Neural networks have shown great potential in compressing volume data for visualization. However, due to the high cost of training and inference, such volumetric neural representations have thus far only been applied to offline data processing and non-interactive rendering. In this paper, we demonstrate that by simultaneously leveraging modern GPU tensor cores, a native CUDA neural network framework, and a well-designed rendering algorithm with macro-cell acceleration, we can interactively ray trace volumetric neural representations (10-60fps). Our neural representations are also high-fidelity (PSNR > 30dB) and compact (10-1000x smaller). Additionally, we show that it is possible to fit the entire training step inside a rendering loop and skip the pre-training process completely. To support extreme-scale volume data, we also develop an efficient out-of-core training strategy, which allows our volumetric neural representation training to potentially scale up to terascale using only an NVIDIA RTX 3090 workstation.
Abdominal organ segmentation has many important clinical applications, such as organ quantification, surgical planning, and disease diagnosis. However, manually annotating organs from CT scans is time-consuming and labor-intensive. Semi-supervised learning has shown the potential to alleviate this challenge by learning from a large set of unlabeled images and limited labeled samples. In this work, we follow the self-training strategy and employ a high-performance hybrid architecture (PHTrans) consisting of CNN and Swin Transformer for the teacher model to generate precise pseudo labels for unlabeled data. Afterward, we introduce them with labeled data together into a two-stage segmentation framework with lightweight PHTrans for training to improve the performance and generalization ability of the model while remaining efficient. Experiments on the validation set of FLARE2022 demonstrate that our method achieves excellent segmentation performance as well as fast and low-resource model inference. The average DSC and NSD are 0.8956 and 0.9316, respectively. Under our development environments, the average inference time is 18.62 s, the average maximum GPU memory is 1995.04 MB, and the area under the GPU memory-time curve and the average area under the CPU utilization-time curve are 23196.84 and 319.67. The code is available at https://github.com/lseventeen/FLARE22-TwoStagePHTrans.
This paper presents $\mu\text{KG}$, an open-source Python library for representation learning over knowledge graphs. $\mu\text{KG}$ supports joint representation learning over multi-source knowledge graphs (and also a single knowledge graph), multiple deep learning libraries (PyTorch and TensorFlow2), multiple embedding tasks (link prediction, entity alignment, entity typing, and multi-source link prediction), and multiple parallel computing modes (multi-process and multi-GPU computing). It currently implements 26 popular knowledge graph embedding models and supports 16 benchmark datasets. $\mu\text{KG}$ provides advanced implementations of embedding techniques with simplified pipelines of different tasks. It also comes with high-quality documentation for ease of use. $\mu\text{KG}$ is more comprehensive than existing knowledge graph embedding libraries. It is useful for a thorough comparison and analysis of various embedding models and tasks. We show that the jointly learned embeddings can greatly help knowledge-powered downstream tasks, such as multi-hop knowledge graph question answering. We will stay abreast of the latest developments in the related fields and incorporate them into $\mu\text{KG}$.
Geometric image transformations that arise in the real world, such as scaling and rotation, have been shown to easily deceive deep neural networks (DNNs). Hence, training DNNs to be certifiably robust to these perturbations is critical. However, no prior work has been able to incorporate the objective of deterministic certified robustness against geometric transformations into the training procedure, as existing verifiers are exceedingly slow. To address these challenges, we propose the first provable defense for deterministic certified geometric robustness. Our framework leverages a novel GPU-optimized verifier that can certify images between 60$\times$ to 42,600$\times$ faster than existing geometric robustness verifiers, and thus unlike existing works, is fast enough for use in training. Across multiple datasets, our results show that networks trained via our framework consistently achieve state-of-the-art deterministic certified geometric robustness and clean accuracy. Furthermore, for the first time, we verify the geometric robustness of a neural network for the challenging, real-world setting of autonomous driving.
We introduce the task of spotting temporally precise, fine-grained events in video (detecting the precise moment in time events occur). Precise spotting requires models to reason globally about the full-time scale of actions and locally to identify subtle frame-to-frame appearance and motion differences that identify events during these actions. Surprisingly, we find that top performing solutions to prior video understanding tasks such as action detection and segmentation do not simultaneously meet both requirements. In response, we propose E2E-Spot, a compact, end-to-end model that performs well on the precise spotting task and can be trained quickly on a single GPU. We demonstrate that E2E-Spot significantly outperforms recent baselines adapted from the video action detection, segmentation, and spotting literature to the precise spotting task. Finally, we contribute new annotations and splits to several fine-grained sports action datasets to make these datasets suitable for future work on precise spotting.
