PyTorchKR Papers

최신 논문

총 5,575편
태그·기간 등 상세 필터

카테고리 탐색

로그인 후 이 태그 구독

arXiv:2007.01192

Digit Image Recognition Using an Ensemble of One-Versus-All Deep Network Classifiers

In multiclass deep network classifiers, the burden of classifying samples of different classes is put on a single classifier. As the result the optimum classification accuracy is not obtained. Also training times are large due to running the CNN training on single CPU/GPU. However it is known that using ensembles of classifiers increases the performance. Also, the training times can be reduced by running each member of the ensemble on a separate processor. Ensemble learning has been used in the past for traditional methods to a varying extent and is a hot topic. With the advent of deep learning, ensemble learning has been applied to the former as well. However, an area which is unexplored and has potential is One-Versus-All (OVA) deep ensemble learning. In this paper we explore it and show that by using OVA ensembles of deep networks, improvements in performance of deep networks can be obtained. As shown in this paper, the classification capability of deep networks can be further increased by using an ensemble of binary classification (OVA) deep networks. We implement a novel technique for the case of digit image recognition and test and evaluate it on the same. In the proposed approach, a single OVA deep network classifier is dedicated to each category. Subsequently, OVA deep network ensembles have been investigated. Every network in an ensemble has been trained by an OVA training technique using the Stochastic Gradient Descent with Momentum Algorithm (SGDMA). For classification of a test sample, the sample is presented to each network in the ensemble. After prediction score voting, the network with the largest score is assumed to have classified the sample. The experimentation has been done on the MNIST digit dataset, the USPS+ digit dataset, and MATLAB digit image dataset. Our proposed technique outperforms the baseline on digit image recognition for all datasets.

arXiv:2007.00784공식 코드 · Papers with Code

Convolutional Neural Network Training with Distributed K-FAC

Training neural networks with many processors can reduce time-to-solution; however, it is challenging to maintain convergence and efficiency at large scales. The Kronecker-factored Approximate Curvature (K-FAC) was recently proposed as an approximation of the Fisher Information Matrix that can be used in natural gradient optimizers. We investigate here a scalable K-FAC design and its applicability in convolutional neural network (CNN) training at scale. We study optimization techniques such as layer-wise distribution strategies, inverse-free second-order gradient evaluation, and dynamic K-FAC update decoupling to reduce training time while preserving convergence. We use residual neural networks (ResNet) applied to the CIFAR-10 and ImageNet-1k datasets to evaluate the correctness and scalability of our K-FAC gradient preconditioner. With ResNet-50 on the ImageNet-1k dataset, our distributed K-FAC implementation converges to the 75.9% MLPerf baseline in 18-25% less time than does the classic stochastic gradient descent (SGD) optimizer across scales on a GPU cluster.

arXiv:2007.01127

Bidirectional Encoder Representations from Transformers (BERT): A sentiment analysis odyssey

The purpose of the study is to investigate the relative effectiveness of four different sentiment analysis techniques: (1) unsupervised lexicon-based model using Sent WordNet; (2) traditional supervised machine learning model using logistic regression; (3) supervised deep learning model using Long Short-Term Memory (LSTM); and, (4) advanced supervised deep learning models using Bidirectional Encoder Representations from Transformers (BERT). We use publicly available labeled corpora of 50,000 movie reviews originally posted on internet movie database (IMDB) for analysis using Sent WordNet lexicon, logistic regression, LSTM, and BERT. The first three models were run on CPU based system whereas BERT was run on GPU based system. The sentiment classification performance was evaluated based on accuracy, precision, recall, and F1 score. The study puts forth two key insights: (1) relative efficacy of four highly advanced and widely used sentiment analysis techniques; (2) undisputed superiority of pre-trained advanced supervised deep learning BERT model in sentiment analysis from text data. This study provides professionals in analytics industry and academicians working on text analysis key insight regarding comparative classification performance evaluation of key sentiment analysis techniques, including the recently developed BERT. This is the first research endeavor to compare the advanced pre-trained supervised deep learning model of BERT vis-\`a-vis other sentiment analysis models of LSTM, logistic regression, and Sent WordNet.

arXiv:2007.01038공식 코드 · Papers with Code

Beyond Signal Propagation: Is Feature Diversity Necessary in Deep Neural Network Initialization?

