PyTorchKR Papers

최신 논문

총 5,576편
태그·기간 등 상세 필터

카테고리 탐색

로그인 후 이 태그 구독

arXiv:2210.04243공식 코드 · Papers with Code

Fine-Tuning Pre-trained Transformers into Decaying Fast Weights

Autoregressive Transformers are strong language models but incur O(T) complexity during per-token generation due to the self-attention mechanism. Recent work proposes kernel-based methods to approximate causal self-attention by replacing it with recurrent formulations with various update rules and feature maps to achieve O(1) time and memory complexity. We explore these approaches and find that they are unnecessarily complex, and propose a simple alternative - decaying fast weights - that runs fast on GPU, outperforms prior methods, and retains 99% of attention's performance for GPT-2. We also show competitive performance on WikiText-103 against more complex attention substitutes.

arXiv:2210.04840공식 코드 · Papers with Code

Rieoptax: Riemannian Optimization in JAX

We present Rieoptax, an open source Python library for Riemannian optimization in JAX. We show that many differential geometric primitives, such as Riemannian exponential and logarithm maps, are usually faster in Rieoptax than existing frameworks in Python, both on CPU and GPU. We support various range of basic and advanced stochastic optimization solvers like Riemannian stochastic gradient, stochastic variance reduction, and adaptive gradient methods. A distinguishing feature of the proposed toolbox is that we also support differentially private optimization on Riemannian manifolds.

arXiv:2210.04112

Leveraging progressive model and overfitting for efficient learned image compression

Deep learning is overwhelmingly dominant in the field of computer vision and image/video processing for the last decade. However, for image and video compression, it lags behind the traditional techniques based on discrete cosine transform (DCT) and linear filters. Built on top of an autoencoder architecture, learned image compression (LIC) systems have drawn enormous attention in recent years. Nevertheless, the proposed LIC systems are still inferior to the state-of-the-art traditional techniques, for example, the Versatile Video Coding (VVC/H.266) standard, due to either their compression performance or decoding complexity. Although claimed to outperform the VVC/H.266 on a limited bit rate range, some proposed LIC systems take over 40 seconds to decode a 2K image on a GPU system. In this paper, we introduce a powerful and flexible LIC framework with multi-scale progressive (MSP) probability model and latent representation overfitting (LOF) technique. With different predefined profiles, the proposed framework can achieve various balance points between compression efficiency and computational complexity. Experiments show that the proposed framework achieves 2.5%, 1.0%, and 1.3% Bjontegaard delta bit rate (BD-rate) reduction over the VVC/H.266 standard on three benchmark datasets on a wide bit rate range. More importantly, the decoding complexity is reduced from O(n) to O(1) compared to many other LIC systems, resulting in over 20 times speedup when decoding 2K images.

arXiv:2210.04195공식 코드 · Papers with Code

Online Training Through Time for Spiking Neural Networks

Spiking neural networks (SNNs) are promising brain-inspired energy-efficient models. Recent progress in training methods has enabled successful deep SNNs on large-scale tasks with low latency. Particularly, backpropagation through time (BPTT) with surrogate gradients (SG) is popularly used to achieve high performance in a very small number of time steps. However, it is at the cost of large memory consumption for training, lack of theoretical clarity for optimization, and inconsistency with the online property of biological learning and rules on neuromorphic hardware. Other works connect spike representations of SNNs with equivalent artificial neural network formulation and train SNNs by gradients from equivalent mappings to ensure descent directions. But they fail to achieve low latency and are also not online. In this work, we propose online training through time (OTTT) for SNNs, which is derived from BPTT to enable forward-in-time learning by tracking presynaptic activities and leveraging instantaneous loss and gradients. Meanwhile, we theoretically analyze and prove that gradients of OTTT can provide a similar descent direction for optimization as gradients based on spike representations under both feedforward and recurrent conditions. OTTT only requires constant training memory costs agnostic to time steps, avoiding the significant memory costs of BPTT for GPU training. Furthermore, the update rule of OTTT is in the form of three-factor Hebbian learning, which could pave a path for online on-chip learning. With OTTT, it is the first time that two mainstream supervised SNN training methods, BPTT with SG and spike representation-based training, are connected, and meanwhile in a biologically plausible form. Experiments on CIFAR-10, CIFAR-100, ImageNet, and CIFAR10-DVS demonstrate the superior performance of our method on large-scale static and neuromorphic datasets in small time steps.