Recent advanced studies have spent considerable human efforts on optimizing network architectures for stereo matching but hardly achieved both high accuracy and fast inference speed. To ease the workload in network design, neural architecture search (NAS) has been applied with great success to various sparse prediction tasks, such as image classification and object detection. However, existing NAS studies on the dense prediction task, especially stereo matching, still cannot be efficiently and effectively deployed on devices of different computing capabilities. To this end, we propose to train an elastic and accurate network for stereo matching (EASNet) that supports various 3D architectural settings on devices with different computing capabilities. Given the deployment latency constraint on the target device, we can quickly extract a sub-network from the full EASNet without additional training while the accuracy of the sub-network can still be maintained. Extensive experiments show that our EASNet outperforms both state-of-the-art human-designed and NAS-based architectures on Scene Flow and MPI Sintel datasets in terms of model accuracy and inference speed. Particularly, deployed on an inference GPU, EASNet achieves a new SOTA 0.73 EPE on the Scene Flow dataset with 100 ms, which is 4.5$\times$ faster than LEAStereo with a better quality model.
Profile hidden Markov models (pHMMs) are widely employed in various bioinformatics applications to identify similarities between biological sequences, such as DNA or protein sequences. In pHMMs, sequences are represented as graph structures. These probabilities are subsequently used to compute the similarity score between a sequence and a pHMM graph. The Baum-Welch algorithm, a prevalent and highly accurate method, utilizes these probabilities to optimize and compute similarity scores. However, the Baum-Welch algorithm is computationally intensive, and existing solutions offer either software-only or hardware-only approaches with fixed pHMM designs. We identify an urgent need for a flexible, high-performance, and energy-efficient HW/SW co-design to address the major inefficiencies in the Baum-Welch algorithm for pHMMs. We introduce ApHMM, the first flexible acceleration framework designed to significantly reduce both computational and energy overheads associated with the Baum-Welch algorithm for pHMMs. ApHMM tackles the major inefficiencies in the Baum-Welch algorithm by 1) designing flexible hardware to accommodate various pHMM designs, 2) exploiting predictable data dependency patterns through on-chip memory with memoization techniques, 3) rapidly filtering out negligible computations using a hardware-based filter, and 4) minimizing redundant computations. ApHMM achieves substantial speedups of 15.55x - 260.03x, 1.83x - 5.34x, and 27.97x when compared to CPU, GPU, and FPGA implementations of the Baum-Welch algorithm, respectively. ApHMM outperforms state-of-the-art CPU implementations in three key bioinformatics applications: 1) error correction, 2) protein family search, and 3) multiple sequence alignment, by 1.29x - 59.94x, 1.03x - 1.75x, and 1.03x - 1.95x, respectively, while improving their energy efficiency by 64.24x - 115.46x, 1.75x, 1.96x.
Video text spotting(VTS) is the task that requires simultaneously detecting, tracking and recognizing text in the video. Existing video text spotting methods typically develop sophisticated pipelines and multiple models, which is not friend for real-time applications. Here we propose a real-time end-to-end video text spotter with Contrastive Representation learning (CoText). Our contributions are three-fold: 1) CoText simultaneously address the three tasks (e.g., text detection, tracking, recognition) in a real-time end-to-end trainable framework. 2) With contrastive learning, CoText models long-range dependencies and learning temporal information across multiple frames. 3) A simple, lightweight architecture is designed for effective and accurate performance, including GPU-parallel detection post-processing, CTC-based recognition head with Masked RoI. Extensive experiments show the superiority of our method. Especially, CoText achieves an video text spotting IDF1 of 72.0% at 41.0 FPS on ICDAR2015video, with 10.5% and 32.0 FPS improvement the previous best method. The code can be found at github.com/weijiawu/CoText.
