Compared to the multi-stage self-supervised multi-view stereo (MVS) method, the end-to-end (E2E) approach has received more attention due to its concise and efficient training pipeline. Recent E2E self-supervised MVS approaches have integrated third-party models (such as optical flow models, semantic segmentation models, NeRF models, etc.) to provide additional consistency constraints, which grows GPU memory consumption and complicates the model's structure and training pipeline. In this work, we propose an efficient framework for end-to-end self-supervised MVS, dubbed ES-MVSNet. To alleviate the high memory consumption of current E2E self-supervised MVS frameworks, we present a memory-efficient architecture that reduces memory usage by 43% without compromising model performance. Furthermore, with the novel design of asymmetric view selection policy and region-aware depth consistency, we achieve state-of-the-art performance among E2E self-supervised MVS methods, without relying on third-party models for additional consistency signals. Extensive experiments on DTU and Tanks&Temples benchmarks demonstrate that the proposed ES-MVSNet approach achieves state-of-the-art performance among E2E self-supervised MVS methods and competitive performance to many supervised and multi-stage self-supervised methods.
Transfomer-based models have significantly advanced natural language processing, in particular the performance in text classification tasks. Nevertheless, these models face challenges in processing large files, primarily due to their input constraints, which are generally restricted to hundreds or thousands of tokens. Attempts to address this issue in existing models usually consist in extracting only a fraction of the essential information from lengthy inputs, while often incurring high computational costs due to their complex architectures. In this work, we address the challenge of classifying large files from the perspective of correlated multiple instance learning. We introduce LaFiCMIL, a method specifically designed for large file classification. LaFiCMIL is optimized for efficient operation on a single GPU, making it a versatile solution for binary, multi-class, and multi-label classification tasks. We conducted extensive experiments using seven diverse and comprehensive benchmark datasets to assess LaFiCMIL's effectiveness. By integrating BERT for feature extraction, LaFiCMIL demonstrates exceptional performance, setting new benchmarks across all datasets. A notable achievement of our approach is its ability to scale BERT to handle nearly 20,000 tokens while operating on a single GPU with 32GB of memory. This efficiency, coupled with its state-of-the-art performance, highlights LaFiCMIL's potential as a groundbreaking approach in the field of large file classification.
In this work, we present a computing platform named digital twin brain (DTB) that can simulate spiking neuronal networks of the whole human brain scale and more importantly, a personalized biological brain structure. In comparison to most brain simulations with a homogeneous global structure, we highlight that the sparseness, couplingness and heterogeneity in the sMRI, DTI and PET data of the brain has an essential impact on the efficiency of brain simulation, which is proved from the scaling experiments that the DTB of human brain simulation is communication-intensive and memory-access intensive computing systems rather than computation-intensive. We utilize a number of optimization techniques to balance and integrate the computation loads and communication traffics from the heterogeneous biological structure to the general GPU-based HPC and achieve leading simulation performance for the whole human brain-scaled spiking neuronal networks. On the other hand, the biological structure, equipped with a mesoscopic data assimilation, enables the DTB to investigate brain cognitive function by a reverse-engineering method, which is demonstrated by a digital experiment of visual evaluation on the DTB. Furthermore, we believe that the developing DTB will be a promising powerful platform for a large of research orients including brain-inspiredintelligence, rain disease medicine and brain-machine interface.
