PyTorchKR Papers

최신 논문

총 5,575편
태그·기간 등 상세 필터

카테고리 탐색

로그인 후 이 태그 구독

arXiv:2011.07962

Performance of Transfer Learning Model vs. Traditional Neural Network in Low System Resource Environment

Recently, the use of pre-trained model to build neural network based on transfer learning methodology is increasingly popular. These pre-trained models present the benefit of using less computing resources to train model with smaller amount of training data. The rise of state-of-the-art models such as BERT, XLNet and GPT boost accuracy and benefit as a base model for transfer leanring. However, these models are still too complex and consume many computing resource to train for transfer learning with low GPU memory. We will compare the performance and cost between lighter transfer learning model and purposely built neural network for NLP application of text classification and NER model.

arXiv:2011.07125공식 코드 · Papers with Code

Better, Faster Fermionic Neural Networks

The Fermionic Neural Network (FermiNet) is a recently-developed neural network architecture that can be used as a wavefunction Ansatz for many-electron systems, and has already demonstrated high accuracy on small systems. Here we present several improvements to the FermiNet that allow us to set new records for speed and accuracy on challenging systems. We find that increasing the size of the network is sufficient to reach chemical accuracy on atoms as large as argon. Through a combination of implementing FermiNet in JAX and simplifying several parts of the network, we are able to reduce the number of GPU hours needed to train the FermiNet on large systems by an order of magnitude. This enables us to run the FermiNet on the challenging transition of bicyclobutane to butadiene and compare against the PauliNet on the automerization of cyclobutadiene, and we achieve results near the state of the art for both.

arXiv:2011.08061

FRDet: Balanced and Lightweight Object Detector based on Fire-Residual Modules for Embedded Processor of Autonomous Driving

For deployment on an embedded processor for autonomous driving, the object detection network should satisfy all of the accuracy, real-time inference, and light model size requirements. Conventional deep CNN-based detectors aim for high accuracy, making their model size heavy for an embedded system with limited memory space. In contrast, lightweight object detectors are greatly compressed but at a significant sacrifice of accuracy. Therefore, we propose FRDet, a lightweight one-stage object detector that is balanced to satisfy all the constraints of accuracy, model size, and real-time processing on an embedded GPU processor for autonomous driving applications. Our network aims to maximize the compression of the model while achieving or surpassing YOLOv3 level of accuracy. This paper proposes the Fire-Residual (FR) module to design a lightweight network with low accuracy loss by adapting fire modules with residual skip connections. In addition, the Gaussian uncertainty modeling of the bounding box is applied to further enhance the localization accuracy. Experiments on the KITTI dataset showed that FRDet reduced the memory size by 50.8% but achieved higher accuracy by 1.12% mAP compared to YOLOv3. Moreover, the real-time detection speed reached 31.3 FPS on an embedded GPU board(NVIDIA Xavier). The proposed network achieved higher compression with comparable accuracy compared to other deep CNN object detectors while showing improved accuracy than the lightweight detector baselines. Therefore, the proposed FRDet is a well-balanced and efficient object detector for practical application in autonomous driving that can satisfies all the criteria of accuracy, real-time inference, and light model size.

arXiv:2011.06315공식 코드 · Papers with Code

Biomedical Named Entity Recognition at Scale

Named entity recognition (NER) is a widely applicable natural language processing task and building block of question answering, topic modeling, information retrieval, etc. In the medical domain, NER plays a crucial role by extracting meaningful chunks from clinical notes and reports, which are then fed to downstream tasks like assertion status detection, entity resolution, relation extraction, and de-identification. Reimplementing a Bi-LSTM-CNN-Char deep learning architecture on top of Apache Spark, we present a single trainable NER model that obtains new state-of-the-art results on seven public biomedical benchmarks without using heavy contextual embeddings like BERT. This includes improving BC4CHEMD to 93.72% (4.1% gain), Species800 to 80.91% (4.6% gain), and JNLPBA to 81.29% (5.2% gain). In addition, this model is freely available within a production-grade code base as part of the open-source Spark NLP library; can scale up for training and inference in any Spark cluster; has GPU support and libraries for popular programming languages such as Python, R, Scala and Java; and can be extended to support other human languages with no code changes.