Histopathology whole slide images (WSIs) play a very important role in clinical studies and serve as the gold standard for many cancer diagnoses. However, generating automatic tools for processing WSIs is challenging due to their enormous sizes. Currently, to deal with this issue, conventional methods rely on a multiple instance learning (MIL) strategy to process a WSI at patch level. Although effective, such methods are computationally expensive, because tiling a WSI into patches takes time and does not explore the spatial relations between these tiles. To tackle these limitations, we propose a locally supervised learning framework which processes the entire slide by exploring the entire local and global information that it contains. This framework divides a pre-trained network into several modules and optimizes each module locally using an auxiliary model. We also introduce a random feature reconstruction unit (RFR) to preserve distinguishing features during training and improve the performance of our method by 1% to 3%. Extensive experiments on three publicly available WSI datasets: TCGA-NSCLC, TCGA-RCC and LKS, highlight the superiority of our method on different classification tasks. Our method outperforms the state-of-the-art MIL methods by 2% to 5% in accuracy, while being 7 to 10 times faster. Additionally, when dividing it into eight modules, our method requires as little as 20% of the total gpu memory required by end-to-end training. Our code is available at https://github.com/cvlab-stonybrook/local_learning_wsi.
Training machine learning (ML) algorithms is a computationally intensive process, which is frequently memory-bound due to repeatedly accessing large training datasets. As a result, processor-centric systems (e.g., CPU, GPU) suffer from costly data movement between memory units and processing units, which consumes large amounts of energy and execution cycles. Memory-centric computing systems, i.e., with processing-in-memory (PIM) capabilities, can alleviate this data movement bottleneck. Our goal is to understand the potential of modern general-purpose PIM architectures to accelerate ML training. To do so, we (1) implement several representative classic ML algorithms (namely, linear regression, logistic regression, decision tree, K-Means clustering) on a real-world general-purpose PIM architecture, (2) rigorously evaluate and characterize them in terms of accuracy, performance and scaling, and (3) compare to their counterpart implementations on CPU and GPU. Our evaluation on a real memory-centric computing system with more than 2500 PIM cores shows that general-purpose PIM architectures can greatly accelerate memory-bound ML workloads, when the necessary operations and datatypes are natively supported by PIM hardware. For example, our PIM implementation of decision tree is $27\times$ faster than a state-of-the-art CPU version on an 8-core Intel Xeon, and $1.34\times$ faster than a state-of-the-art GPU version on an NVIDIA A100. Our K-Means clustering on PIM is $2.8\times$ and $3.2\times$ than state-of-the-art CPU and GPU versions, respectively. To our knowledge, our work is the first one to evaluate ML training on a real-world PIM architecture. We conclude with key observations, takeaways, and recommendations that can inspire users of ML workloads, programmers of PIM architectures, and hardware designers & architects of future memory-centric computing systems.
Low-precision arithmetic has had a transformative effect on the training of neural networks, reducing computation, memory and energy requirements. However, despite its promise, low-precision arithmetic has received little attention for Gaussian processes (GPs), largely because GPs require sophisticated linear algebra routines that are unstable in low-precision. We study the different failure modes that can occur when training GPs in half precision. To circumvent these failure modes, we propose a multi-faceted approach involving conjugate gradients with re-orthogonalization, mixed precision, and preconditioning. Our approach significantly improves the numerical stability and practical performance of conjugate gradients in low-precision over a wide range of settings, enabling GPs to train on $1.8$ million data points in $10$ hours on a single GPU, without any sparse approximations.
