大学伯克利分校）；National University of Singapore（新加坡国立大学）
EfficientNets are a family of state-of-the-art image classification models based on efficiently scaled convolutional neural networks. Currently, EfficientNets can take on the order of days to train; for example, training an EfficientNet-B0 model takes 23 hours on a Cloud TPU v2-8 node. In this paper, we explore techniques to scale up the training of EfficientNets on TPU-v3 Pods with 2048 cores, motivated by speedups that can be achieved when training at such scales. We discuss optimizations required to scale training to a batch size of 65536 on 1024 TPU-v3 cores, such as selecting large batch …
Deep learning thrives with large neural networks and large datasets. However, larger networks and larger datasets result in longer training times that impede research and development progress. Distributed synchronous SGD offers a potential solution to this problem by dividing SGD minibatches over a pool of parallel workers. Yet to make this scheme efficient, the per-worker workload must be large, which implies nontrivial growth in the SGD minibatch size. In this paper, we empirically show that on the ImageNet dataset large minibatches cause optimization difficulties, but when these are address…
University of Technology Sydney（悉尼科技大学）；Baidu Research（百度研究院）
Conventional neural architecture search (NAS) approaches are based on reinforcement learning or evolutionary strategy, which take more than 3000 GPU hours to find a good model on CIFAR-10. We propose an efficient NAS approach learning to search by gradient descent. Our approach represents the search space as a directed acyclic graph (DAG). This DAG contains billions of sub-graphs, each of which indicates a kind of neural architecture. To avoid traversing all the possibilities of the sub-graphs, we develop a differentiable sampler over the DAG. This sampler is learnable and optimized by the val…
Transformers have become central to recent advances in computer vision. However, training a vision Transformer (ViT) model from scratch can be resource intensive and time consuming. In this paper, we aim to explore approaches to reduce the training costs of ViT models. We introduce some algorithmic improvements to enable training a ViT model from scratch with limited hardware (1 GPU) and time (24 hours) resources. First, we propose an efficient approach to add locality to the ViT architecture. Second, we develop a new image size curriculum learning strategy, which allows to reduce the number o…
Neural architecture search (NAS) has a great impact by automatically designing effective neural network architectures. However, the prohibitive computational demand of conventional NAS algorithms (e.g. $10^4$ GPU hours) makes it difficult to \emph{directly} search the architectures on large-scale tasks (e.g. ImageNet). Differentiable NAS can reduce the cost of GPU hours via a continuous representation of network architecture but suffers from the high GPU memory consumption issue (grow linearly w.r.t. candidate set size). As a result, they need to utilize~\emph{proxy} tasks, such as training on…
Carnegie Mellon University（卡内基梅隆大学）；Microsoft（微软）；Harbin Institute of Technology（哈尔滨工业大学）
Can we automatically design a Convolutional Network (ConvNet) with the highest image classification accuracy under the runtime constraint of a mobile device? Neural architecture search (NAS) has revolutionized the design of hardware-efficient ConvNets by automating this process. However, the NAS problem remains challenging due to the combinatorially large design space, causing a significant searching time (at least 200 GPU-hours). To alleviate this complexity, we propose Single-Path NAS, a novel differentiable NAS method for designing hardware-efficient ConvNets in less than 4 hours. Our contr…
Hugging Face（Hugging Face公司）；LightOn（LightOn公司）；NTU, Singapore（新加坡南洋理工大学）；Booz Allen（博思艾伦公司）；EleutherAI（EleutherAI研究院）；Naver Labs Europe（Naver欧洲实验室）；University of Washington（华盛顿大学）；Berkeley University（加州大学伯克利分校）；Big Science（Big Science项目组）；Brown University（布朗大学）；LPENS（巴黎高等师范学院物理实验室）；Allen Institute for AI（艾伦人工智能研究院）；New York University（纽约大学）
The crystallization of modeling methods around the Transformer architecture has been a boon for practitioners. Simple, well-motivated architectural variations can transfer across tasks and scale, increasing the impact of modeling research. However, with the emergence of state-of-the-art 100B+ parameters models, large language models are increasingly expensive to accurately design and train. Notably, it can be difficult to evaluate how modeling decisions may impact emergent capabilities, given that these capabilities arise mainly from sheer scale alone. In the process of building BLOOM--the Big…
We explore efficient neural architecture search methods and show that a simple yet powerful evolutionary algorithm can discover new architectures with excellent performance. Our approach combines a novel hierarchical genetic representation scheme that imitates the modularized design pattern commonly adopted by human experts, and an expressive search space that supports complex topologies. Our algorithm efficiently discovers architectures that outperform a large number of manually designed models for image classification, obtaining top-1 error of 3.6% on CIFAR-10 and 20.3% when transferred to I…
Peking University（北京大学）；Huawei Technologies（华为技术有限公司）；University of Sydney（悉尼大学）