arXiv:2011.06295

When deep learning models on GPU can be accelerated by taking advantage of unstructured sparsity

This paper is focused on the improvement the efficiency of the sparse convolutional neural networks (CNNs) layers on graphic processing units (GPU). The Nvidia deep neural network (cuDnn) library provides the most effective implementation of deep learning (DL) algorithms for GPUs. GPUs are one of the most efficient and commonly used accelerators for deep learning computations. The modern CNN models need megabytes of coefficients and needed millions MAC operations to perform convolution. One of the most common techniques for compressing CNN models is weight pruning. There are two main types of pruning: structural (based on removing whole weight channels) and non-structural (removing individual weights). The first enables much easier acceleration, but with this type it is difficult to achieve a sparsity level and accuracy as high as that obtained with the second type. Non-structural pruning with retraining can generate a matrix-weight up to $\sim90\%$ or more of sparsity in some deep CNN models. This work shows when is worth using a direct sparse operation to speed-up the calculation of the convolution layers. The VGG-16, CNN-non-static and 1x1 layers from ResNet models were used as a benchmarks. In addition, we present the impact of using reduced precision on time efficiency.

arXiv:2011.05497

Understanding Training Efficiency of Deep Learning Recommendation Models at Scale

The use of GPUs has proliferated for machine learning workflows and is now considered mainstream for many deep learning models. Meanwhile, when training state-of-the-art personal recommendation models, which consume the highest number of compute cycles at our large-scale datacenters, the use of GPUs came with various challenges due to having both compute-intensive and memory-intensive components. GPU performance and efficiency of these recommendation models are largely affected by model architecture configurations such as dense and sparse features, MLP dimensions. Furthermore, these models often contain large embedding tables that do not fit into limited GPU memory. The goal of this paper is to explain the intricacies of using GPUs for training recommendation models, factors affecting hardware efficiency at scale, and learnings from a new scale-up GPU server design, Zion.

arXiv:2011.05277

Qualities, challenges and future of genetic algorithms: a literature review

Genetic algorithms, computer programs that simulate natural evolution, are increasingly applied across many disciplines. They have been used to solve various optimisation problems from neural network architecture search to strategic games, and to model phenomena of adaptation and learning. Expertise on the qualities and drawbacks of this technique is largely scattered across the literature or former, motivating an compilation of this knowledge at the light of the most recent developments of the field. In this review, we present genetic algorithms, their qualities, limitations and challenges, as well as some future development perspectives. Genetic algorithms are capable of exploring large and complex spaces of possible solutions, to quickly locate promising elements, and provide an adequate modelling tool to describe evolutionary systems, from games to economies. They however suffer from high computation costs, difficult parameter configuration, and crucial representation of the solutions. Recent developments such as GPU, parallel and quantum computing, conception of powerful parameter control methods, and novel approaches in representation strategies, may be keys to overcome those limitations. This compiling review aims at informing practitioners and newcomers in the field alike in their genetic algorithm research, and at outlining promising avenues for future research. It highlights the potential for interdisciplinary research associating genetic algorithms to pulse original discoveries in social sciences, open ended evolution, artificial life and AI.

arXiv:2011.04943

Simple means Faster: Real-Time Human Motion Forecasting in Monocular First Person Videos on CPU

We present a simple, fast, and light-weight RNN based framework for forecasting future locations of humans in first person monocular videos. The primary motivation for this work was to design a network which could accurately predict future trajectories at a very high rate on a CPU. Typical applications of such a system would be a social robot or a visual assistance system for all, as both cannot afford to have high compute power to avoid getting heavier, less power efficient, and costlier. In contrast to many previous methods which rely on multiple type of cues such as camera ego-motion or 2D pose of the human, we show that a carefully designed network model which relies solely on bounding boxes can not only perform better but also predicts trajectories at a very high rate while being quite low in size of approximately 17 MB. Specifically, we demonstrate that having an auto-encoder in the encoding phase of the past information and a regularizing layer in the end boosts the accuracy of predictions with negligible overhead. We experiment with three first person video datasets: CityWalks, FPL and JAAD. Our simple method trained on CityWalks surpasses the prediction accuracy of state-of-the-art method (STED) while being 9.6x faster on a CPU (STED runs on a GPU). We also demonstrate that our model can transfer zero-shot or after just 15% fine-tuning to other similar datasets and perform on par with the state-of-the-art methods on such datasets (FPL and DTP). To the best of our knowledge, we are the first to accurately forecast trajectories at a very high prediction rate of 78 trajectories per second on CPU.