Denoising diffusion probabilistic models (DDPMs) have recently achieved leading performances in many generative tasks. However, the inherited iterative sampling process costs hinder their applications to text-to-speech deployment. Through the preliminary study on diffusion model parameterization, we find that previous gradient-based TTS models require hundreds or thousands of iterations to guarantee high sample quality, which poses a challenge for accelerating sampling. In this work, we propose ProDiff, on progressive fast diffusion model for high-quality text-to-speech. Unlike previous work estimating the gradient for data density, ProDiff parameterizes the denoising model by directly predicting clean data to avoid distinct quality degradation in accelerating sampling. To tackle the model convergence challenge with decreased diffusion iterations, ProDiff reduces the data variance in the target site via knowledge distillation. Specifically, the denoising model uses the generated mel-spectrogram from an N-step DDIM teacher as the training target and distills the behavior into a new model with N/2 steps. As such, it allows the TTS model to make sharp predictions and further reduces the sampling time by orders of magnitude. Our evaluation demonstrates that ProDiff needs only 2 iterations to synthesize high-fidelity mel-spectrograms, while it maintains sample quality and diversity competitive with state-of-the-art models using hundreds of steps. ProDiff enables a sampling speed of 24x faster than real-time on a single NVIDIA 2080Ti GPU, making diffusion models practically applicable to text-to-speech synthesis deployment for the first time. Our extensive ablation studies demonstrate that each design in ProDiff is effective, and we further show that ProDiff can be easily extended to the multi-speaker setting. Audio samples are available at \url{https://ProDiff.github.io/.}
As demonstrated by GPT-3 and T5, transformers grow in capability as parameter spaces become larger and larger. However, for tasks that require a large amount of knowledge, non-parametric memory allows models to grow dramatically with a sub-linear increase in computational cost and GPU memory requirements. Recent models such as RAG and REALM have introduced retrieval into conditional generation. These models incorporate neural initial retrieval from a corpus of passages. We build on this line of research, proposing Re2G, which combines both neural initial retrieval and reranking into a BART-based sequence-to-sequence generation. Our reranking approach also permits merging retrieval results from sources with incomparable scores, enabling an ensemble of BM25 and neural initial retrieval. To train our system end-to-end, we introduce a novel variation of knowledge distillation to train the initial retrieval, reranker, and generation using only ground truth on the target sequence output. We find large gains in four diverse tasks: zero-shot slot filling, question answering, fact-checking, and dialog, with relative gains of 9% to 34% over the previous state-of-the-art on the KILT leaderboard. We make our code available as open source at https://github.com/IBM/kgi-slot-filling/tree/re2g.
We present a new probabilistic method for detecting, deblending, and cataloging astronomical sources called the Bayesian Light Source Separator (BLISS). BLISS is based on deep generative models, which embed neural networks within a Bayesian model. For posterior inference, BLISS uses a new form of variational inference known as Forward Amortized Variational Inference. The BLISS inference routine is fast, requiring a single forward pass of the encoder networks on a GPU once the encoder networks are trained. BLISS can perform fully Bayesian inference on megapixel images in seconds, and produces highly accurate catalogs. BLISS is highly extensible, and has the potential to directly answer downstream scientific questions in addition to producing probabilistic catalogs.
Vision transformers (ViTs) are usually considered to be less light-weight than convolutional neural networks (CNNs) due to the lack of inductive bias. Recent works thus resort to convolutions as a plug-and-play module and embed them in various ViT counterparts. In this paper, we argue that the convolutional kernels perform information aggregation to connect all tokens; however, they would be actually unnecessary for light-weight ViTs if this explicit aggregation could function in a more homogeneous way. Inspired by this, we present LightViT as a new family of light-weight ViTs to achieve better accuracy-efficiency balance upon the pure transformer blocks without convolution. Concretely, we introduce a global yet efficient aggregation scheme into both self-attention and feed-forward network (FFN) of ViTs, where additional learnable tokens are introduced to capture global dependencies; and bi-dimensional channel and spatial attentions are imposed over token embeddings. Experiments show that our model achieves significant improvements on image classification, object detection, and semantic segmentation tasks. For example, our LightViT-T achieves 78.7% accuracy on ImageNet with only 0.7G FLOPs, outperforming PVTv2-B0 by 8.2% while 11% faster on GPU. Code is available at https://github.com/hunto/LightViT.
Video-to-Video synthesis (Vid2Vid) has achieved remarkable results in generating a photo-realistic video from a sequence of semantic maps. However, this pipeline suffers from high computational cost and long inference latency, which largely depends on two essential factors: 1) network architecture parameters, 2) sequential data stream. Recently, the parameters of image-based generative models have been significantly compressed via more efficient network architectures. Nevertheless, existing methods mainly focus on slimming network architectures and ignore the size of the sequential data stream. Moreover, due to the lack of temporal coherence, image-based compression is not sufficient for the compression of the video task. In this paper, we present a spatial-temporal compression framework, \textbf{Fast-Vid2Vid}, which focuses on data aspects of generative models. It makes the first attempt at time dimension to reduce computational resources and accelerate inference. Specifically, we compress the input data stream spatially and reduce the temporal redundancy. After the proposed spatial-temporal knowledge distillation, our model can synthesize key-frames using the low-resolution data stream. Finally, Fast-Vid2Vid interpolates intermediate frames by motion compensation with slight latency. On standard benchmarks, Fast-Vid2Vid achieves around real-time performance as 20 FPS and saves around 8x computational cost on a single V100 GPU.
