PyTorchKR Papers

최신 논문

총 5,575편
태그·기간 등 상세 필터

카테고리 탐색

로그인 후 이 태그 구독

arXiv:1911.05878

Scientific Image Restoration Anywhere

The use of deep learning models within scientific experimental facilities frequently requires low-latency inference, so that, for example, quality control operations can be performed while data are being collected. Edge computing devices can be useful in this context, as their low cost and compact form factor permit them to be co-located with the experimental apparatus. Can such devices, with their limited resources, can perform neural network feed-forward computations efficiently and effectively? We explore this question by evaluating the performance and accuracy of a scientific image restoration model, for which both model input and output are images, on edge computing devices. Specifically, we evaluate deployments of TomoGAN, an image-denoising model based on generative adversarial networks developed for low-dose x-ray imaging, on the Google Edge TPU and NVIDIA Jetson. We adapt TomoGAN for edge execution, evaluate model inference performance, and propose methods to address the accuracy drop caused by model quantization. We show that these edge computing devices can deliver accuracy comparable to that of a full-fledged CPU or GPU model, at speeds that are more than adequate for use in the intended deployments, denoising a 1024 x 1024 image in less than a second. Our experiments also show that the Edge TPU models can provide 3x faster inference response than a CPU-based model and 1.5x faster than an edge GPU-based model. This combination of high speed and low cost permits image restoration anywhere.

arXiv:1911.04975공식 코드 · Papers with Code

word2ket: Space-efficient Word Embeddings inspired by Quantum Entanglement

Deep learning natural language processing models often use vector word embeddings, such as word2vec or GloVe, to represent words. A discrete sequence of words can be much more easily integrated with downstream neural layers if it is represented as a sequence of continuous vectors. Also, semantic relationships between words, learned from a text corpus, can be encoded in the relative configurations of the embedding vectors. However, storing and accessing embedding vectors for all words in a dictionary requires large amount of space, and may stain systems with limited GPU memory. Here, we used approaches inspired by quantum computing to propose two related methods, {\em word2ket} and {\em word2ketXS}, for storing word embedding matrix during training and inference in a highly efficient way. Our approach achieves a hundred-fold or more reduction in the space required to store the embeddings with almost no relative drop in accuracy in practical natural language processing tasks.

arXiv:1911.04610

XPipe: Efficient Pipeline Model Parallelism for Multi-GPU DNN Training

We propose XPipe, an efficient asynchronous pipeline model parallelism approach for multi-GPU DNN training. XPipe is designed to use multiple GPUs to concurrently and continuously train different parts of a DNN model. To improve GPU utilization and achieve high throughput, it splits a mini-batch into a set of micro-batches. It allows the overlapping of the pipelines of multiple micro-batches, including those belonging to different mini-batches. Most importantly, the novel weight prediction strategy adopted by XPipe enables it to effectively address the weight inconsistency and staleness issues incurred by the asynchronous pipeline parallelism. As a result, XPipe incorporates the advantages of both synchronous and asynchronous pipeline model parallelism approaches. Concretely, it can achieve very comparable (even slightly better) model accuracy as its synchronous counterpart while obtaining higher throughput than it. Experimental results show that XPipe outperforms other state-of-the-art synchronous and asynchronous model parallelism approaches.

arXiv:1911.04477공식 코드 · Papers with Code

A Computing Kernel for Network Binarization on PyTorch

Deep Neural Networks have now achieved state-of-the-art results in a wide range of tasks including image classification, object detection and so on. However, they are both computation consuming and memory intensive, making them difficult to deploy on low-power devices. Network binarization is one of the existing effective techniques for model compression and acceleration, but there is no computing kernel yet to support it on PyTorch. In this paper we developed a computing kernel supporting 1-bit xnor and bitcount computation on PyTorch. Experimental results show that our kernel could accelerate the inference of the binarized neural network by 3 times in GPU and by 4.5 times in CPU compared with the control group.

arXiv:1911.03451

Enabling Highly Efficient Capsule Networks Processing Through A PIM-Based Architecture Design

