PyTorchKR Papers

최신 논문

총 5,576편
태그·기간 등 상세 필터

카테고리 탐색

로그인 후 이 태그 구독

arXiv:2111.12055

Generating GPU Compiler Heuristics using Reinforcement Learning

GPU compilers are complex software programs with many optimizations specific to target hardware. These optimizations are often controlled by heuristics hand-designed by compiler experts using time- and resource-intensive processes. In this paper, we developed a GPU compiler autotuning framework that uses off-policy deep reinforcement learning to generate heuristics that improve the frame rates of graphics applications. Furthermore, we demonstrate the resilience of these learned heuristics to frequent compiler updates by analyzing their stability across a year of code check-ins without retraining. We show that our machine learning-based compiler autotuning framework matches or surpasses the frame rates for 98% of graphics benchmarks with an average uplift of 1.6% up to 15.8%.

arXiv:2111.11840

Subgraph Permutation Equivariant Networks

In this work we develop a new method, named Sub-graph Permutation Equivariant Networks (SPEN), which provides a framework for building graph neural networks that operate on sub-graphs, while using a base update function that is permutation equivariant, that are equivariant to a novel choice of automorphism group. Message passing neural networks have been shown to be limited in their expressive power and recent approaches to over come this either lack scalability or require structural information to be encoded into the feature space. The general framework presented here overcomes the scalability issues associated with global permutation equivariance by operating more locally on sub-graphs. In addition, through operating on sub-graphs the expressive power of higher-dimensional global permutation equivariant networks is improved; this is due to fact that two non-distinguishable graphs often contain distinguishable sub-graphs. Furthermore, the proposed framework only requires a choice of $k$-hops for creating ego-network sub-graphs and a choice of representation space to be used for each layer, which makes the method easily applicable across a range of graph based domains. We experimentally validate the method on a range of graph benchmark classification tasks, demonstrating statistically indistinguishable results from the state-of-the-art on six out of seven benchmarks. Further, we demonstrate that the use of local update functions offers a significant improvement in GPU memory over global methods.

arXiv:2111.11215공식 코드 · Papers with Code

Direct Voxel Grid Optimization: Super-fast Convergence for Radiance Fields Reconstruction

We present a super-fast convergence approach to reconstructing the per-scene radiance field from a set of images that capture the scene with known poses. This task, which is often applied to novel view synthesis, is recently revolutionized by Neural Radiance Field (NeRF) for its state-of-the-art quality and flexibility. However, NeRF and its variants require a lengthy training time ranging from hours to days for a single scene. In contrast, our approach achieves NeRF-comparable quality and converges rapidly from scratch in less than 15 minutes with a single GPU. We adopt a representation consisting of a density voxel grid for scene geometry and a feature voxel grid with a shallow network for complex view-dependent appearance. Modeling with explicit and discretized volume representations is not new, but we propose two simple yet non-trivial techniques that contribute to fast convergence speed and high-quality output. First, we introduce the post-activation interpolation on voxel density, which is capable of producing sharp surfaces in lower grid resolution. Second, direct voxel density optimization is prone to suboptimal geometry solutions, so we robustify the optimization process by imposing several priors. Finally, evaluation on five inward-facing benchmarks shows that our method matches, if not surpasses, NeRF's quality, yet it only takes about 15 minutes to train from scratch for a new scene.

arXiv:2111.10270공식 코드 · Papers with Code

FastDOG: Fast Discrete Optimization on GPU

We present a massively parallel Lagrange decomposition method for solving 0--1 integer linear programs occurring in structured prediction. We propose a new iterative update scheme for solving the Lagrangean dual and a perturbation technique for decoding primal solutions. For representing subproblems we follow Lange et al. (2021) and use binary decision diagrams (BDDs). Our primal and dual algorithms require little synchronization between subproblems and optimization over BDDs needs only elementary operations without complicated control flow. This allows us to exploit the parallelism offered by GPUs for all components of our method. We present experimental results on combinatorial problems from MAP inference for Markov Random Fields, quadratic assignment and cell tracking for developmental biology. Our highly parallel GPU implementation improves upon the running times of the algorithms from Lange et al. (2021) by up to an order of magnitude. In particular, we come close to or outperform some state-of-the-art specialized heuristics while being problem agnostic. Our implementation is available at https://github.com/LPMP/BDD.