Deep neural networks are typically initialized with random weights, with variances chosen to facilitate signal propagation and stable gradients. It is also believed that diversity of features is an important property of these initializations. We construct a deep convolutional network with identical features by initializing almost all the weights to $0$. The architecture also enables perfect signal propagation and stable gradients, and achieves high accuracy on standard benchmarks. This indicates that random, diverse initializations are \textit{not} necessary for training neural networks. An essential element in training this network is a mechanism of symmetry breaking; we study this phenomenon and find that standard GPU operations, which are non-deterministic, can serve as a sufficient source of symmetry breaking to enable training.

arXiv:2007.01016

A Novel DNN Training Framework via Data Sampling and Multi-Task Optimization

Conventional DNN training paradigms typically rely on one training set and one validation set, obtained by partitioning an annotated dataset used for training, namely gross training set, in a certain way. The training set is used for training the model while the validation set is used to estimate the generalization performance of the trained model as the training proceeds to avoid over-fitting. There exist two major issues in this paradigm. Firstly, the validation set may hardly guarantee an unbiased estimate of generalization performance due to potential mismatching with test data. Secondly, training a DNN corresponds to solve a complex optimization problem, which is prone to getting trapped into inferior local optima and thus leads to undesired training results. To address these issues, we propose a novel DNN training framework. It generates multiple pairs of training and validation sets from the gross training set via random splitting, trains a DNN model of a pre-specified structure on each pair while making the useful knowledge (e.g., promising network parameters) obtained from one model training process to be transferred to other model training processes via multi-task optimization, and outputs the best, among all trained models, which has the overall best performance across the validation sets from all pairs. The knowledge transfer mechanism featured in this new framework can not only enhance training effectiveness by helping the model training process to escape from local optima but also improve on generalization performance via implicit regularization imposed on one model training process from other model training processes. We implement the proposed framework, parallelize the implementation on a GPU cluster, and apply it to train several widely used DNN models. Experimental results demonstrate the superiority of the proposed framework over the conventional training paradigm.

arXiv:2007.00183공식 코드 · Papers with Code

Whole-Word Segmental Speech Recognition with Acoustic Word Embeddings

Segmental models are sequence prediction models in which scores of hypotheses are based on entire variable-length segments of frames. We consider segmental models for whole-word ("acoustic-to-word") speech recognition, with the feature vectors defined using vector embeddings of segments. Such models are computationally challenging as the number of paths is proportional to the vocabulary size, which can be orders of magnitude larger than when using subword units like phones. We describe an efficient approach for end-to-end whole-word segmental models, with forward-backward and Viterbi decoding performed on a GPU and a simple segment scoring function that reduces space complexity. In addition, we investigate the use of pre-training via jointly trained acoustic word embeddings (AWEs) and acoustically grounded word embeddings (AGWEs) of written word labels. We find that word error rate can be reduced by a large margin by pre-training the acoustic segment representation with AWEs, and additional (smaller) gains can be obtained by pre-training the word prediction layer with AGWEs. Our final models improve over prior A2W models.

arXiv:2007.00639

End-to-End JPEG Decoding and Artifacts Suppression Using Heterogeneous Residual Convolutional Neural Network

