PyTorchKR Papers

최신 논문

총 5,576편
태그·기간 등 상세 필터

카테고리 탐색

로그인 후 이 태그 구독

arXiv:2103.03663공식 코드 · Papers with Code

Real-time RGBD-based Extended Body Pose Estimation

We present a system for real-time RGBD-based estimation of 3D human pose. We use parametric 3D deformable human mesh model (SMPL-X) as a representation and focus on the real-time estimation of parameters for the body pose, hands pose and facial expression from Kinect Azure RGB-D camera. We train estimators of body pose and facial expression parameters. Both estimators use previously published landmark extractors as input and custom annotated datasets for supervision, while hand pose is estimated directly by a previously published method. We combine the predictions of those estimators into a temporally-smooth human pose. We train the facial expression extractor on a large talking face dataset, which we annotate with facial expression parameters. For the body pose we collect and annotate a dataset of 56 people captured from a rig of 5 Kinect Azure RGB-D cameras and use it together with a large motion capture AMASS dataset. Our RGB-D body pose model outperforms the state-of-the-art RGB-only methods and works on the same level of accuracy compared to a slower RGB-D optimization-based solution. The combined system runs at 30 FPS on a server with a single GPU. The code will be available at https://saic-violet.github.io/rgbd-kinect-pose

arXiv:2103.02800

Hardware Acceleration of Fully Quantized BERT for Efficient Natural Language Processing

BERT is the most recent Transformer-based model that achieves state-of-the-art performance in various NLP tasks. In this paper, we investigate the hardware acceleration of BERT on FPGA for edge computing. To tackle the issue of huge computational complexity and memory footprint, we propose to fully quantize the BERT (FQ-BERT), including weights, activations, softmax, layer normalization, and all the intermediate results. Experiments demonstrate that the FQ-BERT can achieve 7.94x compression for weights with negligible performance loss. We then propose an accelerator tailored for the FQ-BERT and evaluate on Xilinx ZCU102 and ZCU111 FPGA. It can achieve a performance-per-watt of 3.18 fps/W, which is 28.91x and 12.72x over Intel(R) Core(TM) i7-8700 CPU and NVIDIA K80 GPU, respectively.

arXiv:2103.03231공식 코드 · Papers with Code

DONeRF: Towards Real-Time Rendering of Compact Neural Radiance Fields using Depth Oracle Networks

The recent research explosion around implicit neural representations, such as NeRF, shows that there is immense potential for implicitly storing high-quality scene and lighting information in compact neural networks. However, one major limitation preventing the use of NeRF in real-time rendering applications is the prohibitive computational cost of excessive network evaluations along each view ray, requiring dozens of petaFLOPS. In this work, we bring compact neural representations closer to practical rendering of synthetic content in real-time applications, such as games and virtual reality. We show that the number of samples required for each view ray can be significantly reduced when samples are placed around surfaces in the scene without compromising image quality. To this end, we propose a depth oracle network that predicts ray sample locations for each view ray with a single network evaluation. We show that using a classification network around logarithmically discretized and spherically warped depth values is essential to encode surface locations rather than directly estimating depth. The combination of these techniques leads to DONeRF, our compact dual network design with a depth oracle network as its first step and a locally sampled shading network for ray accumulation. With DONeRF, we reduce the inference costs by up to 48x compared to NeRF when conditioning on available ground truth depth information. Compared to concurrent acceleration methods for raymarching-based neural representations, DONeRF does not require additional memory for explicit caching or acceleration structures, and can render interactively (20 frames per second) on a single GPU.