arXiv:2011.04275

Runtime Performances Benchmark for Knowledge Graph Embedding Methods

This paper wants to focus on providing a characterization of the runtime performances of state-of-the-art implementations of KGE alghoritms, in terms of memory footprint and execution time. Despite the rapidly growing interest in KGE methods, so far little attention has been devoted to their comparison and evaluation; in particular, previous work mainly focused on performance in terms of accuracy in specific tasks, such as link prediction. To this extent, a framework is proposed for evaluating available KGE implementations against graphs with different properties, with a particular focus on the effectiveness of the adopted optimization strategies. Graphs and models have been trained leveraging different architectures, in order to enlighten features and properties of both models and the architectures they have been trained on. Some results enlightened with experiments in this document are the fact that multithreading is efficient, but benefit deacreases as the number of threads grows in case of CPU. GPU proves to be the best architecture for the given task, even if CPU with some vectorized instructions still behaves well. Finally, RAM utilization for the loading of the graph never changes between different architectures and depends only on the type of graph, not on the model.

arXiv:2011.03910

Faster object tracking pipeline for real time tracking

Multi-object tracking (MOT) is a challenging practical problem for vision based applications. Most recent approaches for MOT use precomputed detections from models such as Faster RCNN, performing fine-tuning of bounding boxes and association in subsequent phases. However, this is not suitable for actual industrial applications due to unavailability of detections upfront. In their recent work, Wang et al. proposed a tracking pipeline that uses a Joint detection and embedding model and performs target localization and association in realtime. Upon investigating the tracking by detection paradigm, we find that the tracking pipeline can be made faster by performing localization and association tasks parallely with model prediction. This, and other computational optimizations such as using mixed precision model and performing batchwise detection result in a speed-up of the tracking pipeline by 57.8\% (19 FPS to 30 FPS) on FullHD resolution. Moreover, the speed is independent of the object density in image sequence. The main contribution of this paper is showcasing a generic pipeline which can be used to speed up detection based object tracking methods. We also reviewed different batch sizes for optimal performance, taking into consideration GPU memory usage and speed.

arXiv:2011.02265

S3-Net: A Fast and Lightweight Video Scene Understanding Network by Single-shot Segmentation

Real-time understanding in video is crucial in various AI applications such as autonomous driving. This work presents a fast single-shot segmentation strategy for video scene understanding. The proposed net, called S3-Net, quickly locates and segments target sub-scenes, meanwhile extracts structured time-series semantic features as inputs to an LSTM-based spatio-temporal model. Utilizing tensorization and quantization techniques, S3-Net is intended to be lightweight for edge computing. Experiments using CityScapes, UCF11, HMDB51 and MOMENTS datasets demonstrate that the proposed S3-Net achieves an accuracy improvement of 8.1% versus the 3D-CNN based approach on UCF11, a storage reduction of 6.9x and an inference speed of 22.8 FPS on CityScapes with a GTX1080Ti GPU.

arXiv:2011.02242공식 코드 · Papers with Code

BGGAN: Bokeh-Glass Generative Adversarial Network for Rendering Realistic Bokeh

A photo captured with bokeh effect often means objects in focus are sharp while the out-of-focus areas are all blurred. DSLR can easily render this kind of effect naturally. However, due to the limitation of sensors, smartphones cannot capture images with depth-of-field effects directly. In this paper, we propose a novel generator called Glass-Net, which generates bokeh images not relying on complex hardware. Meanwhile, the GAN-based method and perceptual loss are combined for rendering a realistic bokeh effect in the stage of finetuning the model. Moreover, Instance Normalization(IN) is reimplemented in our network, which ensures our tflite model with IN can be accelerated on smartphone GPU. Experiments show that our method is able to render a high-quality bokeh effect and process one $1024 \times 1536$ pixel image in 1.9 seconds on all smartphone chipsets. This approach ranked First in AIM 2020 Rendering Realistic Bokeh Challenge Track 1 \& Track 2.

arXiv:2011.02030공식 코드 · Papers with Code

PyLightcurve-torch: a transit modelling package for deep learning applications in PyTorch