Gradient-based multilevel optimization (MLO) has gained attention as a framework for studying numerous problems, ranging from hyperparameter optimization and meta-learning to neural architecture search and reinforcement learning. However, gradients in MLO, which are obtained by composing best-response Jacobians via the chain rule, are notoriously difficult to implement and memory/compute intensive. We take an initial step towards closing this gap by introducing Betty, a software library for large-scale MLO. At its core, we devise a novel dataflow graph for MLO, which allows us to (1) develop efficient automatic differentiation for MLO that reduces the computational complexity from O(d^3) to O(d^2), (2) incorporate systems support such as mixed-precision and data-parallel training for scalability, and (3) facilitate implementation of MLO programs of arbitrary complexity while allowing a modular interface for diverse algorithmic and systems design choices. We empirically demonstrate that Betty can be used to implement an array of MLO programs, while also observing up to 11% increase in test accuracy, 14% decrease in GPU memory usage, and 20% decrease in training wall time over existing implementations on multiple benchmarks. We also showcase that Betty enables scaling MLO to models with hundreds of millions of parameters. We open-source the code at https://github.com/leopard-ai/betty.
YOLOv7 surpasses all known object detectors in both speed and accuracy in the range from 5 FPS to 160 FPS and has the highest accuracy 56.8% AP among all known real-time object detectors with 30 FPS or higher on GPU V100. YOLOv7-E6 object detector (56 FPS V100, 55.9% AP) outperforms both transformer-based detector SWIN-L Cascade-Mask R-CNN (9.2 FPS A100, 53.9% AP) by 509% in speed and 2% in accuracy, and convolutional-based detector ConvNeXt-XL Cascade-Mask R-CNN (8.6 FPS A100, 55.2% AP) by 551% in speed and 0.7% AP in accuracy, as well as YOLOv7 outperforms: YOLOR, YOLOX, Scaled-YOLOv4, YOLOv5, DETR, Deformable DETR, DINO-5scale-R50, ViT-Adapter-B and many other object detectors in speed and accuracy. Moreover, we train YOLOv7 only on MS COCO dataset from scratch without using any other datasets or pre-trained weights. Source code is released in https://github.com/WongKinYiu/yolov7.
In this work, we propose PolarBEV for vision-based uneven BEV representation learning. To adapt to the foreshortening effect of camera imaging, we rasterize the BEV space both angularly and radially, and introduce polar embedding decomposition to model the associations among polar grids. Polar grids are rearranged to an array-like regular representation for efficient processing. Besides, to determine the 2D-to-3D correspondence, we iteratively update the BEV surface based on a hypothetical plane, and adopt height-based feature transformation. PolarBEV keeps real-time inference speed on a single 2080Ti GPU, and outperforms other methods for both BEV semantic segmentation and BEV instance segmentation. Thorough ablations are presented to validate the design. The code will be released at \url{https://github.com/SuperZ-Liu/PolarBEV}.
The past several years have witnessed the success of transformer-based models, and their scale and application scenarios continue to grow aggressively. The current landscape of transformer models is increasingly diverse: the model size varies drastically with the largest being of hundred-billion parameters; the model characteristics differ due to the sparsity introduced by the Mixture-of-Experts; the target application scenarios can be latency-critical or throughput-oriented; the deployment hardware could be single- or multi-GPU systems with different types of memory and storage, etc. With such increasing diversity and the fast-evolving pace of transformer models, designing a highly performant and efficient inference system is extremely challenging. In this paper, we present DeepSpeed Inference, a comprehensive system solution for transformer model inference to address the above-mentioned challenges. DeepSpeed Inference consists of (1) a multi-GPU inference solution to minimize latency while maximizing the throughput of both dense and sparse transformer models when they fit in aggregate GPU memory, and (2) a heterogeneous inference solution that leverages CPU and NVMe memory in addition to the GPU memory and compute to enable high inference throughput with large models which do not fit in aggregate GPU memory. DeepSpeed Inference reduces latency by up to 7.3X over the state-of-the-art for latency-oriented scenarios and increases throughput by over 1.5x for throughput-oriented scenarios. Moreover, it enables trillion parameter scale inference under real-time latency constraints by leveraging hundreds of GPUs, an unprecedented scale for inference. It can inference 25x larger models than with GPU-only solutions, while delivering a high throughput of 84 TFLOPS (over $50\%$ of A6000 peak).