arXiv:2103.02958

Serverless Data Science -- Are We There Yet? A Case Study of Model Serving

Machine learning (ML) is an important part of modern data science applications. Data scientists today have to manage the end-to-end ML life cycle that includes both model training and model serving, the latter of which is essential, as it makes their works available to end-users. Systems of model serving require high performance, low cost, and ease of management. Cloud providers are already offering model serving choices, including managed services and self-rented servers. Recently, serverless computing, whose advantages include high elasticity and a fine-grained cost model, brings another option for model serving. Our goal in this paper is to examine the viability of serverless as a mainstream model serving platform. To this end, we first conduct a comprehensive evaluation of the performance and cost of serverless against other model serving systems on Amazon Web Service and Google Cloud Platform. We find that serverless outperforms many cloud-based alternatives. Further, there are settings under which it even achieves better performance than GPU-based systems. Next, we present the design space of serverless model serving, which comprises multiple dimensions, including cloud platforms, serving runtimes, and other function-specific parameters. For each dimension, we analyze the impact of different choices and provide suggestions for data scientists to better utilize serverless model serving. Finally, we discuss challenges and opportunities in building a more practical serverless model serving system.

arXiv:2103.01786공식 코드 · Papers with Code

MetaSCI: Scalable and Adaptive Reconstruction for Video Compressive Sensing

To capture high-speed videos using a two-dimensional detector, video snapshot compressive imaging (SCI) is a promising system, where the video frames are coded by different masks and then compressed to a snapshot measurement. Following this, efficient algorithms are desired to reconstruct the high-speed frames, where the state-of-the-art results are achieved by deep learning networks. However, these networks are usually trained for specific small-scale masks and often have high demands of training time and GPU memory, which are hence {\bf \em not flexible} to $i$) a new mask with the same size and $ii$) a larger-scale mask. We address these challenges by developing a Meta Modulated Convolutional Network for SCI reconstruction, dubbed MetaSCI. MetaSCI is composed of a shared backbone for different masks, and light-weight meta-modulation parameters to evolve to different modulation parameters for each mask, thus having the properties of {\bf \em fast adaptation} to new masks (or systems) and ready to {\bf \em scale to large data}. Extensive simulation and real data results demonstrate the superior performance of our proposed approach. Our code is available at {\small\url{https://github.com/xyvirtualgroup/MetaSCI-CVPR2021}}.

arXiv:2103.00241공식 코드 · Papers with Code

Improved Automated Machine Learning from Transfer Learning

In this paper, we propose a neural architecture search framework based on a similarity measure between some baseline tasks and a target task. We first define the notion of the task similarity based on the log-determinant of the Fisher Information matrix. Next, we compute the task similarity from each of the baseline tasks to the target task. By utilizing the relation between a target and a set of learned baseline tasks, the search space of architectures for the target task can be significantly reduced, making the discovery of the best candidates in the set of possible architectures tractable and efficient, in terms of GPU days. This method eliminates the requirement for training the networks from scratch for a given target task as well as introducing the bias in the initialization of the search space from the human domain.

arXiv:2103.00219

Pareto-Frontier-aware Neural Architecture Generation for Diverse Budgets

Designing feasible and effective architectures under diverse computation budgets incurred by different applications/devices is essential for deploying deep models in practice. Existing methods often perform an independent architecture search for each target budget, which is very inefficient yet unnecessary. Moreover, the repeated independent search manner would inevitably ignore the common knowledge among different search processes and hamper the search performance. To address these issues, we seek to train a general architecture generator that automatically produces effective architectures for an arbitrary budget merely via model inference. To this end, we propose a Pareto-Frontier-aware Neural Architecture Generator (NAG) which takes an arbitrary budget as input and produces the Pareto optimal architecture for the target budget. We train NAG by learning the Pareto frontier (i.e., the set of Pareto optimal architectures) over model performance and computational cost (e.g., latency). Extensive experiments on three platforms (i.e., mobile, CPU, and GPU) show the superiority of the proposed method over existing NAS methods.