arXiv:2111.10247공식 코드 · Papers with Code

Fast and Data-Efficient Training of Rainbow: an Experimental Study on Atari

Across the Arcade Learning Environment, Rainbow achieves a level of performance competitive with humans and modern RL algorithms. However, attaining this level of performance requires large amounts of data and hardware resources, making research in this area computationally expensive and use in practical applications often infeasible. This paper's contribution is threefold: We (1) propose an improved version of Rainbow, seeking to drastically reduce Rainbow's data, training time, and compute requirements while maintaining its competitive performance; (2) we empirically demonstrate the effectiveness of our approach through experiments on the Arcade Learning Environment, and (3) we conduct a number of ablation studies to investigate the effect of the individual proposed modifications. Our improved version of Rainbow reaches a median human normalized score close to classic Rainbow's, while using 20 times less data and requiring only 7.5 hours of training time on a single GPU. We also provide our full implementation including pre-trained models.

arXiv:2111.10206공식 코드 · Papers with Code

Subspace Graph Physics: Real-Time Rigid Body-Driven Granular Flow Simulation

An important challenge in robotics is understanding the interactions between robots and deformable terrains that consist of granular material. Granular flows and their interactions with rigid bodies still pose several open questions. A promising direction for accurate, yet efficient, modeling is using continuum methods. Also, a new direction for real-time physics modeling is the use of deep learning. This research advances machine learning methods for modeling rigid body-driven granular flows, for application to terrestrial industrial machines as well as space robotics (where the effect of gravity is an important factor). In particular, this research considers the development of a subspace machine learning simulation approach. To generate training datasets, we utilize our high-fidelity continuum method, material point method (MPM). Principal component analysis (PCA) is used to reduce the dimensionality of data. We show that the first few principal components of our high-dimensional data keep almost the entire variance in data. A graph network simulator (GNS) is trained to learn the underlying subspace dynamics. The learned GNS is then able to predict particle positions and interaction forces with good accuracy. More importantly, PCA significantly enhances the time and memory efficiency of GNS in both training and rollout. This enables GNS to be trained using a single desktop GPU with moderate VRAM. This also makes the GNS real-time on large-scale 3D physics configurations (700x faster than our continuum method).

arXiv:2111.09957공식 코드 · Papers with Code

Rethinking Dilated Convolution for Real-time Semantic Segmentation

The field-of-view is an important metric when designing a model for semantic segmentation. To obtain a large field-of-view, previous approaches generally choose to rapidly downsample the resolution, usually with average poolings or stride 2 convolutions. We take a different approach by using dilated convolutions with large dilation rates throughout the backbone, allowing the backbone to easily tune its field-of-view by adjusting its dilation rates, and show that it's competitive with existing approaches. To effectively use the dilated convolution, we show a simple upper bound on the dilation rate in order to not leave gaps in between the convolutional weights, and design an SE-ResNeXt inspired block structure that uses two parallel $3\times 3$ convolutions with different dilation rates to preserve the local details. Manually tuning the dilation rates for every block can be difficult, so we also introduce a differentiable neural architecture search method that uses gradient descent to optimize the dilation rates. In addition, we propose a lightweight decoder that restores local information better than common alternatives. To demonstrate the effectiveness of our approach, our model RegSeg achieves competitive results on real-time Cityscapes and CamVid datasets. Using a T4 GPU with mixed precision, RegSeg achieves 78.3 mIOU on Cityscapes test set at $37$ FPS, and 80.9 mIOU on CamVid test set at $112$ FPS, both without ImageNet pretraining.

arXiv:2111.09714공식 코드 · Papers with Code

You Only Sample (Almost) Once: Linear Cost Self-Attention Via Bernoulli Sampling

Transformer-based models are widely used in natural language processing (NLP). Central to the transformer model is the self-attention mechanism, which captures the interactions of token pairs in the input sequences and depends quadratically on the sequence length. Training such models on longer sequences is expensive. In this paper, we show that a Bernoulli sampling attention mechanism based on Locality Sensitive Hashing (LSH), decreases the quadratic complexity of such models to linear. We bypass the quadratic cost by considering self-attention as a sum of individual tokens associated with Bernoulli random variables that can, in principle, be sampled at once by a single hash (although in practice, this number may be a small constant). This leads to an efficient sampling scheme to estimate self-attention which relies on specific modifications of LSH (to enable deployment on GPU architectures). We evaluate our algorithm on the GLUE benchmark with standard 512 sequence length where we see favorable performance relative to a standard pretrained Transformer. On the Long Range Arena (LRA) benchmark, for evaluating performance on long sequences, our method achieves results consistent with softmax self-attention but with sizable speed-ups and memory savings and often outperforms other efficient self-attention methods. Our code is available at https://github.com/mlpen/YOSO

