We develop BenchPress, the first ML benchmark generator for compilers that is steerable within feature space representations of source code. BenchPress synthesizes compiling functions by adding new code in any part of an empty or existing sequence by jointly observing its left and right context, achieving excellent compilation rate. BenchPress steers benchmark generation towards desired target features that has been impossible for state of the art synthesizers (or indeed humans) to reach. It performs better in targeting the features of Rodinia benchmarks in 3 different feature spaces compared with (a) CLgen - a state of the art ML synthesizer, (b) CLSmith fuzzer, (c) SRCIROR mutator or even (d) human-written code from GitHub. BenchPress is the first generator to search the feature space with active learning in order to generate benchmarks that will improve a downstream task. We show how using BenchPress, Grewe's et al. CPU vs GPU heuristic model can obtain a higher speedup when trained on BenchPress's benchmarks compared to other techniques. BenchPress is a powerful code generator: Its generated samples compile at a rate of 86%, compared to CLgen's 2.33%. Starting from an empty fixed input, BenchPress produces 10x more unique, compiling OpenCL benchmarks than CLgen, which are significantly larger and more feature diverse.
Training deep neural networks (DNNs) is becoming increasingly more resource- and energy-intensive every year. Unfortunately, existing works primarily focus on optimizing DNN training for faster completion, often without considering the impact on energy efficiency. In this paper, we observe that common practices to improve training performance can often lead to inefficient energy usage. More importantly, we demonstrate that there is a tradeoff between energy consumption and performance optimization. To this end, we propose Zeus, an optimization framework to navigate this tradeoff by automatically finding optimal job- and GPU-level configurations for recurring DNN training jobs. Zeus uses an online exploration-exploitation approach in conjunction with just-in-time energy profiling, averting the need for expensive offline measurements, while adapting to data drifts over time. Our evaluation shows that Zeus can improve the energy efficiency of DNN training by 15.3%-75.8% for diverse workloads.
This paper summarizes model improvements and inference-time optimizations for the popular anchor-based detectors in the scenes of autonomous driving. Based on the high-performing RCNN-RS and RetinaNet-RS detection frameworks designed for common detection scenes, we study a set of framework improvements to adapt the detectors to better detect small objects in crowd scenes. Then, we propose a model scaling strategy by scaling input resolution and model size to achieve a better speed-accuracy trade-off curve. We evaluate our family of models on the real-time 2D detection track of the Waymo Open Dataset (WOD). Within the 70 ms/frame latency constraint on a V100 GPU, our largest Cascade RCNN-RS model achieves 76.9% AP/L1 and 70.1% AP/L2, attaining the new state-of-the-art on WOD real-time 2D detection. Our fastest RetinaNet-RS model achieves 6.3 ms/frame while maintaining a reasonable detection precision at 50.7% AP/L1 and 42.9% AP/L2.
Variational optimization of neural-network representations of quantum states has been successfully applied to solve interacting fermionic problems. Despite rapid developments, significant scalability challenges arise when considering molecules of large scale, which correspond to non-locally interacting quantum spin Hamiltonians consisting of sums of thousands or even millions of Pauli operators. In this work, we introduce scalable parallelization strategies to improve neural-network-based variational quantum Monte Carlo calculations for ab-initio quantum chemistry applications. We establish GPU-supported local energy parallelism to compute the optimization objective for Hamiltonians of potentially complex molecules. Using autoregressive sampling techniques, we demonstrate systematic improvement in wall-clock timings required to achieve CCSD baseline target energies. The performance is further enhanced by accommodating the structure of resultant spin Hamiltonians into the autoregressive sampling ordering. The algorithm achieves promising performance in comparison with the classical approximate methods and exhibits both running time and scalability advantages over existing neural-network based methods.