arXiv:2210.03936

Cloud Native Robotic Applications with GPU Sharing on Kubernetes

In this paper we discuss our experience in teaching the Robotic Applications Programming course at ZHAW combining the use of a Kubernetes (k8s) cluster and real, heterogeneous, robotic hardware. We discuss the main advantages of our solutions in terms of seamless simulation-to-real experience for students and the main shortcomings we encountered with networking and sharing GPUs to support deep learning workloads. We describe the current and foreseen alternatives to avoid these drawbacks in future course editions and propose a more cloud-native approach to deploying multiple robotics applications on a k8s cluster.

arXiv:2210.02574

Privacy-Preserving Text Classification on BERT Embeddings with Homomorphic Encryption

Embeddings, which compress information in raw text into semantics-preserving low-dimensional vectors, have been widely adopted for their efficacy. However, recent research has shown that embeddings can potentially leak private information about sensitive attributes of the text, and in some cases, can be inverted to recover the original input text. To address these growing privacy challenges, we propose a privatization mechanism for embeddings based on homomorphic encryption, to prevent potential leakage of any piece of information in the process of text classification. In particular, our method performs text classification on the encryption of embeddings from state-of-the-art models like BERT, supported by an efficient GPU implementation of CKKS encryption scheme. We show that our method offers encrypted protection of BERT embeddings, while largely preserving their utility on downstream text classification tasks.

arXiv:2210.03052공식 코드 · Papers with Code

ByteTransformer: A High-Performance Transformer Boosted for Variable-Length Inputs

Transformers have become keystone models in natural language processing over the past decade. They have achieved great popularity in deep learning applications, but the increasing sizes of the parameter spaces required by transformer models generate a commensurate need to accelerate performance. Natural language processing problems are also routinely faced with variable-length sequences, as word counts commonly vary among sentences. Existing deep learning frameworks pad variable-length sequences to a maximal length, which adds significant memory and computational overhead. In this paper, we present ByteTransformer, a high-performance transformer boosted for variable-length inputs. We propose a padding-free algorithm that liberates the entire transformer from redundant computations on zero padded tokens. In addition to algorithmic-level optimization, we provide architecture-aware optimizations for transformer functional modules, especially the performance-critical algorithm Multi-Head Attention (MHA). Experimental results on an NVIDIA A100 GPU with variable-length sequence inputs validate that our fused MHA outperforms PyTorch by 6.13x. The end-to-end performance of ByteTransformer for a forward BERT transformer surpasses state-of-the-art transformer frameworks, such as PyTorch JIT, TensorFlow XLA, Tencent TurboTransformer, Microsoft DeepSpeed-Inference and NVIDIA FasterTransformer, by 87\%, 131\%, 138\%, 74\% and 55\%, respectively. We also demonstrate the general applicability of our optimization methods to other BERT-like models, including ALBERT, DistilBERT, and DeBERTa.

arXiv:2210.02023공식 코드 · Papers with Code

DreamShard: Generalizable Embedding Table Placement for Recommender Systems

We study embedding table placement for distributed recommender systems, which aims to partition and place the tables on multiple hardware devices (e.g., GPUs) to balance the computation and communication costs. Although prior work has explored learning-based approaches for the device placement of computational graphs, embedding table placement remains to be a challenging problem because of 1) the operation fusion of embedding tables, and 2) the generalizability requirement on unseen placement tasks with different numbers of tables and/or devices. To this end, we present DreamShard, a reinforcement learning (RL) approach for embedding table placement. DreamShard achieves the reasoning of operation fusion and generalizability with 1) a cost network to directly predict the costs of the fused operation, and 2) a policy network that is efficiently trained on an estimated Markov decision process (MDP) without real GPU execution, where the states and the rewards are estimated with the cost network. Equipped with sum and max representation reductions, the two networks can directly generalize to any unseen tasks with different numbers of tables and/or devices without fine-tuning. Extensive experiments show that DreamShard substantially outperforms the existing human expert and RNN-based strategies with up to 19% speedup over the strongest baseline on large-scale synthetic tables and our production tables. The code is available at https://github.com/daochenzha/dreamshard