Existing deep learning models separate JPEG artifacts suppression from the decoding protocol as independent task. In this work, we take one step forward to design a true end-to-end heterogeneous residual convolutional neural network (HR-CNN) with spectrum decomposition and heterogeneous reconstruction mechanism. Benefitting from the full CNN architecture and GPU acceleration, the proposed model considerably improves the reconstruction efficiency. Numerical experiments show that the overall reconstruction speed reaches to the same magnitude of the standard CPU JPEG decoding protocol, while both decoding and artifacts suppression are completed together. We formulate the JPEG artifacts suppression task as an interactive process of decoding and image detail reconstructions. A heterogeneous, fully convolutional, mechanism is proposed to particularly address the uncorrelated nature of different spectral channels. Directly starting from the JPEG code in k-space, the network first extracts the spectral samples channel by channel, and restores the spectral snapshots with expanded throughput. These intermediate snapshots are then heterogeneously decoded and merged into the pixel space image. A cascaded residual learning segment is designed to further enhance the image details. Experiments verify that the model achieves outstanding performance in JPEG artifacts suppression, while its full convolutional operations and elegant network structure offers higher computational efficiency for practical online usage compared with other deep learning models on this topic.

arXiv:2006.15218

Traditional and accelerated gradient descent for neural architecture search

In this paper we introduce two algorithms for neural architecture search (NASGD and NASAGD) following the theoretical work by two of the authors [5] which used the geometric structure of optimal transport to introduce the conceptual basis for new notions of traditional and accelerated gradient descent algorithms for the optimization of a function on a semi-discrete space. Our algorithms, which use the network morphism framework introduced in [2] as a baseline, can analyze forty times as many architectures as the hill climbing methods [2, 14] while using the same computational resources and time and achieving comparable levels of accuracy. For example, using NASGD on CIFAR-10, our method designs and trains networks with an error rate of 4.06 in only 12 hours on a single GPU.

arXiv:2006.15212공식 코드 · Papers with Code

Hybrid Models for Learning to Branch

A recent Graph Neural Network (GNN) approach for learning to branch has been shown to successfully reduce the running time of branch-and-bound algorithms for Mixed Integer Linear Programming (MILP). While the GNN relies on a GPU for inference, MILP solvers are purely CPU-based. This severely limits its application as many practitioners may not have access to high-end GPUs. In this work, we ask two key questions. First, in a more realistic setting where only a CPU is available, is the GNN model still competitive? Second, can we devise an alternate computationally inexpensive model that retains the predictive power of the GNN architecture? We answer the first question in the negative, and address the second question by proposing a new hybrid architecture for efficient branching on CPU machines. The proposed architecture combines the expressive power of GNNs with computationally inexpensive multi-layer perceptrons (MLP) for branching. We evaluate our methods on four classes of MILP problems, and show that they lead to up to 26% reduction in solver running time compared to state-of-the-art methods without a GPU, while extrapolating to harder problems than it was trained on. The code for this project is publicly available at https://github.com/pg2455/Hybrid-learn2branch.

arXiv:2006.15190공식 코드 · Papers with Code

Making DensePose fast and light

DensePose estimation task is a significant step forward for enhancing user experience computer vision applications ranging from augmented reality to cloth fitting. Existing neural network models capable of solving this task are heavily parameterized and a long way from being transferred to an embedded or mobile device. To enable Dense Pose inference on the end device with current models, one needs to support an expensive server-side infrastructure and have a stable internet connection. To make things worse, mobile and embedded devices do not always have a powerful GPU inside. In this work, we target the problem of redesigning the DensePose R-CNN model's architecture so that the final network retains most of its accuracy but becomes more light-weight and fast. To achieve that, we tested and incorporated many deep learning innovations from recent years, specifically performing an ablation study on 23 efficient backbone architectures, multiple two-stage detection pipeline modifications, and custom model quantization methods. As a result, we achieved $17\times$ model size reduction and $2\times$ latency improvement compared to the baseline model.