arXiv:2111.09451공식 코드 · Papers with Code

Benchmarking and scaling of deep learning models for land cover image classification

The availability of the sheer volume of Copernicus Sentinel-2 imagery has created new opportunities for exploiting deep learning (DL) methods for land use land cover (LULC) image classification. However, an extensive set of benchmark experiments is currently lacking, i.e. DL models tested on the same dataset, with a common and consistent set of metrics, and in the same hardware. In this work, we use the BigEarthNet Sentinel-2 dataset to benchmark for the first time different state-of-the-art DL models for the multi-label, multi-class LULC image classification problem, contributing with an exhaustive zoo of 60 trained models. Our benchmark includes standard CNNs, as well as non-convolutional methods. We put to the test EfficientNets and Wide Residual Networks (WRN) architectures, and leverage classification accuracy, training time and inference rate. Furthermore, we propose to use the EfficientNet framework for the compound scaling of a lightweight WRN. Enhanced with an Efficient Channel Attention mechanism, our scaled lightweight model emerged as the new state-of-the-art. It achieves 4.5% higher averaged F-Score classification accuracy for all 19 LULC classes compared to a standard ResNet50 baseline model, with an order of magnitude less trainable parameters. We provide access to all trained models, along with our code for distributed training on multiple GPU nodes. This model zoo of pre-trained encoders can be used for transfer learning and rapid prototyping in different remote sensing tasks that use Sentinel-2 data, instead of exploiting backbone models trained with data from a different domain, e.g., from ImageNet. We validate their suitability for transfer learning in different datasets of diverse volumes. Our top-performing WRN achieves state-of-the-art performance (71.1% F-Score) on the SEN12MS dataset while being exposed to only a small fraction of the training dataset.

arXiv:2111.08617공식 코드 · Papers with Code

CGX: Adaptive System Support for Communication-Efficient Deep Learning

The ability to scale out training workloads has been one of the key performance enablers of deep learning. The main scaling approach is data-parallel GPU-based training, which has been boosted by hardware and software support for highly efficient point-to-point communication, and in particular via hardware bandwidth overprovisioning. Overprovisioning comes at a cost: there is an order of magnitude price difference between "cloud-grade" servers with such support, relative to their popular "consumer-grade" counterparts, although single server-grade and consumer-grade GPUs can have similar computational envelopes. In this paper, we show that the costly hardware overprovisioning approach can be supplanted via algorithmic and system design, and propose a framework called CGX, which provides efficient software support for compressed communication in ML applications, for both multi-GPU single-node training, as well as larger-scale multi-node training. CGX is based on two technical advances: \emph{At the system level}, it relies on a re-developed communication stack for ML frameworks, which provides flexible, highly-efficient support for compressed communication. \emph{At the application level}, it provides \emph{seamless, parameter-free} integration with popular frameworks, so that end-users do not have to modify training recipes, nor significant training code. This is complemented by a \emph{layer-wise adaptive compression} technique which dynamically balances compression gains with accuracy preservation. CGX integrates with popular ML frameworks, providing up to 3X speedups for multi-GPU nodes based on commodity hardware, and order-of-magnitude improvements in the multi-node setting, with negligible impact on accuracy.

