PyTorchKR Papers

최신 논문

총 5,575편
태그·기간 등 상세 필터

카테고리 탐색

로그인 후 이 태그 구독

arXiv:1910.01010

Design Space Exploration of Hardware Spiking Neurons for Embedded Artificial Intelligence

Machine learning is yielding unprecedented interest in research and industry, due to recent success in many applied contexts such as image classification and object recognition. However, the deployment of these systems requires huge computing capabilities, thus making them unsuitable for embedded systems. To deal with this limitation, many researchers are investigating brain-inspired computing, which would be a perfect alternative to the conventional Von Neumann architecture based computers (CPU/GPU) that meet the requirements for computing performance, but not for energy-efficiency. Therefore, neuromorphic hardware circuits that are adaptable for both parallel and distributed computations need to be designed. In this paper, we focus on Spiking Neural Networks (SNNs) with a comprehensive study of information coding methods and hardware exploration. In this context, we propose a framework for neuromorphic hardware design space exploration, which allows to define a suitable architecture based on application-specific constraints and starting from a wide variety of possible architectural choices. For this framework, we have developed a behavioral level simulator for neuromorphic hardware architectural exploration named NAXT. Moreover, we propose modified versions of the standard Rate Coding technique to make trade-offs with the Time Coding paradigm, which is characterized by the low number of spikes propagating in the network. Thus, we are able to reduce the number of spikes while keeping the same neuron's model, which results in an SNN with fewer events to process. By doing so, we seek to reduce the amount of power consumed by the hardware. Furthermore, we present three neuromorphic hardware architectures in order to quantitatively study the implementation of SNNs. These architectures are derived from a novel funnel-like Design Space Exploration framework for neuromorphic hardware.

arXiv:1910.00932

Training Kinetics in 15 Minutes: Large-scale Distributed Training on Videos

Deep video recognition is more computationally expensive than image recognition, especially on large-scale datasets like Kinetics [1]. Therefore, training scalability is essential to handle a large amount of videos. In this paper, we study the factors that impact the training scalability of video networks. We recognize three bottlenecks, including data loading (data movement from disk to GPU), communication (data movement over networking), and computation FLOPs. We propose three design guidelines to improve the scalability: (1) fewer FLOPs and hardware-friendly operator to increase the computation efficiency; (2) fewer input frames to reduce the data movement and increase the data loading efficiency; (3) smaller model size to reduce the networking traffic and increase the networking efficiency. With these guidelines, we designed a new operator Temporal Shift Module (TSM) that is efficient and scalable for distributed training. TSM model can achieve 1.8x higher throughput compared to previous I3D models. We scale up the training of the TSM model to 1,536 GPUs, with a mini-batch of 12,288 video clips/98,304 images, without losing the accuracy. With such hardware-aware model design, we are able to scale up the training on Summit supercomputer and reduce the training time on Kinetics dataset from 49 hours 55 minutes to 14 minutes 13 seconds, achieving a top-1 accuracy of 74.0%, which is 1.6x and 2.9x faster than previous 3D video models with higher accuracy. The code and more details can be found here: http://tsm-hanlab.mit.edu.

arXiv:1910.00895

Object Parsing in Sequences Using CoordConv Gated Recurrent Networks

We present a monocular object parsing framework for consistent keypoint localization by capturing temporal correlation on sequential data. In this paper, we propose a novel recurrent network based architecture to model long-range dependencies between intermediate features which are highly useful in tasks like keypoint localization and tracking. We leverage the expressiveness of the popular stacked hourglass architecture and augment it by adopting memory units between intermediate layers of the network with weights shared across stages for video frames. We observe that this weight sharing scheme not only enables us to frame hourglass architecture as a recurrent network but also prove to be highly effective in producing increasingly refined estimates for sequential tasks. Furthermore, we propose a new memory cell, we call CoordConvGRU which learns to selectively preserve spatio-temporal correlation and showcase our results on the keypoint localization task. The experiments show that our approach is able to model the motion dynamics between the frames and significantly outperforms the baseline hourglass network. Even though our network is trained on a synthetically rendered dataset, we observe that with minimal fine tuning on 300 real images we are able to achieve performance at par with various state-of-the-art methods trained with the same level of supervisory inputs. By using a simpler architecture than other methods enables us to run it in real time on a standard GPU which is desirable for such applications. Finally, we make our architectures and 524 annotated sequences of cars from KITTI dataset publicly available.