Bound propagation methods, when combined with branch and bound, are among the most effective methods to formally verify properties of deep neural networks such as correctness, robustness, and safety. However, existing works cannot handle the general form of cutting plane constraints widely accepted in traditional solvers, which are crucial for strengthening verifiers with tightened convex relaxations. In this paper, we generalize the bound propagation procedure to allow the addition of arbitrary cutting plane constraints, including those involving relaxed integer variables that do not appear in existing bound propagation formulations. Our generalized bound propagation method, GCP-CROWN, opens up the opportunity to apply general cutting plane methods for neural network verification while benefiting from the efficiency and GPU acceleration of bound propagation methods. As a case study, we investigate the use of cutting planes generated by off-the-shelf mixed integer programming (MIP) solver. We find that MIP solvers can generate high-quality cutting planes for strengthening bound-propagation-based verifiers using our new formulation. Since the branching-focused bound propagation procedure and the cutting-plane-focused MIP solver can run in parallel utilizing different types of hardware (GPUs and CPUs), their combination can quickly explore a large number of branches with strong cutting planes, leading to strong verification performance. Experiments demonstrate that our method is the first verifier that can completely solve the oval20 benchmark and verify twice as many instances on the oval21 benchmark compared to the best tool in VNN-COMP 2021, and also noticeably outperforms state-of-the-art verifiers on a wide range of benchmarks. GCP-CROWN is part of the $\alpha,\!\beta$-CROWN verifier, the VNN-COMP 2022 winner. Code is available at http://PaperCode.cc/GCP-CROWN
Graph neural networks (GNNs) are deep learning models designed specifically for graph data, and they typically rely on node features as the input to the first layer. When applying such a type of network on the graph without node features, one can extract simple graph-based node features (e.g., number of degrees) or learn the input node representations (i.e., embeddings) when training the network. While the latter approach, which trains node embeddings, more likely leads to better performance, the number of parameters associated with the embeddings grows linearly with the number of nodes. It is therefore impractical to train the input node embeddings together with GNNs within graphics processing unit (GPU) memory in an end-to-end fashion when dealing with industrial-scale graph data. Inspired by the embedding compression methods developed for natural language processing (NLP) tasks, we develop a node embedding compression method where each node is compactly represented with a bit vector instead of a floating-point vector. The parameters utilized in the compression method can be trained together with GNNs. We show that the proposed node embedding compression method achieves superior performance compared to the alternatives.
Deep learning recommendation models (DLRMs) have been widely applied in Internet companies. The embedding tables of DLRMs are too large to fit on GPU memory entirely. We propose a GPU-based software cache approaches to dynamically manage the embedding table in the CPU and GPU memory space by leveraging the id's frequency statistics of the target dataset. Our proposed software cache is efficient in training entire DLRMs on GPU in a synchronized update manner. It is also scaled to multiple GPUs in combination with the widely used hybrid parallel training approaches. Evaluating our prototype system shows that we can keep only 1.5% of the embedding parameters in the GPU to obtain a decent end-to-end training speed.
Real-time viral genome detection, taxonomic classification and phylogenetic analysis are critical for efficient tracking and control of viral pandemics such as Covid-19. However, the unprecedented and still growing amounts of viral genome data create a computational bottleneck, which effectively prevents the real-time pandemic tracking. For genomic tracing to work effectively, each new viral genome sequence must be placed in its pangenomic context. Re-inferring the full phylogeny of SARS-CoV-2, with datasets containing millions of samples, is prohibitively slow even using powerful computational resources. We are attempting to alleviate the computational bottleneck by modifying and applying Vision Transformer, a recently developed neural network model for image recognition, to taxonomic classification and placement of viral genomes, such as SARS-CoV-2. Our solution, CoViT, places SARS-CoV-2 genome accessions onto SARS-CoV-2 phylogenetic tree with the accuracy of 94.2%. Since CoViT is a classification neural network, it provides more than one likely placement. Specifically, one of the two most likely placements suggested by CoViT is correct with the probability of 97.9%. The probability of the correct placement to be found among the five most likely placements generated by CoViT is 99.8%. The placement time is 0.055s per individual genome running on NVIDIAs GeForce RTX 2080 Ti GPU. We make CoViT available to research community through GitHub: https://github.com/zuherJahshan/covit.
We present CG-NeRF, a cascade and generalizable neural radiance fields method for view synthesis. Recent generalizing view synthesis methods can render high-quality novel views using a set of nearby input views. However, the rendering speed is still slow due to the nature of uniformly-point sampling of neural radiance fields. Existing scene-specific methods can train and render novel views efficiently but can not generalize to unseen data. Our approach addresses the problems of fast and generalizing view synthesis by proposing two novel modules: a coarse radiance fields predictor and a convolutional-based neural renderer. This architecture infers consistent scene geometry based on the implicit neural fields and renders new views efficiently using a single GPU. We first train CG-NeRF on multiple 3D scenes of the DTU dataset, and the network can produce high-quality and accurate novel views on unseen real and synthetic data using only photometric losses. Moreover, our method can leverage a denser set of reference images of a single scene to produce accurate novel views without relying on additional explicit representations and still maintains the high-speed rendering of the pre-trained model. Experimental results show that CG-NeRF outperforms state-of-the-art generalizable neural rendering methods on various synthetic and real datasets.