arXiv:2006.15373공식 코드 · Papers with Code

MTStereo 2.0: improved accuracy of stereo depth estimation withMax-trees

Efficient yet accurate extraction of depth from stereo image pairs is required by systems with low power resources, such as robotics and embedded systems. State-of-the-art stereo matching methods based on convolutional neural networks require intensive computations on GPUs and are difficult to deploy on embedded systems. In this paper, we propose a stereo matching method, called MTStereo 2.0, for limited-resource systems that require efficient and accurate depth estimation. It is based on a Max-tree hierarchical representation of image pairs, which we use to identify matching regions along image scan-lines. The method includes a cost function that considers similarity of region contextual information based on the Max-trees and a disparity border preserving cost aggregation approach. MTStereo 2.0 improves on its predecessor MTStereo 1.0 as it a) deploys a more robust cost function, b) performs more thorough detection of incorrect matches, c) computes disparity maps with pixel-level rather than node-level precision. MTStereo provides accurate sparse and semi-dense depth estimation and does not require intensive GPU computations like methods based on CNNs. Thus it can run on embedded and robotics devices with low-power requirements. We tested the proposed approach on several benchmark data sets, namely KITTI 2015, Driving, FlyingThings3D, Middlebury 2014, Monkaa and the TrimBot2020 garden data sets, and achieved competitive accuracy and efficiency. The code is available at https://github.com/rbrandt1/MaxTreeS.

arXiv:2006.15350

MiniNet: An extremely lightweight convolutional neural network for real-time unsupervised monocular depth estimation

Predicting depth from a single image is an attractive research topic since it provides one more dimension of information to enable machines to better perceive the world. Recently, deep learning has emerged as an effective approach to monocular depth estimation. As obtaining labeled data is costly, there is a recent trend to move from supervised learning to unsupervised learning to obtain monocular depth. However, most unsupervised learning methods capable of achieving high depth prediction accuracy will require a deep network architecture which will be too heavy and complex to run on embedded devices with limited storage and memory spaces. To address this issue, we propose a new powerful network with a recurrent module to achieve the capability of a deep network while at the same time maintaining an extremely lightweight size for real-time high performance unsupervised monocular depth prediction from video sequences. Besides, a novel efficient upsample block is proposed to fuse the features from the associated encoder layer and recover the spatial size of features with the small number of model parameters. We validate the effectiveness of our approach via extensive experiments on the KITTI dataset. Our new model can run at a speed of about 110 frames per second (fps) on a single GPU, 37 fps on a single CPU, and 2 fps on a Raspberry Pi 3. Moreover, it achieves higher depth accuracy with nearly 33 times fewer model parameters than state-of-the-art models. To the best of our knowledge, this work is the first extremely lightweight neural network trained on monocular video sequences for real-time unsupervised monocular depth estimation, which opens up the possibility of implementing deep learning-based real-time unsupervised monocular depth prediction on low-cost embedded devices.

arXiv:2006.14794공식 코드 · Papers with Code

The Signature Kernel is the solution of a Goursat PDE

Recently, there has been an increased interest in the development of kernel methods for learning with sequential data. The signature kernel is a learning tool with potential to handle irregularly sampled, multivariate time series. In "Kernels for sequentially ordered data" the authors introduced a kernel trick for the truncated version of this kernel avoiding the exponential complexity that would have been involved in a direct computation. Here we show that for continuously differentiable paths, the signature kernel solves a hyperbolic PDE and recognize the connection with a well known class of differential equations known in the literature as Goursat problems. This Goursat PDE only depends on the increments of the input sequences, does not require the explicit computation of signatures and can be solved efficiently using state-of-the-arthyperbolic PDE numerical solvers, giving a kernel trick for the untruncated signature kernel, with the same raw complexity as the method from "Kernels for sequentially ordered data", but with the advantage that the PDE numerical scheme is well suited for GPU parallelization, which effectively reduces the complexity by a full order of magnitude in the length of the input sequences. In addition, we extend the previous analysis to the space of geometric rough paths and establish, using classical results from rough path theory, that the rough version of the signature kernel solves a rough integral equation analogous to the aforementioned Goursat PDE. Finally, we empirically demonstrate the effectiveness of our PDE kernel as a machine learning tool in various machine learning applications dealing with sequential data. We release the library sigkernel publicly available at https://github.com/crispitagorico/sigkernel.