arXiv:1910.00572

Map as The Hidden Sensor: Fast Odometry-Based Global Localization

Accurate and robust global localization is essential to robotics applications. We propose a novel global localization method that employs the map traversability as a hidden observation. The resulting map-corrected odometry localization is able to provide an accurate belief tensor of the robot state. Our method can be used for blind robots in dark or highly reflective areas. In contrast to odometry drift in long-term, our method using only odometry and the map converges in longterm. Our method can also be integrated with other sensors to boost the localization performance. The algorithm does not have any initial state assumption and tracks all possible robot states at all times. Therefore, our method is global and is robust in the event of ambiguous observations. We parallel each step of our algorithm such that it can be performed in real-time (up to ~ 300 Hz) using GPU. We validate our algorithm in different publicly available floor-plans and show that it is able to converge to the ground truth fast while being robust to ambiguities.

arXiv:1910.00078

MIOpen: An Open Source Library For Deep Learning Primitives

Deep Learning has established itself to be a common occurrence in the business lexicon. The unprecedented success of deep learning in recent years can be attributed to: abundance of data, availability of gargantuan compute capabilities offered by GPUs, and adoption of open-source philosophy by the researchers and industry. Deep neural networks can be decomposed into a series of different operators. MIOpen, AMD's open-source deep learning primitives library for GPUs, provides highly optimized implementations of such operators, shielding researchers from internal implementation details and hence, accelerating the time to discovery. This paper introduces MIOpen and provides details about the internal workings of the library and supported features. MIOpen innovates on several fronts, such as implementing fusion to optimize for memory bandwidth and GPU launch overheads, providing an auto-tuning infrastructure to overcome the large design space of problem configurations, and implementing different algorithms to optimize convolutions for different filter and input sizes. MIOpen is one of the first libraries to publicly support the bfloat16 data-type for convolutions, allowing efficient training at lower precision without the loss of accuracy.

arXiv:1909.13654

Serving Recurrent Neural Networks Efficiently with a Spatial Accelerator

Recurrent Neural Network (RNN) applications form a major class of AI-powered, low-latency data center workloads. Most execution models for RNN acceleration break computation graphs into BLAS kernels, which lead to significant inter-kernel data movement and resource underutilization. We show that by supporting more general loop constructs that capture design parameters in accelerators, it is possible to improve resource utilization using cross-kernel optimization without sacrificing programmability. Such abstraction level enables a design space search that can lead to efficient usage of on-chip resources on a spatial architecture across a range of problem sizes. We evaluate our optimization strategy on such abstraction with DeepBench using a configurable spatial accelerator. We demonstrate that this implementation provides a geometric speedup of 30x in performance, 1.6x in area, and 2x in power efficiency compared to a Tesla V100 GPU, and a geometric speedup of 2x compared to Microsoft Brainwave implementation on a Stratix 10 FPGA.

arXiv:1909.12978공식 코드 · Papers with Code

MutualNet: Adaptive ConvNet via Mutual Learning from Network Width and Resolution

We propose the width-resolution mutual learning method (MutualNet) to train a network that is executable at dynamic resource constraints to achieve adaptive accuracy-efficiency trade-offs at runtime. Our method trains a cohort of sub-networks with different widths using different input resolutions to mutually learn multi-scale representations for each sub-network. It achieves consistently better ImageNet top-1 accuracy over the state-of-the-art adaptive network US-Net under different computation constraints, and outperforms the best compound scaled MobileNet in EfficientNet by 1.5%. The superiority of our method is also validated on COCO object detection and instance segmentation as well as transfer learning. Surprisingly, the training strategy of MutualNet can also boost the performance of a single network, which substantially outperforms the powerful AutoAugmentation in both efficiency (GPU search hours: 15000 vs. 0) and accuracy (ImageNet: 77.6% vs. 78.6%). Code is available at \url{https://github.com/taoyang1122/MutualNet}.

arXiv:1909.11942공식 코드 · Papers with Code

ALBERT: A Lite BERT for Self-supervised Learning of Language Representations

Increasing model size when pretraining natural language representations often results in improved performance on downstream tasks. However, at some point further model increases become harder due to GPU/TPU memory limitations and longer training times. To address these problems, we present two parameter-reduction techniques to lower memory consumption and increase the training speed of BERT. Comprehensive empirical evidence shows that our proposed methods lead to models that scale much better compared to the original BERT. We also use a self-supervised loss that focuses on modeling inter-sentence coherence, and show it consistently helps downstream tasks with multi-sentence inputs. As a result, our best model establishes new state-of-the-art results on the GLUE, RACE, and \squad benchmarks while having fewer parameters compared to BERT-large. The code and the pretrained models are available at https://github.com/google-research/ALBERT.