Neural Architecture Search (NAS) refers to automatically design the architecture. We propose an hourglass-inspired approach (HourNAS) for this problem that is motivated by the fact that the effects of the architecture often proceed from the vital few blocks. Acting like the narrow neck of an hourglass, vital blocks in the guaranteed path from the input to the output of a deep neural network restrict the information flow and influence the network accuracy. The other blocks occupy the major volume of the network and determine the overall network complexity, corresponding to the bulbs of an hourg…
Many mission-critical systems are based on GPU for inference. It requires not only high recognition accuracy but also low latency in responding time. Although many studies are devoted to optimizing the structure of deep models for efficient inference, most of them do not leverage the architecture of \textbf{modern GPU} for fast inference, leading to suboptimal performance. To address this issue, we propose a general principle for designing GPU-efficient networks based on extensive empirical studies. This design principle enables us to search for GPU-efficient network structures effectively by …
University of South Australia（南澳大学）；Royal University of Bhutan（不丹皇家大学）；Naresuan University（那黎宣大学）
The goals of this research were to search for Convolutional Neural Network (CNN) architectures, suitable for an on-device processor with limited computing resources, performing at substantially lower Network Architecture Search (NAS) costs. A new algorithm entitled an Early Exit Population Initialisation (EE-PI) for Evolutionary Algorithm (EA) was developed to achieve both goals. The EE-PI reduces the total number of parameters in the search process by filtering the models with fewer parameters than the maximum threshold. It will look for a new model to replace those models with parameters mor…
Understanding the patterns of misclassified ImageNet images is particularly important, as it could guide us to design deep neural networks (DNN) that generalize better. However, the richness of ImageNet imposes difficulties for researchers to visually find any useful patterns of misclassification. Here, to help find these patterns, we propose "Superclassing ImageNet dataset". It is a subset of ImageNet which consists of 10 superclasses, each containing 7-116 related subclasses (e.g., 52 bird types, 116 dog types). By training neural networks on this dataset, we found that: (i) Misclassificatio…
The evolution of MobileNets has laid a solid foundation for neural network applications on mobile end. With the latest MobileNetV3, neural architecture search again claimed its supremacy in network design. Unfortunately, till today all mobile methods mainly focus on CPU latencies instead of GPU, the latter, however, is much preferred in practice for it has faster speed, lower overhead and less interference. Bearing the target hardware in mind, we propose the first Mobile GPU-Aware (MoGA) neural architecture search in order to be precisely tailored for real-world applications. Further, the ulti…
Boost Vision Transformer with GPU-Friendly Sparsity and Quantization
Academy for Engineering and Technology, Fudan University（复旦大学工程与应用技术研究院）；NVIDIA Corporation（英伟达公司）；School for Information Science and Technology, Fudan University（复旦大学信息科学与工程学院）
University of Virginia（弗吉尼亚大学）；Facebook AI Research（Facebook AI研究院）
We introduce CryptGPU, a system for privacy-preserving machine learning that implements all operations on the GPU (graphics processing unit). Just as GPUs played a pivotal role in the success of modern deep learning, they are also essential for realizing scalable privacy-preserving deep learning. In this work, we start by introducing a new interface to losslessly embed cryptographic operations over secret-shared values (in a discrete domain) into floating-point operations that can be processed by highly-optimized CUDA kernels for linear algebra. We then identify a sequence of "GPU-friendly" cr…
Previous works on meta-learning either relied on elaborately hand-designed network structures or adopted specialized learning rules to a particular domain. We propose a universal framework to optimize the meta-learning process automatically by adopting neural architecture search technique (NAS). NAS automatically generates and evaluates meta-learner's architecture for few-shot learning problems, while the meta-learner uses meta-learning algorithm to optimize its parameters based on the distribution of learning tasks. Parameter sharing and experience replay are adopted to accelerate the archite…
Scaling the distributed deep learning to a massive GPU cluster level is challenging due to the instability of the large mini-batch training and the overhead of the gradient synchronization. We address the instability of the large mini-batch training with batch-size control and label smoothing. We address the overhead of the gradient synchronization with 2D-Torus all-reduce. Specifically, 2D-Torus all-reduce arranges GPUs in a logical 2D grid and performs a series of collective operation in different orientations. These two techniques are implemented with Neural Network Libraries (NNL). We have…
Xiamen University（厦门大学）；Peng Cheng Laboratory（鹏城实验室）；Beihang University（北京航空航天大学）；Huawei Noah’s Ark Lab（华为诺亚方舟实验室）；University of Texas at San Antonio（德克萨斯大学圣安东尼奥分校）