In recent years, the CNNs have achieved great successes in the image processing tasks, e.g., image recognition and object detection. Unfortunately, traditional CNN's classification is found to be easily misled by increasingly complex image features due to the usage of pooling operations, hence unable to preserve accurate position and pose information of the objects. To address this challenge, a novel neural network structure called Capsule Network has been proposed, which introduces equivariance through capsules to significantly enhance the learning ability for image segmentation and object detection. Due to its requirement of performing a high volume of matrix operations, CapsNets have been generally accelerated on modern GPU platforms that provide highly optimized software library for common deep learning tasks. However, based on our performance characterization on modern GPUs, CapsNets exhibit low efficiency due to the special program and execution features of their routing procedure, including massive unshareable intermediate variables and intensive synchronizations, which are very difficult to optimize at software level. To address these challenges, we propose a hybrid computing architecture design named \textit{PIM-CapsNet}. It preserves GPU's on-chip computing capability for accelerating CNN types of layers in CapsNet, while pipelining with an off-chip in-memory acceleration solution that effectively tackles routing procedure's inefficiency by leveraging the processing-in-memory capability of today's 3D stacked memory. Using routing procedure's inherent parallellization feature, our design enables hierarchical improvements on CapsNet inference efficiency through minimizing data movement and maximizing parallel processing in memory.

arXiv:1911.03667공식 코드 · Papers with Code

Factored Latent-Dynamic Conditional Random Fields for Single and Multi-label Sequence Modeling

Conditional Random Fields (CRF) are frequently applied for labeling and segmenting sequence data. Morency et al. (2007) introduced hidden state variables in a labeled CRF structure in order to model the latent dynamics within class labels, thus improving the labeling performance. Such a model is known as Latent-Dynamic CRF (LDCRF). We present Factored LDCRF (FLDCRF), a structure that allows multiple latent dynamics of the class labels to interact with each other. Including such latent-dynamic interactions leads to improved labeling performance on single-label and multi-label sequence modeling tasks. We apply our FLDCRF models on two single-label (one nested cross-validation) and one multi-label sequence tagging (nested cross-validation) experiments across two different datasets - UCI gesture phase data and UCI opportunity data. FLDCRF outperforms all state-of-the-art sequence models, i.e., CRF, LDCRF, LSTM, LSTM-CRF, Factorial CRF, Coupled CRF and a multi-label LSTM model in all our experiments. In addition, LSTM based models display inconsistent performance across validation and test data, and pose diffculty to select models on validation data during our experiments. FLDCRF offers easier model selection, consistency across validation and test performance and lucid model intuition. FLDCRF is also much faster to train compared to LSTM, even without a GPU. FLDCRF outshines the best LSTM model by ~4% on a single-label task on UCI gesture phase data and outperforms LSTM performance by ~2% on average across nested cross-validation test sets on the multi-label sequence tagging experiment on UCI opportunity data. The idea of FLDCRF can be extended to joint (multi-agent interactions) and heterogeneous (discrete and continuous) state space models.

arXiv:1911.02733

Porous Lattice-based Transformer Encoder for Chinese NER

Incorporating lattices into character-level Chinese named entity recognition is an effective method to exploit explicit word information. Recent works extend recurrent and convolutional neural networks to model lattice inputs. However, due to the DAG structure or the variable-sized potential word set for lattice inputs, these models prevent the convenient use of batched computation, resulting in serious inefficient. In this paper, we propose a porous lattice-based transformer encoder for Chinese named entity recognition, which is capable to better exploit the GPU parallelism and batch the computation owing to the mask mechanism in transformer. We first investigate the lattice-aware self-attention coupled with relative position representations to explore effective word information in the lattice structure. Besides, to strengthen the local dependencies among neighboring tokens, we propose a novel porous structure during self-attentional computation processing, in which every two non-neighboring tokens are connected through a shared pivot node. Experimental results on four datasets show that our model performs up to 9.47 times faster than state-of-the-art models, while is roughly on a par with its performance. The source code of this paper can be obtained from https://github.com/xxx/xxx.