arXiv:1909.11469공식 코드 · Papers with Code

Message Scheduling for Performant, Many-Core Belief Propagation

Belief Propagation (BP) is a message-passing algorithm for approximate inference over Probabilistic Graphical Models (PGMs), finding many applications such as computer vision, error-correcting codes, and protein-folding. While general, the convergence and speed of the algorithm has limited its practical use on difficult inference problems. As an algorithm that is highly amenable to parallelization, many-core Graphical Processing Units (GPUs) could significantly improve BP performance. Improving BP through many-core systems is non-trivial: the scheduling of messages in the algorithm strongly affects performance. We present a study of message scheduling for BP on GPUs. We demonstrate that BP exhibits a tradeoff between speed and convergence based on parallelism and show that existing message schedulings are not able to utilize this tradeoff. To this end, we present a novel randomized message scheduling approach, Randomized BP (RnBP), which outperforms existing methods on the GPU.

arXiv:1909.10798공식 코드 · Papers with Code

Development of Fast Refinement Detectors on AI Edge Platforms

With the improvements in the object detection networks, several variations of object detection networks have been achieved impressive performance. However, the performance evaluation of most models has focused on detection accuracy, and performance verification is mostly based on high-end GPU hardware. In this paper, we propose real-time object detectors that guarantee balanced performance for real-time systems on embedded platforms. The proposed model utilizes the basic head structure of the RefineDet model, which is a variant of the single-shot object detector (SSD). In order to ensure real-time performance, CNN models with relatively shallow layers or fewer parameters have been used as the backbone structure. In addition to the basic VGGNet and ResNet structures, various backbone structures such as MobileNet, Xception, ResNeXt, Inception-SENet, and SE-ResNeXt have been used for this purpose. Successful training of object detection networks was achieved through an appropriate combination of intermediate layers. The accuracy of the proposed detector was estimated by the evaluation of the MS-COCO 2017 object detection dataset and the inference speed on the NVIDIA Drive PX2 and Jetson Xavier boards were tested to verify real-time performance in the embedded systems. The experiments show that the proposed models ensure balanced performance in terms of accuracy and inference speed in the embedded system environments. In addition, unlike the high-end GPUs, the use of embedded GPUs involves several additional concerns for efficient inference, which have been identified in this work. The codes and models are publicly available on the web (link).

arXiv:1909.10726

Multi-scale fully convolutional neural networks for histopathology image segmentation: from nuclear aberrations to the global tissue architecture

Histopathologic diagnosis relies on simultaneous integration of information from a broad range of scales, ranging from nuclear aberrations ($\approx \mathcal{O}(0.1{\mu m})$) through cellular structures ($\approx \mathcal{O}(10{\mu m})$) to the global tissue architecture ($\gtrapprox \mathcal{O}(1{mm})$). To explicitly mimic how human pathologists combine multi-scale information, we introduce a family of multi-encoder FCNs with deep fusion. We present a simple block for merging model paths with differing spatial scales in a spatial relationship-preserving fashion, which can readily be included in standard encoder-decoder networks. Additionally, a context classification gate block is proposed as an alternative for the incorporation of global context. Our experiments were performed on three publicly available whole-slide images of recent challenges (PAIP 2019, BACH 2020, CAMELYON 2016). The multi-scale architectures consistently outperformed the baseline single-scale U-Nets by a large margin. They benefit from local as well as global context and particularly a combination of both. If feature maps from different scales are fused, doing so in a manner preserving spatial relationships was found to be beneficial. Deep guidance by a context classification loss appeared to improve model training at low computational costs. All multi-scale models had a reduced GPU memory footprint compared to ensembles of individual U-Nets trained on different image scales. Additional path fusions were shown to be possible at low computational cost, opening up possibilities for further, systematic and task-specific architecture optimization. The findings demonstrate the potential of the presented family of human-inspired, end-to-end trainable, multi-scale multi-encoder FCNs to improve deep histopathologic diagnosis by extensive integration of largely different spatial scales.

arXiv:1909.10690

Analysis of Generalized Entropies in Mutual Information Medical Image Registration

Mutual information (MI) is the standard method used in image registration and the most studied one but can diverge and produce wrong results when used in an automated manner. In this study we compared the results of the ITK Mattes MI function, used in 3D Slicer and ITK derived software solutions, and our own MICUDA Shannon and Tsallis MI functions under the translation, rotation and scale transforms in a 3D mathematical space. This comparison allows to understand why registration fails in some circumstances and how to produce a more robust automated algorithm to register medical images. Since our algorithms were designed to use GPU computations we also have a huge gain in speed while improving the quality of registration.