arXiv:2006.14652

Constant-Depth and Subcubic-Size Threshold Circuits for Matrix Multiplication

Boolean circuits of McCulloch-Pitts threshold gates are a classic model of neural computation studied heavily in the late 20th century as a model of general computation. Recent advances in large-scale neural computing hardware has made their practical implementation a near-term possibility. We describe a theoretical approach for multiplying two $N$ by $N$ matrices that integrates threshold gate logic with conventional fast matrix multiplication algorithms, that perform $O(N^\omega)$ arithmetic operations for a positive constant $\omega < 3$. Our approach converts such a fast matrix multiplication algorithm into a constant-depth threshold circuit with approximately $O(N^\omega)$ gates. Prior to our work, it was not known whether the $\Theta(N^3)$-gate barrier for matrix multiplication was surmountable by constant-depth threshold circuits. Dense matrix multiplication is a core operation in convolutional neural network training. Performing this work on a neural architecture instead of off-loading it to a GPU may be an appealing option.

arXiv:2006.15100

E2GC: Energy-efficient Group Convolution in Deep Neural Networks

The number of groups ($g$) in group convolution (GConv) is selected to boost the predictive performance of deep neural networks (DNNs) in a compute and parameter efficient manner. However, we show that naive selection of $g$ in GConv creates an imbalance between the computational complexity and degree of data reuse, which leads to suboptimal energy efficiency in DNNs. We devise an optimum group size model, which enables a balance between computational cost and data movement cost, thus, optimize the energy-efficiency of DNNs. Based on the insights from this model, we propose an "energy-efficient group convolution" (E2GC) module where, unlike the previous implementations of GConv, the group size ($G$) remains constant. Further, to demonstrate the efficacy of the E2GC module, we incorporate this module in the design of MobileNet-V1 and ResNeXt-50 and perform experiments on two GPUs, P100 and P4000. We show that, at comparable computational complexity, DNNs with constant group size (E2GC) are more energy-efficient than DNNs with a fixed number of groups (F$g$GC). For example, on P100 GPU, the energy-efficiency of MobileNet-V1 and ResNeXt-50 is increased by 10.8% and 4.73% (respectively) when E2GC modules substitute the F$g$GC modules in both the DNNs. Furthermore, through our extensive experimentation with ImageNet-1K and Food-101 image classification datasets, we show that the E2GC module enables a trade-off between generalization ability and representational power of DNN. Thus, the predictive performance of DNNs can be optimized by selecting an appropriate $G$. The code and trained models are available at https://github.com/iithcandle/E2GC-release.

arXiv:2006.15098

The Ramifications of Making Deep Neural Networks Compact

The recent trend in deep neural networks (DNNs) research is to make the networks more compact. The motivation behind designing compact DNNs is to improve energy efficiency since by virtue of having lower memory footprint, compact DNNs have lower number of off-chip accesses which improves energy efficiency. However, we show that making DNNs compact has indirect and subtle implications which are not well-understood. Reducing the number of parameters in DNNs increases the number of activations which, in turn, increases the memory footprint. We evaluate several recently-proposed compact DNNs on Tesla P100 GPU and show that their "activations to parameters ratio" ranges between 1.4 to 32.8. Further, the "memory-footprint to model size ratio" ranges between 15 to 443. This shows that a higher number of activations causes large memory footprint which increases on-chip/off-chip data movements. Furthermore, these parameter-reducing techniques reduce the arithmetic intensity which increases on-chip/off-chip memory bandwidth requirement. Due to these factors, the energy efficiency of compact DNNs may be significantly reduced which is against the original motivation for designing compact DNNs.