arXiv:1911.00632

FDDWNet: A Lightweight Convolutional Neural Network for Real-time Sementic Segmentation

This paper introduces a lightweight convolutional neural network, called FDDWNet, for real-time accurate semantic segmentation. In contrast to recent advances of lightweight networks that prefer to utilize shallow structure, FDDWNet makes an effort to design more deeper network architecture, while maintains faster inference speed and higher segmentation accuracy. Our network uses factorized dilated depth-wise separable convolutions (FDDWC) to learn feature representations from different scale receptive fields with fewer model parameters. Additionally, FDDWNet has multiple branches of skipped connections to gather context cues from intermediate convolution layers. The experiments show that FDDWNet only has 0.8M model size, while achieves 60 FPS running speed on a single RTX 2080Ti GPU with a 1024x512 input image. The comprehensive experiments demonstrate that our model achieves state-of-the-art results in terms of available speed and accuracy trade-off on CityScapes and CamVid datasets.

arXiv:1911.01258

SHARP: An Adaptable, Energy-Efficient Accelerator for Recurrent Neural Network

The effectiveness of Recurrent Neural Networks (RNNs) for tasks such as Automatic Speech Recognition has fostered interest in RNN inference acceleration. Due to the recurrent nature and data dependencies of RNN computations, prior work has designed customized architectures specifically tailored to the computation pattern of RNN, getting high computation efficiency for certain chosen model sizes. However, given that the dimensionality of RNNs varies a lot for different tasks, it is crucial to generalize this efficiency to diverse configurations. In this work, we identify adaptiveness as a key feature that is missing from today's RNN accelerators. In particular, we first show the problem of low resource-utilization and low adaptiveness for the state-of-the-art RNN implementations on GPU, FPGA and ASIC architectures. To solve these issues, we propose an intelligent tiled-based dispatching mechanism for increasing the adaptiveness of RNN computation, in order to efficiently handle the data dependencies. To do so, we propose Sharp as a hardware accelerator, which pipelines RNN computation using an effective scheduling scheme to hide most of the dependent serialization. Furthermore, Sharp employs dynamic reconfigurable architecture to adapt to the model's characteristics. Sharp achieves 2x, 2.8x, and 82x speedups on average, considering different RNN models and resource budgets, compared to the state-of-the-art ASIC, FPGA, and GPU implementations, respectively. Furthermore, we provide significant energy-reduction with respect to the previous solutions, due to the low power dissipation of Sharp (321 GFLOPS/Watt).

arXiv:1911.00119

ALERT: Accurate Learning for Energy and Timeliness

An increasing number of software applications incorporate runtime Deep Neural Networks (DNNs) to process sensor data and return inference results to humans. Effective deployment of DNNs in these interactive scenarios requires meeting latency and accuracy constraints while minimizing energy, a problem exacerbated by common system dynamics. Prior approaches handle dynamics through either (1) system-oblivious DNN adaptation, which adjusts DNN latency/accuracy tradeoffs, or (2) application-oblivious system adaptation, which adjusts resources to change latency/energy tradeoffs. In contrast, this paper improves on the state-of-the-art by coordinating application- and system-level adaptation. ALERT, our runtime scheduler, uses a probabilistic model to detect environmental volatility and then simultaneously select both a DNN and a system resource configuration to meet latency, accuracy, and energy constraints. We evaluate ALERT on CPU and GPU platforms for image and speech tasks in dynamic environments. ALERT's holistic approach achieves more than 13% energy reduction, and 27% error reduction over prior approaches that adapt solely at the application or system level. Furthermore, ALERT incurs only 3% more energy consumption and 2% higher DNN-inference error than an oracle scheme with perfect application and system knowledge.