arXiv:2103.00860

Learning to Enhance Low-Light Image via Zero-Reference Deep Curve Estimation

This paper presents a novel method, Zero-Reference Deep Curve Estimation (Zero-DCE), which formulates light enhancement as a task of image-specific curve estimation with a deep network. Our method trains a lightweight deep network, DCE-Net, to estimate pixel-wise and high-order curves for dynamic range adjustment of a given image. The curve estimation is specially designed, considering pixel value range, monotonicity, and differentiability. Zero-DCE is appealing in its relaxed assumption on reference images, i.e., it does not require any paired or even unpaired data during training. This is achieved through a set of carefully formulated non-reference loss functions, which implicitly measure the enhancement quality and drive the learning of the network. Despite its simplicity, we show that it generalizes well to diverse lighting conditions. Our method is efficient as image enhancement can be achieved by an intuitive and simple nonlinear curve mapping. We further present an accelerated and light version of Zero-DCE, called Zero-DCE++, that takes advantage of a tiny network with just 10K parameters. Zero-DCE++ has a fast inference speed (1000/11 FPS on a single GPU/CPU for an image of size 1200*900*3) while keeping the enhancement performance of Zero-DCE. Extensive experiments on various benchmarks demonstrate the advantages of our method over state-of-the-art methods qualitatively and quantitatively. Furthermore, the potential benefits of our method to face detection in the dark are discussed. The source code will be made publicly available at https://li-chongyi.github.io/Proj_Zero-DCE++.html.

arXiv:2103.00686공식 코드 · Papers with Code

Accelerating Recommendation System Training by Leveraging Popular Choices

Recommender models are commonly used to suggest relevant items to a user for e-commerce and online advertisement-based applications. These models use massive embedding tables to store numerical representation of items' and users' categorical variables (memory intensive) and employ neural networks (compute intensive) to generate final recommendations. Training these large-scale recommendation models is evolving to require increasing data and compute resources. The highly parallel neural networks portion of these models can benefit from GPU acceleration however, large embedding tables often cannot fit in the limited-capacity GPU device memory. Hence, this paper deep dives into the semantics of training data and obtains insights about the feature access, transfer, and usage patterns of these models. We observe that, due to the popularity of certain inputs, the accesses to the embeddings are highly skewed with a few embedding entries being accessed up to 10000x more. This paper leverages this asymmetrical access pattern to offer a framework, called FAE, and proposes a hot-embedding aware data layout for training recommender models. This layout utilizes the scarce GPU memory for storing the highly accessed embeddings, thus reduces the data transfers from CPU to GPU. At the same time, FAE engages the GPU to accelerate the executions of these hot embedding entries. Experiments on production-scale recommendation models with real datasets show that FAE reduces the overall training time by 2.3x and 1.52x in comparison to XDL CPU-only and XDL CPU-GPU execution while maintaining baseline accuracy

arXiv:2102.12985

A Novel Framework for Neural Architecture Search in the Hill Climbing Domain

Neural networks have now long been used for solving complex problems of image domain, yet designing the same needs manual expertise. Furthermore, techniques for automatically generating a suitable deep learning architecture for a given dataset have frequently made use of reinforcement learning and evolutionary methods which take extensive computational resources and time. We propose a new framework for neural architecture search based on a hill-climbing procedure using morphism operators that makes use of a novel gradient update scheme. The update is based on the aging of neural network layers and results in the reduction in the overall training time. This technique can search in a broader search space which subsequently yields competitive results. We achieve a 4.96% error rate on the CIFAR-10 dataset in 19.4 hours of a single GPU training.

arXiv:2102.12178공식 코드 · Papers with Code

Learning to Generate Wasserstein Barycenters

Optimal transport is a notoriously difficult problem to solve numerically, with current approaches often remaining intractable for very large scale applications such as those encountered in machine learning. Wasserstein barycenters -- the problem of finding measures in-between given input measures in the optimal transport sense -- is even more computationally demanding as it requires to solve an optimization problem involving optimal transport distances. By training a deep convolutional neural network, we improve by a factor of 60 the computational speed of Wasserstein barycenters over the fastest state-of-the-art approach on the GPU, resulting in milliseconds computational times on $512\times512$ regular grids. We show that our network, trained on Wasserstein barycenters of pairs of measures, generalizes well to the problem of finding Wasserstein barycenters of more than two measures. We demonstrate the efficiency of our approach for computing barycenters of sketches and transferring colors between multiple images.

arXiv:2102.12103

FIXAR: A Fixed-Point Deep Reinforcement Learning Platform with Quantization-Aware Training and Adaptive Parallelism