Architectures obtained by Neural Architecture Search (NAS) have achieved highly competitive performance in various computer vision tasks. However, the prohibitive computation demand of forward-backward propagation in deep neural networks and searching algorithms makes it difficult to apply NAS in practice. In this paper, we propose a Multinomial Distribution Learning for extremely effective NAS,which considers the search space as a joint multinomial distribution, i.e., the operation between two nodes is sampled from this distribution, and the optimal network structure is obtained by the operat…
Southern University of Science and Technology（南方科技大学）
In the recent past, neural architecture search (NAS) has attracted increasing attention from both academia and industries. Despite the steady stream of impressive empirical results, most existing NAS algorithms are computationally prohibitive to execute due to the costly iterations of stochastic gradient descent (SGD) training. In this work, we propose an effective alternative, dubbed Random-Weight Evaluation (RWE), to rapidly estimate the performance of network architectures. By just training the last linear classification layer, RWE reduces the computational cost of evaluating an architectur…
Carnegie Mellon University（卡内基梅隆大学）；Microsoft（微软）；Harbin Institute of Technology（哈尔滨工业大学）
Can we automatically design a Convolutional Network (ConvNet) with the highest image classification accuracy under the latency constraint of a mobile device? Neural Architecture Search (NAS) for ConvNet design is a challenging problem due to the combinatorially large design space and search time (at least 200 GPU-hours). To alleviate this complexity, we propose Single-Path NAS, a novel differentiable NAS method for designing device-efficient ConvNets in less than 4 hours. 1. Novel NAS formulation: our method introduces a single-path, over-parameterized ConvNet to encode all architectural decis…
Is ImageNet worth 1 video? Learning strong image encoders from 1 long unlabelled video
Inria（法国国家信息与自动化研究所）；Univ Rennes（雷恩大学）；CNRS（法国国家科学研究中心）；IRISA（法国计算机科学与随机系统研究所）；University of Central Florida（中佛罗里达大学）；Google DeepMind（谷歌DeepMind）；University of Amsterdam（阿姆斯特丹大学）；Institute of Advanced Research on Artificial Intelligence (IARAI)（人工智能高级研究所）
We extract an optimal subset of architectural parameters for the BERT architecture from Devlin et al. (2018) by applying recent breakthroughs in algorithms for neural architecture search. This optimal subset, which we refer to as "Bort", is demonstrably smaller, having an effective (that is, not counting the embedding layer) size of $5.5\%$ the original BERT-large architecture, and $16\%$ of the net size. Bort is also able to be pretrained in $288$ GPU hours, which is $1.2\%$ of the time required to pretrain the highest-performing BERT parametric architectural variant, RoBERTa-large (Liu et al…
University of Notre Dame（圣母大学）；University of New Mexico（新墨西哥大学）；Edgecortix Inc.（Edgecortix公司）；University of Pittsburgh（匹兹堡大学）
Hardware and neural architecture co-search that automatically generates Artificial Intelligence (AI) solutions from a given dataset is promising to promote AI democratization; however, the amount of time that is required by current co-search frameworks is in the order of hundreds of GPU hours for one target hardware. This inhibits the use of such frameworks on commodity hardware. The root cause of the low efficiency in existing co-search frameworks is the fact that they start from a "cold" state (i.e., search from scratch). In this paper, we propose a novel framework, namely HotNAS, that start…
National University of Singapore（新加坡国立大学）；HPC-AI Technology Inc.（HPC-AI科技公司）；University of California, San Diego（加州大学圣地亚哥分校）；Xi’an Jiaotong University（西安交通大学）；HeliXon（HeliXon公司）
Protein structure prediction helps to understand gene translation and protein function, which is of growing interest and importance in structural biology. The AlphaFold model, which used transformer architecture to achieve atomic-level accuracy in protein structure prediction, was a significant breakthrough. However, training and inference of the AlphaFold model are challenging due to its high computation and memory cost. In this work, we present FastFold, an efficient implementation of AlphaFold for both training and inference. We propose Dynamic Axial Parallelism and Duality Async Operations…
Many deep learning models, developed in recent years, reach higher ImageNet accuracy than ResNet50, with fewer or comparable FLOPS count. While FLOPs are often seen as a proxy for network efficiency, when measuring actual GPU training and inference throughput, vanilla ResNet50 is usually significantly faster than its recent competitors, offering better throughput-accuracy trade-off. In this work, we introduce a series of architecture modifications that aim to boost neural networks' accuracy, while retaining their GPU training and inference efficiency. We first demonstrate and discuss the bottl…
We demonstrate that training ResNet-50 on ImageNet for 90 epochs can be achieved in 15 minutes with 1024 Tesla P100 GPUs. This was made possible by using a large minibatch size of 32k. To maintain accuracy with this large minibatch size, we employed several techniques such as RMSprop warm-up, batch normalization without moving averages, and a slow-start learning rate schedule. This paper also describes the details of the hardware and software of the system used to achieve the above performance.