arXiv:2111.07684공식 코드 · Papers with Code

AutoGMap: Learning to Map Large-scale Sparse Graphs on Memristive Crossbars

The sparse representation of graphs has shown great potential for accelerating the computation of graph applications (e.g., Social Networks, Knowledge Graphs) on traditional computing architectures (CPU, GPU, or TPU). But the exploration of large-scale sparse graph computing on processing-in-memory (PIM) platforms (typically with memristive crossbars) is still in its infancy. To implement the computation or storage of large-scale or batch graphs on memristive crossbars, a natural assumption is that a large-scale crossbar is demanded, but with low utilization. Some recent works question this assumption, to avoid the waste of storage and computational resource, the fixed-size or progressively scheduled ''block partition'' schemes are proposed. However, these methods are coarse-grained or static, and are not effectively sparsity-aware. This work proposes the dynamic sparsity-aware mapping scheme generating method that models the problem with a sequential decision-making model, and optimizes it by reinforcement learning (RL) algorithm (REINFORCE). Our generating model (LSTM, combined with the dynamic-fill scheme) generates remarkable mapping performance on a small-scale graph/matrix data (complete mapping costs 43% area of the original matrix) and two large-scale matrix data (costing 22.5% area on qh882 and 17.1% area on qh1484). Our method may be extended to sparse graph computing on other PIM architectures, not limited to the memristive device-based platforms.

arXiv:2111.07722

Stacked BNAS: Rethinking Broad Convolutional Neural Network for Neural Architecture Search

Different from other deep scalable architecture-based NAS approaches, Broad Neural Architecture Search (BNAS) proposes a broad scalable architecture which consists of convolution and enhancement blocks, dubbed Broad Convolutional Neural Network (BCNN), as the search space for amazing efficiency improvement. BCNN reuses the topologies of cells in the convolution block so that BNAS can employ few cells for efficient search. Moreover, multi-scale feature fusion and knowledge embedding are proposed to improve the performance of BCNN with shallow topology. However, BNAS suffers some drawbacks: 1) insufficient representation diversity for feature fusion and enhancement and 2) time consumption of knowledge embedding design by human experts. This paper proposes Stacked BNAS, whose search space is a developed broad scalable architecture named Stacked BCNN, with better performance than BNAS. On the one hand, Stacked BCNN treats mini BCNN as a basic block to preserve comprehensive representation and deliver powerful feature extraction ability. For multi-scale feature enhancement, each mini BCNN feeds the outputs of deep and broad cells to the enhancement cell. For multi-scale feature fusion, each mini BCNN feeds the outputs of deep, broad and enhancement cells to the output node. On the other hand, Knowledge Embedding Search (KES) is proposed to learn appropriate knowledge embeddings in a differentiable way. Moreover, the basic unit of KES is an over-parameterized knowledge embedding module that consists of all possible candidate knowledge embeddings. Experimental results show that 1) Stacked BNAS obtains better performance than BNAS-v2 on both CIFAR-10 and ImageNet, 2) the proposed KES algorithm contributes to reducing the parameters of the learned architecture with satisfactory performance, and 3) Stacked BNAS delivers a state-of-the-art efficiency of 0.02 GPU days.

arXiv:2111.06340공식 코드 · Papers with Code

ParaFold: Paralleling AlphaFold for Large-Scale Predictions

AlphaFold predicts protein structures from the amino acid sequence at or near experimental resolution, solving the 50-year-old protein folding challenge, leading to progress by transforming large-scale genomics data into protein structures. AlphaFold will also greatly change the scientific research model from low-throughput to high-throughput manner. The AlphaFold framework is a mixture of two types of workloads: MSA construction based on CPUs and model inference on GPUs. The first CPU stage dominates the overall runtime, taking hours for a single protein due to the large database sizes and I/O bottlenecks. However, GPUs in this CPU stage remain idle, resulting in low GPU utilization and restricting the capacity of large-scale structure predictions. Therefore, we proposed ParaFold, an open-source parallel version of AlphaFold for high throughput protein structure predictions. ParaFold separates the CPU and GPU parts to enable large-scale structure predictions. ParaFold also effectively reduces the CPU and GPU runtime with two optimizations without compromising the quality of prediction results: using multi-threaded parallelism on CPUs and using optimized JAX compilation on GPUs. We evaluated ParaFold with three datasets of different size and protein lengths. We evaluated the accuracy and efficiency of optimizations on CPUs and GPUs, and showed the large-scale prediction capability by running ParaFold inferences of 19,704 small proteins in five hours on one NVIDIA DGX-2. Using the JAX compile optimization, ParaFold attained a 13.8X average speedup over AlphaFold. ParaFold offers a rapid and effective approach for high-throughput structure predictions, leveraging the predictive power by running on supercomputers, with shorter time, and at a lower cost. The development of ParaFold will greatly speed up high-throughput studies and render the protein "structure-omics" feasible.