We present a new open source python package, based on PyLightcurve and PyTorch, tailored for efficient computation and automatic differentiation of exoplanetary transits. The classes and functions implemented are fully vectorised, natively GPU-compatible and differentiable with respect to the stellar and planetary parameters. This makes PyLightcurve-torch suitable for traditional forward computation of transits, but also extends the range of possible applications with inference and optimisation algorithms requiring access to the gradients of the physical model. This endeavour is aimed at fostering the use of deep learning in exoplanets research, motivated by an ever increasing amount of stellar light curves data and various incentives for the improvement of detection and characterisation techniques.

arXiv:2011.01890공식 코드 · Papers with Code

RealHePoNet: a robust single-stage ConvNet for head pose estimation in the wild

Human head pose estimation in images has applications in many fields such as human-computer interaction or video surveillance tasks. In this work, we address this problem, defined here as the estimation of both vertical (tilt/pitch) and horizontal (pan/yaw) angles, through the use of a single Convolutional Neural Network (ConvNet) model, trying to balance precision and inference speed in order to maximize its usability in real-world applications. Our model is trained over the combination of two datasets: 'Pointing'04' (aiming at covering a wide range of poses) and 'Annotated Facial Landmarks in the Wild' (in order to improve robustness of our model for its use on real-world images). Three different partitions of the combined dataset are defined and used for training, validation and testing purposes. As a result of this work, we have obtained a trained ConvNet model, coined RealHePoNet, that given a low-resolution grayscale input image, and without the need of using facial landmarks, is able to estimate with low error both tilt and pan angles (~4.4{\deg} average error on the test partition). Also, given its low inference time (~6 ms per head), we consider our model usable even when paired with medium-spec hardware (i.e. GTX 1060 GPU). * Code available at: https://github.com/rafabs97/headpose_final * Demo video at: https://www.youtube.com/watch?v=2UeuXh5DjAE

arXiv:2011.01731공식 코드 · Papers with Code

RecBole: Towards a Unified, Comprehensive and Efficient Framework for Recommendation Algorithms

In recent years, there are a large number of recommendation algorithms proposed in the literature, from traditional collaborative filtering to deep learning algorithms. However, the concerns about how to standardize open source implementation of recommendation algorithms continually increase in the research community. In the light of this challenge, we propose a unified, comprehensive and efficient recommender system library called RecBole, which provides a unified framework to develop and reproduce recommendation algorithms for research purpose. In this library, we implement 73 recommendation models on 28 benchmark datasets, covering the categories of general recommendation, sequential recommendation, context-aware recommendation and knowledge-based recommendation. We implement the RecBole library based on PyTorch, which is one of the most popular deep learning frameworks. Our library is featured in many aspects, including general and extensible data structures, comprehensive benchmark models and datasets, efficient GPU-accelerated execution, and extensive and standard evaluation protocols. We provide a series of auxiliary functions, tools, and scripts to facilitate the use of this library, such as automatic parameter tuning and break-point resume. Such a framework is useful to standardize the implementation and evaluation of recommender systems. The project and documents are released at https://recbole.io/.

arXiv:2011.01112

Scheduling Real-time Deep Learning Services as Imprecise Computations

The paper presents an efficient real-time scheduling algorithm for intelligent real-time edge services, defined as those that perform machine intelligence tasks, such as voice recognition, LIDAR processing, or machine vision, on behalf of local embedded devices that are themselves unable to support extensive computations. The work contributes to a recent direction in real-time computing that develops scheduling algorithms for machine intelligence tasks with anytime prediction. We show that deep neural network workflows can be cast as imprecise computations, each with a mandatory part and (several) optional parts whose execution utility depends on input data. The goal of the real-time scheduler is to maximize the average accuracy of deep neural network outputs while meeting task deadlines, thanks to opportunistic shedding of the least necessary optional parts. The work is motivated by the proliferation of increasingly ubiquitous but resource-constrained embedded devices (for applications ranging from autonomous cars to the Internet of Things) and the desire to develop services that endow them with intelligence. Experiments on recent GPU hardware and a state of the art deep neural network for machine vision illustrate that our scheme can increase the overall accuracy by 10%-20% while incurring (nearly) no deadline misses.