In this paper, we present a deep reinforcement learning platform named FIXAR which employs fixed-point data types and arithmetic units for the first time using a SW/HW co-design approach. Starting from 32-bit fixed-point data, Quantization-Aware Training (QAT) reduces its data precision based on the range of activations and performs retraining to minimize the reward degradation. FIXAR proposes the adaptive array processing core composed of configurable processing elements to support both intra-layer parallelism and intra-batch parallelism for high-throughput inference and training. Finally, FIXAR was implemented on Xilinx U50 and achieves 25293.3 inferences per second (IPS) training throughput and 2638.0 IPS/W accelerator efficiency, which is 2.7 times faster and 15.4 times more energy efficient than those of the CPU-GPU platform without any accuracy degradation.

arXiv:2102.12459공식 코드 · Papers with Code

When Attention Meets Fast Recurrence: Training Language Models with Reduced Compute

Large language models have become increasingly difficult to train because of the growing computation time and cost. In this work, we present SRU++, a highly-efficient architecture that combines fast recurrence and attention for sequence modeling. SRU++ exhibits strong modeling capacity and training efficiency. On standard language modeling tasks such as Enwik8, Wiki-103 and Billion Word datasets, our model obtains better bits-per-character and perplexity while using 3x-10x less training cost compared to top-performing Transformer models. For instance, our model achieves a state-of-the-art result on the Enwik8 dataset using 1.6 days of training on an 8-GPU machine. We further demonstrate that SRU++ requires minimal attention for near state-of-the-art performance. Our results suggest jointly leveraging fast recurrence with little attention as a promising direction for accelerating model training and inference.

arXiv:2102.11535공식 코드 · Papers with Code

Neural Architecture Search on ImageNet in Four GPU Hours: A Theoretically Inspired Perspective

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.

arXiv:2102.11026

High-order Differentiable Autoencoder for Nonlinear Model Reduction

This paper provides a new avenue for exploiting deep neural networks to improve physics-based simulation. Specifically, we integrate the classic Lagrangian mechanics with a deep autoencoder to accelerate elastic simulation of deformable solids. Due to the inertia effect, the dynamic equilibrium cannot be established without evaluating the second-order derivatives of the deep autoencoder network. This is beyond the capability of off-the-shelf automatic differentiation packages and algorithms, which mainly focus on the gradient evaluation. Solving the nonlinear force equilibrium is even more challenging if the standard Newton's method is to be used. This is because we need to compute a third-order derivative of the network to obtain the variational Hessian. We attack those difficulties by exploiting complex-step finite difference, coupled with reverse automatic differentiation. This strategy allows us to enjoy the convenience and accuracy of complex-step finite difference and in the meantime, to deploy complex-value perturbations as collectively as possible to save excessive network passes. With a GPU-based implementation, we are able to wield deep autoencoders (e.g., $10+$ layers) with a relatively high-dimension latent space in real-time. Along this pipeline, we also design a sampling network and a weighting network to enable \emph{weight-varying} Cubature integration in order to incorporate nonlinearity in the model reduction. We believe this work will inspire and benefit future research efforts in nonlinearly reduced physical simulation problems.

arXiv:2102.10424

GIST: Distributed Training for Large-Scale Graph Convolutional Networks

The graph convolutional network (GCN) is a go-to solution for machine learning on graphs, but its training is notoriously difficult to scale both in terms of graph size and the number of model parameters. Although some work has explored training on large-scale graphs (e.g., GraphSAGE, ClusterGCN, etc.), we pioneer efficient training of large-scale GCN models (i.e., ultra-wide, overparameterized models) with the proposal of a novel, distributed training framework. Our proposed training methodology, called GIST, disjointly partitions the parameters of a GCN model into several, smaller sub-GCNs that are trained independently and in parallel. In addition to being compatible with all GCN architectures and existing sampling techniques for efficient GCN training, GIST i) improves model performance, ii) scales to training on arbitrarily large graphs, iii) decreases wall-clock training time, and iv) enables the training of markedly overparameterized GCN models. Remarkably, with GIST, we train an astonishgly-wide 32,768-dimensional GraphSAGE model, which exceeds the capacity of a single GPU by a factor of 8x, to SOTA performance on the Amazon2M dataset.