arXiv:2006.14090공식 코드 · Papers with Code

Neural Architecture Design for GPU-Efficient Networks

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 a simple and lightweight method as opposed to most Neural Architecture Search (NAS) methods that are complicated and computationally expensive. Based on the proposed framework, we design a family of GPU-Efficient Networks, or GENets in short. We did extensive evaluations on multiple GPU platforms and inference engines. While achieving $\geq 81.3\%$ top-1 accuracy on ImageNet, GENet is up to $6.4$ times faster than EfficienNet on GPU. It also outperforms most state-of-the-art models that are more efficient than EfficientNet in high precision regimes. Our source code and pre-trained models are available from \url{https://github.com/idstcv/GPU-Efficient-Networks}.

arXiv:2006.14618

Parametric Instance Classification for Unsupervised Visual Feature Learning

This paper presents parametric instance classification (PIC) for unsupervised visual feature learning. Unlike the state-of-the-art approaches which do instance discrimination in a dual-branch non-parametric fashion, PIC directly performs a one-branch parametric instance classification, revealing a simple framework similar to supervised classification and without the need to address the information leakage issue. We show that the simple PIC framework can be as effective as the state-of-the-art approaches, i.e. SimCLR and MoCo v2, by adapting several common component settings used in the state-of-the-art approaches. We also propose two novel techniques to further improve effectiveness and practicality of PIC: 1) a sliding-window data scheduler, instead of the previous epoch-based data scheduler, which addresses the extremely infrequent instance visiting issue in PIC and improves the effectiveness; 2) a negative sampling and weight update correction approach to reduce the training time and GPU memory consumption, which also enables application of PIC to almost unlimited training images. We hope that the PIC framework can serve as a simple baseline to facilitate future study.

arXiv:2006.13486

Ramanujan Bipartite Graph Products for Efficient Block Sparse Neural Networks

Sparse neural networks are shown to give accurate predictions competitive to denser versions, while also minimizing the number of arithmetic operations performed. However current hardware like GPU's can only exploit structured sparsity patterns for better efficiency. Hence the run time of a sparse neural network may not correspond to the arithmetic operations required. In this work, we propose RBGP( Ramanujan Bipartite Graph Product) framework for generating structured multi level block sparse neural networks by using the theory of Graph products. We also propose to use products of Ramanujan graphs which gives the best connectivity for a given level of sparsity. This essentially ensures that the i.) the networks has the structured block sparsity for which runtime efficient algorithms exists ii.) the model gives high prediction accuracy, due to the better expressive power derived from the connectivity of the graph iii.) the graph data structure has a succinct representation that can be stored efficiently in memory. We use our framework to design a specific connectivity pattern called RBGP4 which makes efficient use of the memory hierarchy available on GPU. We benchmark our approach by experimenting on image classification task over CIFAR dataset using VGG19 and WideResnet-40-4 networks and achieve 5-9x and 2-5x runtime gains over unstructured and block sparsity patterns respectively, while achieving the same level of accuracy.

arXiv:2006.13365공식 코드 · Papers with Code

Bringing Light Into the Dark: A Large-scale Evaluation of Knowledge Graph Embedding Models Under a Unified Framework

The heterogeneity in recently published knowledge graph embedding models' implementations, training, and evaluation has made fair and thorough comparisons difficult. In order to assess the reproducibility of previously published results, we re-implemented and evaluated 21 interaction models in the PyKEEN software package. Here, we outline which results could be reproduced with their reported hyper-parameters, which could only be reproduced with alternate hyper-parameters, and which could not be reproduced at all as well as provide insight as to why this might be the case. We then performed a large-scale benchmarking on four datasets with several thousands of experiments and 24,804 GPU hours of computation time. We present insights gained as to best practices, best configurations for each model, and where improvements could be made over previously published best configurations. Our results highlight that the combination of model architecture, training approach, loss function, and the explicit modeling of inverse relations is crucial for a model's performances, and not only determined by the model architecture. We provide evidence that several architectures can obtain results competitive to the state-of-the-art when configured carefully. We have made all code, experimental configurations, results, and analyses that lead to our interpretations available at https://github.com/pykeen/pykeen and https://github.com/pykeen/benchmarking