arXiv:1911.00081공식 코드 · Papers with Code

Scaling structural learning with NO-BEARS to infer causal transcriptome networks

Constructing gene regulatory networks is a critical step in revealing disease mechanisms from transcriptomic data. In this work, we present NO-BEARS, a novel algorithm for estimating gene regulatory networks. The NO-BEARS algorithm is built on the basis of the NOTEARS algorithm with two improvements. First, we propose a new constraint and its fast approximation to reduce the computational cost of the NO-TEARS algorithm. Next, we introduce a polynomial regression loss to handle non-linearity in gene expressions. Our implementation utilizes modern GPU computation that can decrease the time of hours-long CPU computation to seconds. Using synthetic data, we demonstrate improved performance, both in processing time and accuracy, on inferring gene regulatory networks from gene expression data.

arXiv:1911.00357공식 코드 · Papers with Code

DD-PPO: Learning Near-Perfect PointGoal Navigators from 2.5 Billion Frames

We present Decentralized Distributed Proximal Policy Optimization (DD-PPO), a method for distributed reinforcement learning in resource-intensive simulated environments. DD-PPO is distributed (uses multiple machines), decentralized (lacks a centralized server), and synchronous (no computation is ever stale), making it conceptually simple and easy to implement. In our experiments on training virtual robots to navigate in Habitat-Sim, DD-PPO exhibits near-linear scaling -- achieving a speedup of 107x on 128 GPUs over a serial implementation. We leverage this scaling to train an agent for 2.5 Billion steps of experience (the equivalent of 80 years of human experience) -- over 6 months of GPU-time training in under 3 days of wall-clock time with 64 GPUs. This massive-scale training not only sets the state of art on Habitat Autonomous Navigation Challenge 2019, but essentially solves the task --near-perfect autonomous navigation in an unseen environment without access to a map, directly from an RGB-D camera and a GPS+Compass sensor. Fortuitously, error vs computation exhibits a power-law-like distribution; thus, 90% of peak performance is obtained relatively early (at 100 million steps) and relatively cheaply (under 1 day with 8 GPUs). Finally, we show that the scene understanding and navigation policies learned can be transferred to other navigation tasks -- the analog of ImageNet pre-training + task-specific fine-tuning for embodied AI. Our model outperforms ImageNet pre-trained CNNs on these transfer tasks and can serve as a universal resource (all models and code are publicly available).

arXiv:1910.13598공식 코드 · Papers with Code

Local SGD with Periodic Averaging: Tighter Analysis and Adaptive Synchronization

Communication overhead is one of the key challenges that hinders the scalability of distributed optimization algorithms. In this paper, we study local distributed SGD, where data is partitioned among computation nodes, and the computation nodes perform local updates with periodically exchanging the model among the workers to perform averaging. While local SGD is empirically shown to provide promising results, a theoretical understanding of its performance remains open. We strengthen convergence analysis for local SGD, and show that local SGD can be far less expensive and applied far more generally than current theory suggests. Specifically, we show that for loss functions that satisfy the Polyak-{\L}ojasiewicz condition, $O((pT)^{1/3})$ rounds of communication suffice to achieve a linear speed up, that is, an error of $O(1/pT)$, where $T$ is the total number of model updates at each worker. This is in contrast with previous work which required higher number of communication rounds, as well as was limited to strongly convex loss functions, for a similar asymptotic performance. We also develop an adaptive synchronization scheme that provides a general condition for linear speed up. Finally, we validate the theory with experimental results, running over AWS EC2 clouds and an internal GPU cluster.

arXiv:1910.11480

Parallel WaveGAN: A fast waveform generation model based on generative adversarial networks with multi-resolution spectrogram

We propose Parallel WaveGAN, a distillation-free, fast, and small-footprint waveform generation method using a generative adversarial network. In the proposed method, a non-autoregressive WaveNet is trained by jointly optimizing multi-resolution spectrogram and adversarial loss functions, which can effectively capture the time-frequency distribution of the realistic speech waveform. As our method does not require density distillation used in the conventional teacher-student framework, the entire model can be easily trained. Furthermore, our model is able to generate high-fidelity speech even with its compact architecture. In particular, the proposed Parallel WaveGAN has only 1.44 M parameters and can generate 24 kHz speech waveform 28.68 times faster than real-time on a single GPU environment. Perceptual listening test results verify that our proposed method achieves 4.16 mean opinion score within a Transformer-based text-to-speech framework, which is comparative to the best distillation-based Parallel WaveNet system.