The training of modern speech processing systems often requires a large amount of simulated room impulse response (RIR) data in order to allow the systems to generalize well in real-world, reverberant environments. However, simulating realistic RIR data typically requires accurate physical modeling, and the acceleration of such simulation process typically requires certain computational platforms such as a graphics processing unit (GPU). In this paper, we propose FRA-RIR, a fast random approximation method of the widely-used image-source method (ISM), to efficiently generate realistic RIR data without specific computational devices. FRA-RIR replaces the physical simulation in the standard ISM by a series of random approximations, which significantly speeds up the simulation process and enables its application in on-the-fly data generation pipelines. Experiments show that FRA-RIR can not only be significantly faster than other existing ISM-based RIR simulation tools on standard computational platforms, but also improves the performance of speech denoising systems evaluated on real-world RIR when trained with simulated RIR. A Python implementation of FRA-RIR is available online\footnote{\url{https://github.com/yluo42/FRA-RIR}}.
Transformers are considered one of the most important deep learning models since 2018, in part because it establishes state-of-the-art (SOTA) records and could potentially replace existing Deep Neural Networks (DNNs). Despite the remarkable triumphs, the prolonged turnaround time of Transformer models is a widely recognized roadblock. The variety of sequence lengths imposes additional computing overhead where inputs need to be zero-padded to the maximum sentence length in the batch to accommodate the parallel computing platforms. This paper targets the field-programmable gate array (FPGA) and proposes a coherent sequence length adaptive algorithm-hardware co-design for Transformer acceleration. Particularly, we develop a hardware-friendly sparse attention operator and a length-aware hardware resource scheduling algorithm. The proposed sparse attention operator brings the complexity of attention-based models down to linear complexity and alleviates the off-chip memory traffic. The proposed length-aware resource hardware scheduling algorithm dynamically allocates the hardware resources to fill up the pipeline slots and eliminates bubbles for NLP tasks. Experiments show that our design has very small accuracy loss and has 80.2 $\times$ and 2.6 $\times$ speedup compared to CPU and GPU implementation, and 4 $\times$ higher energy efficiency than state-of-the-art GPU accelerator optimized via CUBLAS GEMM.
Robots working in unstructured environments must be capable of sensing and interpreting their surroundings. One of the main obstacles of deep-learning-based models in the field of robotics is the lack of domain-specific labeled data for different industrial applications. In this article, we propose a sim2real transfer learning method based on domain randomization for object detection with which labeled synthetic datasets of arbitrary size and object types can be automatically generated. Subsequently, a state-of-the-art convolutional neural network, YOLOv4, is trained to detect the different types of industrial objects. With the proposed domain randomization method, we could shrink the reality gap to a satisfactory level, achieving 86.32% and 97.38% mAP50 scores, respectively, in the case of zero-shot and one-shot transfers, on our manually annotated dataset containing 190 real images. Our solution fits for industrial use as the data generation process takes less than 0.5 s per image and the training lasts only around 12 h, on a GeForce RTX 2080 Ti GPU. Furthermore, it can reliably differentiate similar classes of objects by having access to only one real image for training. To our best knowledge, this is the only work thus far satisfying these constraints.
This paper presents a state-of-the-art overview on how to architect, design, and optimize Deep Neural Networks (DNNs) such that performance is improved and accuracy is preserved. The paper covers a set of optimizations that span the entire Machine Learning processing pipeline. We introduce two types of optimizations. The first alters the DNN model and requires NN re-training, while the second does not. We focus on GPU optimizations, but we believe the presented techniques can be used with other AI inference platforms. To demonstrate the DNN model optimizations, we improve one of the most advanced deep network architectures for optical flow, RAFT arXiv:2003.12039, on a popular edge AI inference platform (Nvidia Jetson AGX Xavier).
Boosting the runtime performance of deep neural networks (DNNs) is critical due to their wide adoption in real-world tasks. Existing approaches to optimizing the tensor algebra expression of a DNN only consider expressions representable by a fixed set of predefined operators, missing possible optimization opportunities between general expressions. We propose OLLIE, the first derivation-based tensor program optimizer. OLLIE optimizes tensor programs by leveraging transformations between general tensor algebra expressions, enabling a significantly larger expression search space that includes those supported by prior work as special cases. OLLIE uses a hybrid derivation-based optimizer that effectively combines explorative and guided derivations to quickly discover highly optimized expressions. Evaluation on seven DNNs shows that OLLIE can outperform existing optimizers by up to 2.73$\times$ (1.46$\times$ on average) on an A100 GPU and up to 2.68$\times$ (1.51$\times$) on a V100 GPU, respectively.
Training deep neural networks consumes increasing computational resource shares in many compute centers. Often, a brute force approach to obtain hyperparameter values is employed. Our goal is (1) to enhance this by enabling second-order optimization methods with fewer hyperparameters for large-scale neural networks and (2) to perform a survey of the performance optimizers for specific tasks to suggest users the best one for their problem. We introduce a novel second-order optimization method that requires the effect of the Hessian on a vector only and avoids the huge cost of explicitly setting up the Hessian for large-scale networks. We compare the proposed second-order method with two state-of-the-art optimizers on five representative neural network problems, including regression and very deep networks from computer vision or variational autoencoders. For the largest setup, we efficiently parallelized the optimizers with Horovod and applied it to a 8 GPU NVIDIA P100 (DGX-1) machine.