arXiv:2210.01402

Streaming Video Analytics On The Edge With Asynchronous Cloud Support

Emerging Internet of Things (IoT) and mobile computing applications are expected to support latency-sensitive deep neural network (DNN) workloads. To realize this vision, the Internet is evolving towards an edge-computing architecture, where computing infrastructure is located closer to the end device to help achieve low latency. However, edge computing may have limited resources compared to cloud environments and thus, cannot run large DNN models that often have high accuracy. In this work, we develop REACT, a framework that leverages cloud resources to execute large DNN models with higher accuracy to improve the accuracy of models running on edge devices. To do so, we propose a novel edge-cloud fusion algorithm that fuses edge and cloud predictions, achieving low latency and high accuracy. We extensively evaluate our approach and show that our approach can significantly improve the accuracy compared to baseline approaches. We focus specifically on object detection in videos (applicable in many video analytics scenarios) and show that the fused edge-cloud predictions can outperform the accuracy of edge-only and cloud-only scenarios by as much as 50%. We also show that REACT can achieve good performance across tradeoff points by choosing a wide range of system parameters to satisfy use-case specific constraints, such as limited network bandwidth or GPU cycles.

arXiv:2210.00286

NeuroEvo: A Cloud-based Platform for Automated Design and Training of Neural Networks using Evolutionary and Particle Swarm Algorithms

Evolutionary algorithms (EAs) provide unique advantages for optimizing neural networks in complex search spaces. This paper introduces a new web platform, NeuroEvo (neuroevo.io), that allows users to interactively design and train neural network classifiers using evolutionary and particle swarm algorithms. The classification problem and training data are provided by the user and, upon completion of the training process, the best classifier is made available to download and implement in Python, Java, and JavaScript. NeuroEvo is a cloud-based application that leverages GPU parallelization to improve the speed with which the independent evolutionary steps, such as mutation, crossover, and fitness evaluation, are executed across the population. This paper outlines the training algorithms and opportunities for users to specify design decisions and hyperparameter settings. The algorithms described in this paper are also made available as a Python package, neuroevo (PyPI: https://pypi.org/project/neuroevo/).

arXiv:2210.00893

Combining Efficient and Precise Sign Language Recognition: Good pose estimation library is all you need

Sign language recognition could significantly improve the user experience for d/Deaf people with the general consumer technology, such as IoT devices or videoconferencing. However, current sign language recognition architectures are usually computationally heavy and require robust GPU-equipped hardware to run in real-time. Some models aim for lower-end devices (such as smartphones) by minimizing their size and complexity, which leads to worse accuracy. This highly scrutinizes accurate in-the-wild applications. We build upon the SPOTER architecture, which belongs to the latter group of light methods, as it came close to the performance of large models employed for this task. By substituting its original third-party pose estimation module with the MediaPipe library, we achieve an overall state-of-the-art result on the WLASL100 dataset. Significantly, our method beats previous larger architectures while still being twice as computationally efficient and almost $11$ times faster on inference when compared to a relevant benchmark. To demonstrate our method's combined efficiency and precision, we built an online demo that enables users to translate sign lemmas of American sign language in their browsers. This is the first publicly available online application demonstrating this task to the best of our knowledge.

arXiv:2210.00129

An In-depth Study of Stochastic Backpropagation

In this paper, we provide an in-depth study of Stochastic Backpropagation (SBP) when training deep neural networks for standard image classification and object detection tasks. During backward propagation, SBP calculates the gradients by only using a subset of feature maps to save the GPU memory and computational cost. We interpret SBP as an efficient way to implement stochastic gradient decent by performing backpropagation dropout, which leads to considerable memory saving and training process speedup, with a minimal impact on the overall model accuracy. We offer some good practices to apply SBP in training image recognition models, which can be adopted in learning a wide range of deep neural networks. Experiments on image classification and object detection show that SBP can save up to 40% of GPU memory with less than 1% accuracy degradation.

arXiv:2210.00973

NCVX: A General-Purpose Optimization Solver for Constrained Machine and Deep Learning