arXiv:1910.11414

Reconstruction of Undersampled 3D Non-Cartesian Image-Based Navigators for Coronary MRA Using an Unrolled Deep Learning Model

Purpose: To rapidly reconstruct undersampled 3D non-Cartesian image-based navigators (iNAVs) using an unrolled deep learning (DL) model for non-rigid motion correction in coronary magnetic resonance angiography (CMRA). Methods: An unrolled network is trained to reconstruct beat-to-beat 3D iNAVs acquired as part of a CMRA sequence. The unrolled model incorporates a non-uniform FFT operator to perform the data consistency operation, and the regularization term is learned by a convolutional neural network (CNN) based on the proximal gradient descent algorithm. The training set includes 6,000 3D iNAVs acquired from 7 different subjects and 11 scans using a variable-density (VD) cones trajectory. For testing, 3D iNAVs from 4 additional subjects are reconstructed using the unrolled model. To validate reconstruction accuracy, global and localized motion estimates from DL model-based 3D iNAVs are compared with those extracted from 3D iNAVs reconstructed with $\textit{l}_{1}$-ESPIRiT. Then, the high-resolution coronary MRA images motion corrected with autofocusing using the $\textit{l}_{1}$-ESPIRiT and DL model-based 3D iNAVs are assessed for differences. Results: 3D iNAVs reconstructed using the DL model-based approach and conventional $\textit{l}_{1}$-ESPIRiT generate similar global and localized motion estimates and provide equivalent coronary image quality. Reconstruction with the unrolled network completes in a fraction of the time compared to CPU and GPU implementations of $\textit{l}_{1}$-ESPIRiT (20x and 3x speed increases, respectively). Conclusion: We have developed a deep neural network architecture to reconstruct undersampled 3D non-Cartesian VD cones iNAVs. Our approach decreases reconstruction time for 3D iNAVs, while preserving the accuracy of non-rigid motion information offered by them for correction.

arXiv:1910.11818

Real-time Memory Efficient Large-pose Face Alignment via Deep Evolutionary Network

There is an urgent need to apply face alignment in a memory-efficient and real-time manner due to the recent explosion of face recognition applications. However, impact factors such as large pose variation and computational inefficiency, still hinder its broad implementation. To this end, we propose a computationally efficient deep evolutionary model integrated with 3D Diffusion Heap Maps (DHM). First, we introduce a sparse 3D DHM to assist the initial modeling process under extreme pose conditions. Afterward, a simple and effective CNN feature is extracted and fed to Recurrent Neural Network (RNN) for evolutionary learning. To accelerate the model, we propose an efficient network structure to accelerate the evolutionary learning process through a factorization strategy. Extensive experiments on three popular alignment databases demonstrate the advantage of the proposed models over the state-of-the-art, especially under large-pose conditions. Notably, the computational speed of our model is 6 times faster than the state-of-the-art on CPU and 14 times on GPU. We also discuss and analyze the limitations of our models and future research work.