Pure vision transformer architectures are highly effective for short video classification and action recognition tasks. However, due to the quadratic complexity of self attention and lack of inductive bias, transformers are resource intensive and suffer from data inefficiencies. Long form video understanding tasks amplify data and memory efficiency problems in transformers making current approaches unfeasible to implement on data or memory restricted domains. This paper introduces an efficient Spatio-Temporal Attention Network (STAN) which uses a two-stream transformer architecture to model dependencies between static image features and temporal contextual features. Our proposed approach can classify videos up to two minutes in length on a single GPU, is data efficient, and achieves SOTA performance on several long video understanding tasks.
Learned image compression allows achieving state-of-the-art accuracy and compression ratios, but their relatively slow runtime performance limits their usage. While previous attempts on optimizing learned image codecs focused more on the neural model and entropy coding, we present an alternative method to improving the runtime performance of various learned image compression models. We introduce multi-threaded pipelining and an optimized memory model to enable GPU and CPU workloads asynchronous execution, fully taking advantage of computational resources. Our architecture alone already produces excellent performance without any change to the neural model itself. We also demonstrate that combining our architecture with previous tweaks to the neural models can further improve runtime performance. We show that our implementations excel in throughput and latency compared to the baseline and demonstrate the performance of our implementations by creating a real-time video streaming encoder-decoder sample application, with the encoder running on an embedded device.
Solid results from Transformers have made them prevailing architectures in various natural language and vision tasks. As a default component in Transformers, Layer Normalization (LN) normalizes activations within each token to boost the robustness. However, LN requires on-the-fly statistics calculation in inference as well as division and square root operations, leading to inefficiency on hardware. What is more, replacing LN with other hardware-efficient normalization schemes (e.g., Batch Normalization) results in inferior performance, even collapse in training. We find that this dilemma is caused by abnormal behaviors of activation statistics, including large fluctuations over iterations and extreme outliers across layers. To tackle these issues, we propose Unified Normalization (UN), which can speed up the inference by being fused with other linear operations and achieve comparable performance on par with LN. UN strives to boost performance by calibrating the activation and gradient statistics with a tailored fluctuation smoothing strategy. Meanwhile, an adaptive outlier filtration strategy is applied to avoid collapse in training whose effectiveness is theoretically proved and experimentally verified in this paper. We demonstrate that UN can be an efficient drop-in alternative to LN by conducting extensive experiments on language and vision tasks. Besides, we evaluate the efficiency of our method on GPU. Transformers equipped with UN enjoy about 31% inference speedup and nearly 18% memory reduction. Code will be released at https://github.com/hikvision-research/Unified-Normalization.
The FastGeodis package provides an efficient implementation for computing Geodesic and Euclidean distance transforms (or a mixture of both), targeting efficient utilisation of CPU and GPU hardware. In particular, it implements the paralellisable raster scan method from Criminisi et al. (2009), where elements in a row (2D) or plane (3D) can be computed with parallel threads. This package is able to handle 2D as well as 3D data, where it achieves up to a 20x speedup on a CPU and up to a 74x speedup on a GPU as compared to an existing open-source library (Wang, 2020) that uses a non-parallelisable single-thread CPU implementation. The performance speedups reported here were evaluated using 3D volume data on an Nvidia GeForce Titan X (12 GB) with a 6-Core Intel Xeon E5-1650 CPU. Further in-depth comparison of performance improvements are discussed in the FastGeodis documentation: https://fastgeodis.readthedocs.io
With the popularity of deep learning, the hardware implementation platform of deep learning has received increasing interest. Unlike the general purpose devices, e.g., CPU, or GPU, where the deep learning algorithms are executed at the software level, neural network hardware accelerators directly execute the algorithms to achieve higher both energy efficiency and performance improvements. However, as the deep learning algorithms evolve frequently, the engineering effort and cost of designing the hardware accelerators are greatly increased. To improve the design quality while saving the cost, design automation for neural network accelerators was proposed, where design space exploration algorithms are used to automatically search the optimized accelerator design within a design space. Nevertheless, the increasing complexity of the neural network accelerators brings the increasing dimensions to the design space. As a result, the previous design space exploration algorithms are no longer effective enough to find an optimized design. In this work, we propose a neural network accelerator design automation framework named GANDSE, where we rethink the problem of design space exploration, and propose a novel approach based on the generative adversarial network (GAN) to support an optimized exploration for high dimension large design space. The experiments show that GANDSE is able to find the more optimized designs in negligible time compared with approaches including multilayer perceptron and deep reinforcement learning.