It is commonly accepted that the Vision Transformer model requires sophisticated regularization techniques to excel at ImageNet-1k scale data. Surprisingly, we find this is not the case and standard data augmentation is sufficient. This note presents a few minor modifications to the original Vision Transformer (ViT) vanilla training setting that dramatically improve the performance of plain ViT models. Notably, 90 epochs of training surpass 76% top-1 accuracy in under seven hours on a TPUv3-8, similar to the classic ResNet50 baseline, and 300 epochs of training reach 80% in less than one day.
GigaVideo-1: Advancing Video Generation via Automatic Feedback with 4 GPU-Hours Fine-Tuning
GigaAI（极佳科技）；Institute of Automation, Chinese Academy of Sciences（中国科学院自动化研究所）；School of Artificial Intelligence, University of Chinese Academy of Sciences（中国科学院大学人工智能学院）；School of Computer Science, Sichuan University（四川大学计算机学院）
Allen Institute for AI（艾伦人工智能研究所）；University of Washington（华盛顿大学）
We propose two efficient approximations to standard convolutional neural networks: Binary-Weight-Networks and XNOR-Networks. In Binary-Weight-Networks, the filters are approximated with binary values resulting in 32x memory saving. In XNOR-Networks, both the filters and the input to convolutional layers are binary. XNOR-Networks approximate convolutions using primarily binary operations. This results in 58x faster convolutional operations and 32x memory savings. XNOR-Nets offer the possibility of running state-of-the-art networks on CPUs (rather than GPUs) in real-time. Our binary networks are…
Institute of Software, Chinese Academy of Sciences（中国科学院软件研究所）；Yahoo Labs（雅虎实验室）
Fisher vector has been widely used in many multimedia retrieval and visual recognition applications with good performance. However, the computation complexity prevents its usage in real-time video monitoring. In this work, we proposed and implemented GPU-FV, a fast Fisher vector extraction method with the help of modern GPUs. The challenge of implementing Fisher vector on GPUs lies in the data dependency in feature extraction and expensive memory access in Fisher vector computing. To handle these challenges, we carefully designed GPU-FV in a way that utilizes the computing power of GPU as much…
How far can we go with ImageNet for Text-to-Image generation?
École Polytechnique（巴黎综合理工学院）；CNRS（法国国家科学研究中心）；IP Paris（巴黎理工学院）；Kyutai；École Nationale des Ponts et Chaussées（巴黎国立桥路学校）；Univ Gustave Eiffel（古斯塔夫·埃菲尔大学）
Generative adversarial networks (GAN) have shown remarkable results in image generation tasks. High fidelity class-conditional GAN methods often rely on stabilization techniques by constraining the global Lipschitz continuity. Such regularization leads to less expressive models and slower convergence speed; other techniques, such as the large batch training, require unconventional computing power and are not widely accessible. In this paper, we develop an efficient algorithm, namely FastGAN (Free AdverSarial Training), to improve the speed and quality of GAN training based on the adversarial t…
University of Pisa（比萨大学）；ISTI-CNR（意大利国家研究委员会信息科学与技术研究所）；Joanneum Research（约阿尼姆研究中心）
Learning algorithms for Deep Neural Networks are typically based on supervised end-to-end Stochastic Gradient Descent (SGD) training with error backpropagation (backprop). Backprop algorithms require a large number of labelled training samples to achieve high performance. However, in many realistic applications, even if there is plenty of image samples, very few of them are labelled, and semi-supervised sample-efficient training strategies have to be used. Hebbian learning represents a possible approach towards sample efficient training; however, in current solutions, it does not scale well to…
Finishing 90-epoch ImageNet-1k training with ResNet-50 on a NVIDIA M40 GPU takes 14 days. This training requires 10^18 single precision operations in total. On the other hand, the world's current fastest supercomputer can finish 2 * 10^17 single precision operations per second (Dongarra et al 2017, https://www.top500.org/lists/2017/06/). If we can make full use of the supercomputer for DNN training, we should be able to finish the 90-epoch ResNet-50 training in one minute. However, the current bottleneck for fast DNN training is in the algorithm level. Specifically, the current batch size (e.g…
Carnegie Mellon University（卡内基梅隆大学）；Seoul National University（首尔国立大学）；Institute for Infocomm Research（资讯通信研究院）
Deep learning (DL) has achieved notable successes in many machine learning tasks. A number of frameworks have been developed to expedite the process of designing and training deep neural networks (DNNs), such as Caffe, Torch and Theano. Currently they can harness multiple GPUs on a single machine, but are unable to use GPUs that are distributed across multiple machines; as even average-sized DNNs can take days to train on a single GPU with 100s of GBs to TBs of data, distributed GPUs present a prime opportunity for scaling up DL. However, the limited bandwidth available on commodity Ethernet n…