Imposing explicit constraints is relatively new but increasingly pressing in deep learning, stimulated by, e.g., trustworthy AI that performs robust optimization over complicated perturbation sets and scientific applications that need to respect physical laws and constraints. However, it can be hard to reliably solve constrained deep learning problems without optimization expertise. The existing deep learning frameworks do not admit constraints. General-purpose optimization packages can handle constraints but do not perform auto-differentiation and have trouble dealing with nonsmoothness. In this paper, we introduce a new software package called NCVX, whose initial release contains the solver PyGRANSO, a PyTorch-enabled general-purpose optimization package for constrained machine/deep learning problems, the first of its kind. NCVX inherits auto-differentiation, GPU acceleration, and tensor variables from PyTorch, and is built on freely available and widely used open-source frameworks. NCVX is available at https://ncvx.org, with detailed documentation and numerous examples from machine/deep learning and other fields.

arXiv:2210.01035공식 코드 · Papers with Code

Expediting Large-Scale Vision Transformer for Dense Prediction without Fine-tuning

Vision transformers have recently achieved competitive results across various vision tasks but still suffer from heavy computation costs when processing a large number of tokens. Many advanced approaches have been developed to reduce the total number of tokens in large-scale vision transformers, especially for image classification tasks. Typically, they select a small group of essential tokens according to their relevance with the class token, then fine-tune the weights of the vision transformer. Such fine-tuning is less practical for dense prediction due to the much heavier computation and GPU memory cost than image classification. In this paper, we focus on a more challenging problem, i.e., accelerating large-scale vision transformers for dense prediction without any additional re-training or fine-tuning. In response to the fact that high-resolution representations are necessary for dense prediction, we present two non-parametric operators, a token clustering layer to decrease the number of tokens and a token reconstruction layer to increase the number of tokens. The following steps are performed to achieve this: (i) we use the token clustering layer to cluster the neighboring tokens together, resulting in low-resolution representations that maintain the spatial structures; (ii) we apply the following transformer layers only to these low-resolution representations or clustered tokens; and (iii) we use the token reconstruction layer to re-create the high-resolution representations from the refined low-resolution representations. The results obtained by our method are promising on five dense prediction tasks, including object detection, semantic segmentation, panoptic segmentation, instance segmentation, and depth estimation.

arXiv:2210.00882

MSRL: Distributed Reinforcement Learning with Dataflow Fragments

Reinforcement learning (RL) trains many agents, which is resource-intensive and must scale to large GPU clusters. Different RL training algorithms offer different opportunities for distributing and parallelising the computation. Yet, current distributed RL systems tie the definition of RL algorithms to their distributed execution: they hard-code particular distribution strategies and only accelerate specific parts of the computation (e.g. policy network updates) on GPU workers. Fundamentally, current systems lack abstractions that decouple RL algorithms from their execution. We describe MindSpore Reinforcement Learning (MSRL), a distributed RL training system that supports distribution policies that govern how RL training computation is parallelised and distributed on cluster resources, without requiring changes to the algorithm implementation. MSRL introduces the new abstraction of a fragmented dataflow graph, which maps Python functions from an RL algorithm's training loop to parallel computational fragments. Fragments are executed on different devices by translating them to low-level dataflow representations, e.g. computational graphs as supported by deep learning engines, CUDA implementations or multi-threaded CPU processes. We show that MSRL subsumes the distribution strategies of existing systems, while scaling RL training to 64 GPUs.