arXiv:1910.10892공식 코드 · Papers with Code

Fast and Differentiable Message Passing on Pairwise Markov Random Fields

Despite the availability of many Markov Random Field (MRF) optimization algorithms, their widespread usage is currently limited due to imperfect MRF modelling arising from hand-crafted model parameters and the selection of inferior inference algorithm. In addition to differentiability, the two main aspects that enable learning these model parameters are the forward and backward propagation time of the MRF optimization algorithm and its inference capabilities. In this work, we introduce two fast and differentiable message passing algorithms, namely, Iterative Semi-Global Matching Revised (ISGMR) and Parallel Tree-Reweighted Message Passing (TRWP) which are greatly sped up on a GPU by exploiting massive parallelism. Specifically, ISGMR is an iterative and revised version of the standard SGM for general pairwise MRFs with improved optimization effectiveness, and TRWP is a highly parallel version of Sequential TRW (TRWS) for faster optimization. Our experiments on the standard stereo and denoising benchmarks demonstrated that ISGMR and TRWP achieve much lower energies than SGM and Mean-Field (MF), and TRWP is two orders of magnitude faster than TRWS without losing effectiveness in optimization. We further demonstrated the effectiveness of our algorithms on end-to-end learning for semantic segmentation. Notably, our CUDA implementations are at least $7$ and $700$ times faster than PyTorch GPU implementations for forward and backward propagation respectively, enabling efficient end-to-end learning with message passing.

논문 ID: document/8867874

Octave Deep Plane-Sweeping Network: Reducing Spatial Redundancy for Learning-Based Plane-Sweeping Stereo

In this paper, we propose the octave deep plane-sweeping network (OctDPSNet). OctDPSNet is a novel learning-based plane-sweeping stereo, which drastically reduces the required GPU memory and computation time while achieving a state-of-the-art depth estimation accuracy. Inspired by octave convolution, we divide image features into high and low spatial frequency features, and two cost volumes are generated from these using our proposed plane-sweeping module. To reduce spatial redundancy, the resolution of the cost volume from the low spatial frequency features is set to half that of the high spatial frequency features, which enables the memory consumption and computational cost to be reduced. After refinement, the two cost volumes are integrated into a final cost volume through our proposed pixel-wise “squeeze-and-excitation” based attention mechanism, and the depth maps are estimated from the final cost volume. We evaluate the proposed model on five datasets: SUN3D, RGB-D SLAM, MVS, Scenes11, and ETH3D. Our model outperforms previous methods on five datasets while drastically reducing the memory consumption and computational cost. Our source code is available at https://github.com/matsuren/octDPSNet.

arXiv:1910.11127

Reversible designs for extreme memory cost reduction of CNN training

Training Convolutional Neural Networks (CNN) is a resource intensive task that requires specialized hardware for efficient computation. One of the most limiting bottleneck of CNN training is the memory cost associated with storing the activation values of hidden layers needed for the computation of the weights gradient during the backward pass of the backpropagation algorithm. Recently, reversible architectures have been proposed to reduce the memory cost of training large CNN by reconstructing the input activation values of hidden layers from their output during the backward pass, circumventing the need to accumulate these activations in memory during the forward pass. In this paper, we push this idea to the extreme and analyze reversible network designs yielding minimal training memory footprint. We investigate the propagation of numerical errors in long chains of invertible operations and analyze their effect on training. We introduce the notion of pixel-wise memory cost to characterize the memory footprint of model training, and propose a new model architecture able to efficiently train arbitrarily deep neural networks with a minimum memory cost of 352 bytes per input pixel. This new kind of architecture enables training large neural networks on very limited memory, opening the door for neural network training on embedded devices or non-specialized hardware. For instance, we demonstrate training of our model to 93.3% accuracy on the CIFAR10 dataset within 67 minutes on a low-end Nvidia GTX750 GPU with only 1GB of memory.

arXiv:1910.10398

Random 2.5D U-net for Fully 3D Segmentation

Convolutional neural networks are state-of-the-art for various segmentation tasks. While for 2D images these networks are also computationally efficient, 3D convolutions have huge storage requirements and therefore, end-to-end training is limited by GPU memory and data size. To overcome this issue, we introduce a network structure for volumetric data without 3D convolution layers. The main idea is to include projections from different directions to transform the volumetric data to a sequence of images, where each image contains information of the full data. We then apply 2D convolutions to these projection images and lift them again to volumetric data using a trainable reconstruction algorithm. The proposed architecture can be applied end-to-end to very large data volumes without cropping or sliding-window techniques. For a tested sparse binary segmentation task, it outperforms already known standard approaches and is more resistant to generation of artefacts.