arXiv:1909.10389공식 코드 · Papers with Code

Machine Learning Pipelines with Modern Big Data Tools for High Energy Physics

The effective utilization at scale of complex machine learning (ML) techniques for HEP use cases poses several technological challenges, most importantly on the actual implementation of dedicated end-to-end data pipelines. A solution to these challenges is presented, which allows training neural network classifiers using solutions from the Big Data and data science ecosystems, integrated with tools, software, and platforms common in the HEP environment. In particular, Apache Spark is exploited for data preparation and feature engineering, running the corresponding (Python) code interactively on Jupyter notebooks. Key integrations and libraries that make Spark capable of ingesting data stored using ROOT format and accessed via the XRootD protocol, are described and discussed. Training of the neural network models, defined using the Keras API, is performed in a distributed fashion on Spark clusters by using BigDL with Analytics Zoo and also by using TensorFlow, notably for distributed training on CPU and GPU resourcess. The implementation and the results of the distributed training are described in detail in this work.

arXiv:1909.09757

Positive-Unlabeled Compression on the Cloud

Many attempts have been done to extend the great success of convolutional neural networks (CNNs) achieved on high-end GPU servers to portable devices such as smart phones. Providing compression and acceleration service of deep learning models on the cloud is therefore of significance and is attractive for end users. However, existing network compression and acceleration approaches usually fine-tuning the svelte model by requesting the entire original training data (\eg ImageNet), which could be more cumbersome than the network itself and cannot be easily uploaded to the cloud. In this paper, we present a novel positive-unlabeled (PU) setting for addressing this problem. In practice, only a small portion of the original training set is required as positive examples and more useful training examples can be obtained from the massive unlabeled data on the cloud through a PU classifier with an attention based multi-scale feature extractor. We further introduce a robust knowledge distillation (RKD) scheme to deal with the class imbalance problem of these newly augmented training examples. The superiority of the proposed method is verified through experiments conducted on the benchmark models and datasets. We can use only $8\%$ of uniformly selected data from the ImageNet to obtain an efficient model with comparable performance to the baseline ResNet-34.

arXiv:1909.09690

A Deep Learning-Based Approach for Measuring the Domain Similarity of Persian Texts

In this paper, we propose a novel approach for measuring the degree of similarity between categories of two pieces of Persian text, which were published as descriptions of two separate advertisements. We built an appropriate dataset for this work using a dataset which consists of advertisements posted on an e-commerce website. We generated a significant number of paired texts from this dataset and assigned each pair a score from 0 to 3, which demonstrates the degree of similarity between the domains of the pair. In this work, we represent words with word embedding vectors derived from word2vec. Then deep neural network models are used to represent texts. Eventually, we employ concatenation of absolute difference and bit-wise multiplication and a fully-connected neural network to produce a probability distribution vector for the score of the pairs. Through a supervised learning approach, we trained our model on a GPU, and our best model achieved an F1 score of 0.9865.

arXiv:1909.09709공식 코드 · Papers with Code

SkyNet: a Hardware-Efficient Method for Object Detection and Tracking on Embedded Systems

Object detection and tracking are challenging tasks for resource-constrained embedded systems. While these tasks are among the most compute-intensive tasks from the artificial intelligence domain, they are only allowed to use limited computation and memory resources on embedded devices. In the meanwhile, such resource-constrained implementations are often required to satisfy additional demanding requirements such as real-time response, high-throughput performance, and reliable inference accuracy. To overcome these challenges, we propose SkyNet, a hardware-efficient neural network to deliver the state-of-the-art detection accuracy and speed for embedded systems. Instead of following the common top-down flow for compact DNN (Deep Neural Network) design, SkyNet provides a bottom-up DNN design approach with comprehensive understanding of the hardware constraints at the very beginning to deliver hardware-efficient DNNs. The effectiveness of SkyNet is demonstrated by winning the competitive System Design Contest for low power object detection in the 56th IEEE/ACM Design Automation Conference (DAC-SDC), where our SkyNet significantly outperforms all other 100+ competitors: it delivers 0.731 Intersection over Union (IoU) and 67.33 frames per second (FPS) on a TX2 embedded GPU; and 0.716 IoU and 25.05 FPS on an Ultra96 embedded FPGA. The evaluation of SkyNet is also extended to GOT-10K, a recent large-scale high-diversity benchmark for generic object tracking in the wild. For state-of-the-art object trackers SiamRPN++ and SiamMask, where ResNet-50 is employed as the backbone, implementations using our SkyNet as the backbone DNN are 1.60X and 1.73X faster with better or similar accuracy when running on a 1080Ti GPU, and 37.20X smaller in terms of parameter size for significantly better memory and storage footprint.