arXiv:2011.00869

Receptive Field Size Optimization with Continuous Time Pooling

The pooling operation is a cornerstone element of convolutional neural networks. These elements generate receptive fields for neurons, in which local perturbations should have minimal effect on the output activations, increasing robustness and invariance of the network. In this paper we will present an altered version of the most commonly applied method, maximum pooling, where pooling in theory is substituted by a continuous time differential equation, which generates a location sensitive pooling operation, more similar to biological receptive fields. We will present how this continuous method can be approximated numerically using discrete operations which fit ideally on a GPU. In our approach the kernel size is substituted by diffusion strength which is a continuous valued parameter, this way it can be optimized by gradient descent algorithms. We will evaluate the effect of continuous pooling on accuracy and computational need using commonly applied network architectures and datasets.

arXiv:2010.16407공식 코드 · Papers with Code

TopicBERT for Energy Efficient Document Classification

Prior research notes that BERT's computational cost grows quadratically with sequence length thus leading to longer training times, higher GPU memory constraints and carbon emissions. While recent work seeks to address these scalability issues at pre-training, these issues are also prominent in fine-tuning especially for long sequence tasks like document classification. Our work thus focuses on optimizing the computational cost of fine-tuning for document classification. We achieve this by complementary learning of both topic and language models in a unified framework, named TopicBERT. This significantly reduces the number of self-attention operations - a main performance bottleneck. Consequently, our model achieves a 1.4x ($\sim40\%$) speedup with $\sim40\%$ reduction in $CO_2$ emission while retaining $99.9\%$ performance over 5 datasets.

arXiv:2010.15884

Systolic Computing on GPUs for Productive Performance

We propose a language and compiler to productively build high-performance {\it software systolic arrays} that run on GPUs. Based on a rigorous mathematical foundation (uniform recurrence equations and space-time transform), our language has a high abstraction level and covers a wide range of applications. A programmer {\it specifies} a projection of a dataflow compute onto a linear systolic array, while leaving the detailed implementation of the projection to a compiler; the compiler implements the specified projection and maps the linear systolic array to the SIMD execution units and vector registers of GPUs. In this way, both productivity and performance are achieved in the same time. This approach neatly combines loop transformations, data shuffling, and vector register allocation into a single framework. Meanwhile, many other optimizations can be applied as well; the compiler composes the optimizations together to generate efficient code. We implemented the approach on Intel GPUs. This is the first system that allows productive construction of systolic arrays on GPUs. We allow multiple projections, arbitrary projection directions and linear schedules, which can express most, if not all, systolic arrays in practice. Experiments with 1- and 2-D convolution on an Intel GEN9.5 GPU have demonstrated the generality of the approach, and its productivity in expressing various systolic designs for finding the best candidate. Although our systolic arrays are purely software running on generic SIMD hardware, compared with the GPU's specialized, hardware samplers that perform the same convolutions, some of our best designs are up to 59\% faster. Overall, this approach holds promise for productive high-performance computing on GPUs.

arXiv:2010.15156

Diagnostic data integration using deep neural networks for real-time plasma analysis

Recent advances in acquisition equipment is providing experiments with growing amounts of precise yet affordable sensors. At the same time an improved computational power, coming from new hardware resources (GPU, FPGA, ACAP), has been made available at relatively low costs. This led us to explore the possibility of completely renewing the chain of acquisition for a fusion experiment, where many high-rate sources of data, coming from different diagnostics, can be combined in a wide framework of algorithms. If on one hand adding new data sources with different diagnostics enriches our knowledge about physical aspects, on the other hand the dimensions of the overall model grow, making relations among variables more and more opaque. A new approach for the integration of such heterogeneous diagnostics, based on composition of deep variational autoencoders, could ease this problem, acting as a structural sparse regularizer. This has been applied to RFX-mod experiment data, integrating the soft X-ray linear images of plasma temperature with the magnetic state. However to ensure a real-time signal analysis, those algorithmic techniques must be adapted to run in well suited hardware. In particular it is shown that, attempting a quantization of neurons transfer functions, such models can be modified to create an embedded firmware. This firmware, approximating the deep inference model to a set of simple operations, fits well with the simple logic units that are largely abundant in FPGAs. This is the key factor that permits the use of affordable hardware with complex deep neural topology and operates them in real-time.