논문 ID: document/9293017

GEM: Online Globally consistent dense elevation mapping for unstructured terrain

Online dense mapping gives a representation of the unstructured terrain, which is indispensable for safe robotic motion planning. In this article, we propose such an elevation mapping system, namely GEM, to generate a dense local elevation map in constant real time for fast responsive local planning, and maintain a globally consistent dense map for path routing at the same time. We model the global elevation map as a collection of submaps. When the trajectory estimation of the robot is corrected by simultaneous localization and mapping (SLAM), only relative poses between submaps are updated without rebuilding the submap. As a result, this deformable global dense map representation is able to keep the global consistency online. Besides, we accelerate the local mapping by integrating traversability analysis into the mapping system to save the computation cost by obstacle awareness. The system is implemented by CPU-GPU coordinated processing to guarantee constant real-time performance for in-time handling of dynamic obstacles. Substantial experimental results on both simulated and real-world data set validate the efficiency and effectiveness of GEM.

arXiv:2102.09964공식 코드 · Papers with Code

Temporal Gaussian Process Regression in Logarithmic Time

The aim of this article is to present a novel parallelization method for temporal Gaussian process (GP) regression problems. The method allows for solving GP regression problems in logarithmic O(log N) time, where N is the number of time steps. Our approach uses the state-space representation of GPs which in its original form allows for linear O(N) time GP regression by leveraging the Kalman filtering and smoothing methods. By using a recently proposed parallelization method for Bayesian filters and smoothers, we are able to reduce the linear computational complexity of the temporal GP regression problems into logarithmic span complexity. This ensures logarithmic time complexity when run on parallel hardware such as a graphics processing unit (GPU). We experimentally demonstrate the computational benefits on simulated and real datasets via our open-source implementation leveraging the GPflow framework.

arXiv:2102.08473공식 코드 · Papers with Code

COCO-LM: Correcting and Contrasting Text Sequences for Language Model Pretraining

We present a self-supervised learning framework, COCO-LM, that pretrains Language Models by COrrecting and COntrasting corrupted text sequences. Following ELECTRA-style pretraining, COCO-LM employs an auxiliary language model to corrupt text sequences, upon which it constructs two new tasks for pretraining the main model. The first token-level task, Corrective Language Modeling, is to detect and correct tokens replaced by the auxiliary model, in order to better capture token-level semantics. The second sequence-level task, Sequence Contrastive Learning, is to align text sequences originated from the same source input while ensuring uniformity in the representation space. Experiments on GLUE and SQuAD demonstrate that COCO-LM not only outperforms recent state-of-the-art pretrained models in accuracy, but also improves pretraining efficiency. It achieves the MNLI accuracy of ELECTRA with 50% of its pretraining GPU hours. With the same pretraining steps of standard base/large-sized models, COCO-LM outperforms the previous best models by 1+ GLUE average points.

arXiv:2102.08099공식 코드 · Papers with Code

EPE-NAS: Efficient Performance Estimation Without Training for Neural Architecture Search

Neural Architecture Search (NAS) has shown excellent results in designing architectures for computer vision problems. NAS alleviates the need for human-defined settings by automating architecture design and engineering. However, NAS methods tend to be slow, as they require large amounts of GPU computation. This bottleneck is mainly due to the performance estimation strategy, which requires the evaluation of the generated architectures, mainly by training them, to update the sampler method. In this paper, we propose EPE-NAS, an efficient performance estimation strategy, that mitigates the problem of evaluating networks, by scoring untrained networks and creating a correlation with their trained performance. We perform this process by looking at intra and inter-class correlations of an untrained network. We show that EPE-NAS can produce a robust correlation and that by incorporating it into a simple random sampling strategy, we are able to search for competitive networks, without requiring any training, in a matter of seconds using a single GPU. Moreover, EPE-NAS is agnostic to the search method, since it focuses on the evaluation of untrained networks, making it easy to integrate into almost any NAS method.