arXiv:1909.09252

HyperLearn: A Distributed Approach for Representation Learning in Datasets With Many Modalities

Multimodal datasets contain an enormous amount of relational information, which grows exponentially with the introduction of new modalities. Learning representations in such a scenario is inherently complex due to the presence of multiple heterogeneous information channels. These channels can encode both (a) inter-relations between the items of different modalities and (b) intra-relations between the items of the same modality. Encoding multimedia items into a continuous low-dimensional semantic space such that both types of relations are captured and preserved is extremely challenging, especially if the goal is a unified end-to-end learning framework. The two key challenges that need to be addressed are: 1) the framework must be able to merge complex intra and inter relations without losing any valuable information and 2) the learning model should be invariant to the addition of new and potentially very different modalities. In this paper, we propose a flexible framework which can scale to data streams from many modalities. To that end we introduce a hypergraph-based model for data representation and deploy Graph Convolutional Networks to fuse relational information within and across modalities. Our approach provides an efficient solution for distributing otherwise extremely computationally expensive or even unfeasible training processes across multiple-GPUs, without any sacrifices in accuracy. Moreover, adding new modalities to our model requires only an additional GPU unit keeping the computational time unchanged, which brings representation learning to truly multimodal datasets. We demonstrate the feasibility of our approach in the experiments on multimedia datasets featuring second, third and fourth order relations.

arXiv:1909.09213

Experimenting with Constraint Programming on GPU

The focus of my PhD thesis is on exploring parallel approaches to efficiently solve problems modeled by constraints and presenting a new proposal. Current solvers are very advanced; they are carefully designed to effectively manage the high-level problems' description and include refined strategies to avoid useless work. Despite this, finding a solution can take an unacceptable amount of time. Parallelization can mitigate this problem when the instance of the problem modeled is large, as it happens in real world problems. It is done by propagating constraints in parallel and concurrently exploring different parts of the search space. I am developing on a constraint solver that exploits the many cores available on Graphics Processing Units (GPU) to speed up the search.

arXiv:1909.09010

A Random Gossip BMUF Process for Neural Language Modeling

Neural network language model (NNLM) is an essential component of industrial ASR systems. One important challenge of training an NNLM is to leverage between scaling the learning process and handling big data. Conventional approaches such as block momentum provides a blockwise model update filtering (BMUF) process and achieves almost linear speedups with no performance degradation for speech recognition. However, it needs to calculate the model average from all computing nodes (e.g., GPUs) and when the number of computing nodes is large, the learning suffers from the severe communication latency. As a consequence, BMUF is not suitable under restricted network conditions. In this paper, we present a decentralized BMUF process, in which the model is split into different components, each of which is updated by communicating to some randomly chosen neighbor nodes with the same component, followed by a BMUF-like process. We apply this method to several LSTM language modeling tasks. Experimental results show that our approach achieves consistently better performance than conventional BMUF. In particular, we obtain a lower perplexity than the single-GPU baseline on the wiki-text-103 benchmark using 4 GPUs. In addition, no performance degradation is observed when scaling to 8 and 16 GPUs.

arXiv:1909.08989공식 코드 · Papers with Code

Grid Anchor based Image Cropping: A New Benchmark and An Efficient Model

Image cropping aims to improve the composition as well as aesthetic quality of an image by removing extraneous content from it. Most of the existing image cropping databases provide only one or several human-annotated bounding boxes as the groundtruths, which can hardly reflect the non-uniqueness and flexibility of image cropping in practice. The employed evaluation metrics such as intersection-over-union cannot reliably reflect the real performance of a cropping model, either. This work revisits the problem of image cropping, and presents a grid anchor based formulation by considering the special properties and requirements (e.g., local redundancy, content preservation, aspect ratio) of image cropping. Our formulation reduces the searching space of candidate crops from millions to no more than ninety. Consequently, a grid anchor based cropping benchmark is constructed, where all crops of each image are annotated and more reliable evaluation metrics are defined. To meet the practical demands of robust performance and high efficiency, we also design an effective and lightweight cropping model. By simultaneously considering the region of interest and region of discard, and leveraging multi-scale information, our model can robustly output visually pleasing crops for images of different scenes. With less than 2.5M parameters, our model runs at a speed of 200 FPS on one single GTX 1080Ti GPU and 12 FPS on one i7-6800K CPU. The code is available at: \url{https://github.com/HuiZeng/Grid-Anchor-based-Image-Cropping-Pytorch}.