arXiv:2111.05936

SPA-GCN: Efficient and Flexible GCN Accelerator with an Application for Graph Similarity Computation

While there have been many studies on hardware acceleration for deep learning on images, there has been a rather limited focus on accelerating deep learning applications involving graphs. The unique characteristics of graphs, such as the irregular memory access and dynamic parallelism, impose several challenges when the algorithm is mapped to a CPU or GPU. To address these challenges while exploiting all the available sparsity, we propose a flexible architecture called SPA-GCN for accelerating Graph Convolutional Networks (GCN), the core computation unit in deep learning algorithms on graphs. The architecture is specialized for dealing with many small graphs since the graph size has a significant impact on design considerations. In this context, we use SimGNN, a neural-network-based graph matching algorithm, as a case study to demonstrate the effectiveness of our architecture. The experimental results demonstrate that SPA-GCN can deliver a high speedup compared to a multi-core CPU implementation and a GPU implementation, showing the efficiency of our design.

arXiv:2111.05894

Graph Neural Network Training with Data Tiering

Graph Neural Networks (GNNs) have shown success in learning from graph-structured data, with applications to fraud detection, recommendation, and knowledge graph reasoning. However, training GNN efficiently is challenging because: 1) GPU memory capacity is limited and can be insufficient for large datasets, and 2) the graph-based data structure causes irregular data access patterns. In this work, we provide a method to statistical analyze and identify more frequently accessed data ahead of GNN training. Our data tiering method not only utilizes the structure of input graph, but also an insight gained from actual GNN training process to achieve a higher prediction result. With our data tiering method, we additionally provide a new data placement and access strategy to further minimize the CPU-GPU communication overhead. We also take into account of multi-GPU GNN training as well and we demonstrate the effectiveness of our strategy in a multi-GPU system. The evaluation results show that our work reduces CPU-GPU traffic by 87-95% and improves the training speed of GNN over the existing solutions by 1.6-2.1x on graphs with hundreds of millions of nodes and billions of edges.

arXiv:2111.05188

FinRL-Podracer: High Performance and Scalable Deep Reinforcement Learning for Quantitative Finance

Machine learning techniques are playing more and more important roles in finance market investment. However, finance quantitative modeling with conventional supervised learning approaches has a number of limitations. The development of deep reinforcement learning techniques is partially addressing these issues. Unfortunately, the steep learning curve and the difficulty in quick modeling and agile development are impeding finance researchers from using deep reinforcement learning in quantitative trading. In this paper, we propose an RLOps in finance paradigm and present a FinRL-Podracer framework to accelerate the development pipeline of deep reinforcement learning (DRL)-driven trading strategy and to improve both trading performance and training efficiency. FinRL-Podracer is a cloud solution that features high performance and high scalability and promises continuous training, continuous integration, and continuous delivery of DRL-driven trading strategies, facilitating a rapid transformation from algorithmic innovations into a profitable trading strategy. First, we propose a generational evolution mechanism with an ensemble strategy to improve the trading performance of a DRL agent, and schedule the training of a DRL algorithm onto a GPU cloud via multi-level mapping. Then, we carry out the training of DRL components with high-performance optimizations on GPUs. Finally, we evaluate the FinRL-Podracer framework for a stock trend prediction task on an NVIDIA DGX SuperPOD cloud. FinRL-Podracer outperforms three popular DRL libraries Ray RLlib, Stable Baseline 3 and FinRL, i.e., 12% \sim 35% improvements in annual return, 0.1 \sim 0.6 improvements in Sharpe ratio and 3 times \sim 7 times speed-up in training time. We show the high scalability by training a trading agent in 10 minutes with $80$ A100 GPUs, on NASDAQ-100 constituent stocks with minute-level data over 10 years.