Cornell University（康奈尔大学）；Fudan University（复旦大学）；Facebook AI Research（Facebook AI研究院）
The DenseNet architecture is highly computationally efficient as a result of feature reuse. However, a naive DenseNet implementation can require a significant amount of GPU memory: If not properly managed, pre-activation batch normalization and contiguous convolution operations can produce feature maps that grow quadratically with network depth. In this technical report, we introduce strategies to reduce the memory consumption of DenseNets during training. By strategically using shared memory allocations, we reduce the memory cost for storing feature maps from quadratic to linear. Without the …
Global covariance pooling in convolutional neural networks has achieved impressive improvement over the classical first-order pooling. Recent works have shown matrix square root normalization plays a central role in achieving state-of-the-art performance. However, existing methods depend heavily on eigendecomposition (EIG) or singular value decomposition (SVD), suffering from inefficient training due to limited support of EIG and SVD on GPU. Towards addressing this problem, we propose an iterative matrix square root normalization method for fast end-to-end training of global covariance pooling…
Deep neural network models used for medical image segmentation are large because they are trained with high-resolution three-dimensional (3D) images. Graphics processing units (GPUs) are widely used to accelerate the trainings. However, the memory on a GPU is not large enough to train the models. A popular approach to tackling this problem is patch-based method, which divides a large image into small patches and trains the models with these small patches. However, this method would degrade the segmentation quality if a target object spans multiple patches. In this paper, we propose a novel app…
Large, pre-trained generative models have been increasingly popular and useful to both the research and wider communities. Specifically, BigGANs a class-conditional Generative Adversarial Networks trained on ImageNet---achieved excellent, state-of-the-art capability in generating realistic photos. However, fine-tuning or training BigGANs from scratch is practically impossible for most researchers and engineers because (1) GAN training is often unstable and suffering from mode-collapse; and (2) the training requires a significant amount of computation, 256 Google TPUs for 2 days or 8xV100 GPUs …
Mobile and embedded platforms are increasingly required to efficiently execute computationally demanding DNNs across heterogeneous processing elements. At runtime, the available hardware resources to DNNs can vary considerably due to other concurrently running applications. The performance requirements of the applications could also change under different scenarios. To achieve the desired performance, dynamic DNNs have been proposed in which the number of channels/layers can be scaled in real time to meet different requirements under varying resource constraints. However, the training process …
BMW Autonomous Driving（宝马自动驾驶部门）；Technical University of Munich（慕尼黑工业大学）
Various applications in the field of autonomous driving are based on convolutional neural networks (CNNs), especially for processing camera data. The optimization of such CNNs is a major challenge in continuous development. Newly learned features must be brought into vehicles as quickly as possible, and as such, it is not feasible to spend redundant GPU hours during compression. In this context, we present Learning to Prune Faster which details a multi-task, try-and-learn method, discretely learning redundant filters of the CNN and a continuous action of how long the layers have to be fine-tun…
This paper studies the neural architecture search (NAS) problem for developing efficient generator networks. Compared with deep models for visual recognition tasks, generative adversarial network (GAN) are usually designed to conduct various complex image generation. We first discover an intact search space of generator networks including three dimensionalities, i.e., path, operator, channel for fully excavating the network performance. To reduce the huge search cost, we explore a coarse-to-fine search strategy which divides the overall search process into three sub-optimization problems accor…
Glance Forcing: Making Bidirectional Diffusion Autoregressive at a Glance
Wuhan University（武汉大学）；University of Electronic Science and Technology of China（电子科技大学）；Sun Yat-sen University（中山大学）；University of Washington（华盛顿大学）；National University of Singapore（新加坡国立大学）；Central South University（中南大学）