arXiv:2209.15486공식 코드 · Papers with Code

Graph Neural Networks for Link Prediction with Subgraph Sketching

Many Graph Neural Networks (GNNs) perform poorly compared to simple heuristics on Link Prediction (LP) tasks. This is due to limitations in expressive power such as the inability to count triangles (the backbone of most LP heuristics) and because they can not distinguish automorphic nodes (those having identical structural roles). Both expressiveness issues can be alleviated by learning link (rather than node) representations and incorporating structural features such as triangle counts. Since explicit link representations are often prohibitively expensive, recent works resorted to subgraph-based methods, which have achieved state-of-the-art performance for LP, but suffer from poor efficiency due to high levels of redundancy between subgraphs. We analyze the components of subgraph GNN (SGNN) methods for link prediction. Based on our analysis, we propose a novel full-graph GNN called ELPH (Efficient Link Prediction with Hashing) that passes subgraph sketches as messages to approximate the key components of SGNNs without explicit subgraph construction. ELPH is provably more expressive than Message Passing GNNs (MPNNs). It outperforms existing SGNN models on many standard LP benchmarks while being orders of magnitude faster. However, it shares the common GNN limitation that it is only efficient when the dataset fits in GPU memory. Accordingly, we develop a highly scalable model, called BUDDY, which uses feature precomputation to circumvent this limitation without sacrificing predictive performance. Our experiments show that BUDDY also outperforms SGNNs on standard LP benchmarks while being highly scalable and faster than ELPH.

arXiv:2209.14569공식 코드 · Papers with Code

COLO: A Contrastive Learning based Re-ranking Framework for One-Stage Summarization

Traditional training paradigms for extractive and abstractive summarization systems always only use token-level or sentence-level training objectives. However, the output summary is always evaluated from summary-level which leads to the inconsistency in training and evaluation. In this paper, we propose a Contrastive Learning based re-ranking framework for one-stage summarization called COLO. By modeling a contrastive objective, we show that the summarization model is able to directly generate summaries according to the summary-level score without additional modules and parameters. Extensive experiments demonstrate that COLO boosts the extractive and abstractive results of one-stage systems on CNN/DailyMail benchmark to 44.58 and 46.33 ROUGE-1 score while preserving the parameter efficiency and inference efficiency. Compared with state-of-the-art multi-stage systems, we save more than 100 GPU training hours and obtaining 3~8 speed-up ratio during inference while maintaining comparable results.

arXiv:2209.14499

NVRadarNet: Real-Time Radar Obstacle and Free Space Detection for Autonomous Driving

Detecting obstacles is crucial for safe and efficient autonomous driving. To this end, we present NVRadarNet, a deep neural network (DNN) that detects dynamic obstacles and drivable free space using automotive RADAR sensors. The network utilizes temporally accumulated data from multiple RADAR sensors to detect dynamic obstacles and compute their orientation in a top-down bird's-eye view (BEV). The network also regresses drivable free space to detect unclassified obstacles. Our DNN is the first of its kind to utilize sparse RADAR signals in order to perform obstacle and free space detection in real time from RADAR data only. The network has been successfully used for perception on our autonomous vehicles in real self-driving scenarios. The network runs faster than real time on an embedded GPU and shows good generalization across geographic regions.

arXiv:2209.14829

Lightweight Monocular Depth Estimation with an Edge Guided Network

Monocular depth estimation is an important task that can be applied to many robotic applications. Existing methods focus on improving depth estimation accuracy via training increasingly deeper and wider networks, however these suffer from large computational complexity. Recent studies found that edge information are important cues for convolutional neural networks (CNNs) to estimate depth. Inspired by the above observations, we present a novel lightweight Edge Guided Depth Estimation Network (EGD-Net) in this study. In particular, we start out with a lightweight encoder-decoder architecture and embed an edge guidance branch which takes as input image gradients and multi-scale feature maps from the backbone to learn the edge attention features. In order to aggregate the context information and edge attention features, we design a transformer-based feature aggregation module (TRFA). TRFA captures the long-range dependencies between the context information and edge attention features through cross-attention mechanism. We perform extensive experiments on the NYU depth v2 dataset. Experimental results show that the proposed method runs about 96 fps on a Nvidia GTX 1080 GPU whilst achieving the state-of-the-art performance in terms of accuracy.

arXiv:2209.14981공식 코드 · Papers with Code

Stop Wasting My Time! Saving Days of ImageNet and BERT Training with Latest Weight Averaging

Training vision or language models on large datasets can take days, if not weeks. We show that averaging the weights of the k latest checkpoints, each collected at the end of an epoch, can speed up the training progression in terms of loss and accuracy by dozens of epochs, corresponding to time savings up to ~68 and ~30 GPU hours when training a ResNet50 on ImageNet and RoBERTa-Base model on WikiText-103, respectively. We also provide the code and model checkpoint trajectory to reproduce the results and facilitate research on reusing historical weights for faster convergence.