arXiv:2111.04628

Accelerating GAN training using highly parallel hardware on public cloud

With the increasing number of Machine and Deep Learning applications in High Energy Physics, easy access to dedicated infrastructure represents a requirement for fast and efficient R&D. This work explores different types of cloud services to train a Generative Adversarial Network (GAN) in a parallel environment, using Tensorflow data parallel strategy. More specifically, we parallelize the training process on multiple GPUs and Google Tensor Processing Units (TPU) and we compare two algorithms: the TensorFlow built-in logic and a custom loop, optimised to have higher control of the elements assigned to each GPU worker or TPU core. The quality of the generated data is compared to Monte Carlo simulation. Linear speed-up of the training process is obtained, while retaining most of the performance in terms of physics results. Additionally, we benchmark the aforementioned approaches, at scale, over multiple GPU nodes, deploying the training process on different public cloud providers, seeking for overall efficiency and cost-effectiveness. The combination of data science, cloud deployment options and associated economics allows to burst out heterogeneously, exploring the full potential of cloud-based services.

arXiv:2111.04060공식 코드 · Papers with Code

Are we ready for a new paradigm shift? A Survey on Visual Deep MLP

Recently, the proposed deep MLP models have stirred up a lot of interest in the vision community. Historically, the availability of larger datasets combined with increased computing capacity leads to paradigm shifts. This review paper provides detailed discussions on whether MLP can be a new paradigm for computer vision. We compare the intrinsic connections and differences between convolution, self-attention mechanism, and Token-mixing MLP in detail. Advantages and limitations of Token-mixing MLP are provided, followed by careful analysis of recent MLP-like variants, from module design to network architecture, and their applications. In the GPU era, the locally and globally weighted summations are the current mainstreams, represented by the convolution and self-attention mechanism, as well as MLP. We suggest the further development of paradigm to be considered alongside the next-generation computing devices.

arXiv:2111.03892

TND-NAS: Towards Non-differentiable Objectives in Progressive Differentiable NAS Framework

Differentiable architecture search has gradually become the mainstream research topic in the field of Neural Architecture Search (NAS) for its high efficiency compared with the early NAS methods. Recent differentiable NAS also aims at further improving the search performance and reducing the GPU-memory consumption. However, these methods are no longer naturally capable of tackling the non-differentiable objectives, e.g., energy, resource-constrained efficiency, and other metrics, let alone the multi-objective search demands. Researches in the multi-objective NAS field target this but requires vast computational resources cause of the sole optimization of each candidate architecture. In light of this discrepancy, we propose the TND-NAS, which is with the merits of the high efficiency in differentiable NAS framework and the compatibility among non-differentiable metrics in Multi-objective NAS. Under the differentiable NAS framework, with the continuous relaxation of the search space, TND-NAS has the architecture parameters been optimized in discrete space, while resorting to the progressive search space shrinking by architecture parameters. Our representative experiment takes two objectives (Parameters, Accuracy) as an example, we achieve a series of high-performance compact architectures on CIFAR10 (1.09M/3.3%, 2.4M/2.95%, 9.57M/2.54%) and CIFAR100 (2.46M/18.3%, 5.46/16.73%, 12.88/15.20%) datasets. Favorably, compared with other multi-objective NAS methods, TND-NAS is less time-consuming (1.3 GPU-days on NVIDIA 1080Ti, 1/6 of that in NSGA-Net), and can be conveniently adapted to real-world NAS scenarios (resource-constrained, platform-specialized).

arXiv:2111.03759공식 코드 · Papers with Code

MQBench: Towards Reproducible and Deployable Model Quantization Benchmark

Model quantization has emerged as an indispensable technique to accelerate deep learning inference. While researchers continue to push the frontier of quantization algorithms, existing quantization work is often unreproducible and undeployable. This is because researchers do not choose consistent training pipelines and ignore the requirements for hardware deployments. In this work, we propose Model Quantization Benchmark (MQBench), a first attempt to evaluate, analyze, and benchmark the reproducibility and deployability for model quantization algorithms. We choose multiple different platforms for real-world deployments, including CPU, GPU, ASIC, DSP, and evaluate extensive state-of-the-art quantization algorithms under a unified training pipeline. MQBench acts like a bridge to connect the algorithm and the hardware. We conduct a comprehensive analysis and find considerable intuitive or counter-intuitive insights. By aligning the training settings, we find existing algorithms have about the same performance on the conventional academic track. While for the hardware-deployable quantization, there is a huge accuracy gap which remains unsettled. Surprisingly, no existing algorithm wins every challenge in MQBench, and we hope this work could inspire future research directions.