With the popularity of cloud computing and machine learning, it has been a trend to outsource machine learning processes (including model training and model-based inference) to cloud. By the outsourcing, other than utilizing the extensive and scalable resource offered by the cloud service provider, it will also be attractive to users if the cloud servers can manage the machine learning processes autonomously on behalf of the users. Such a feature will be especially salient when the machine learning is expected to be a long-term continuous process and the users are not always available to participate. Due to security and privacy concerns, it is also desired that the autonomous learning preserves the confidentiality of users' data and models involved. Hence, in this paper, we aim to design a scheme that enables autonomous and confidential model refining in cloud. Homomorphic encryption and trusted execution environment technology can protect confidentiality for autonomous computation, but each of them has their limitations respectively and they are complementary to each other. Therefore, we further propose to integrate these two techniques in the design of the model refining scheme. Through implementation and experiments, we evaluate the feasibility of our proposed scheme. The results indicate that, with our proposed scheme the cloud server can autonomously refine an encrypted model with newly provided encrypted training data to continuously improve its accuracy. Though the efficiency is still significantly lower than the baseline scheme that refines plaintext-model with plaintext-data, we expect that it can be improved by fully utilizing the higher level of parallelism and the computational power of GPU at the cloud server.
The morphology and distribution of airway tree abnormalities enables diagnosis and disease characterisation across a variety of chronic respiratory conditions. In this regard, airway segmentation plays a critical role in the production of the outline of the entire airway tree to enable estimation of disease extent and severity. In this study, we propose a data-centric deep learning technique to segment the airway tree. The proposed technique utilises interpolation and image split to improve data usefulness and quality. Then, an ensemble learning strategy is implemented to aggregate the segmented airway trees at different scales. In terms of segmentation performance (dice similarity coefficient), our method outperforms the baseline model by 2.5% on average when a combined loss is used. Further, our proposed technique has a low GPU usage and high flexibility enabling it to be deployed on any 2D deep learning model.
Datacenters are increasingly becoming heterogeneous, and are starting to include specialized hardware for networking, video processing, and especially deep learning. To leverage the heterogeneous compute capability of modern datacenters, we develop an approach for compiler-level partitioning of deep neural networks (DNNs) onto multiple interconnected hardware devices. We present a general framework for heterogeneous DNN compilation, offering automatic partitioning and device mapping. Our scheduler integrates both an exact solver, through a mixed integer linear programming (MILP) formulation, and a modularity-based heuristic for scalability. Furthermore, we propose a theoretical lower bound formula for the optimal solution, which enables the assessment of the heuristic solutions' quality. We evaluate our scheduler in optimizing both conventional DNNs and randomly-wired neural networks, subject to latency and throughput constraints, on a heterogeneous system comprised of a CPU and two distinct GPUs. Compared to na\"ively running DNNs on the fastest GPU, he proposed framework can achieve more than 3$\times$ times lower latency and up to 2.9$\times$ higher throughput by automatically leveraging both data and model parallelism to deploy DNNs on our sample heterogeneous server node. Moreover, our modularity-based "splitting" heuristic improves the solution runtime up to 395$\times$ without noticeably sacrificing solution quality compared to an exact MILP solution, and outperforms all other heuristics by 30-60% solution quality. Finally, our case study shows how we can extend our framework to schedule large language models across multiple heterogeneous servers by exploiting symmetry in the hardware setup. Our code can be easily plugged in to existing frameworks, and is available at https://github.com/abdelfattah-lab/diviml.
The safety of children in children home has become an increasing social concern, and the purpose of this experiment is to use machine learning applied to detect the scenarios of child abuse to increase the safety of children. This experiment uses machine learning to classify and recognize a child's voice and predict whether the current sound made by the child is crying, screaming or laughing. If a child is found to be crying or screaming, an alert is immediately sent to the relevant personnel so that they can perceive what the child may be experiencing in a surveillance blind spot and respond in a timely manner. Together with a hybrid use of video image classification, the accuracy of child abuse detection can be significantly increased. This greatly reduces the likelihood that a child will receive violent abuse in the nursery and allows personnel to stop an imminent or incipient child abuse incident in time. The datasets collected from this experiment is entirely from sounds recorded on site at the children home, including crying, laughing, screaming sound and background noises. These sound files are transformed into spectrograms using Short-Time Fourier Transform, and then these image data are imported into a CNN neural network for classification, and the final trained model can achieve an accuracy of about 92% for sound detection.
Recently, learned image compression has achieved impressive performance. The entropy model, which estimates the distribution of the latent representation, plays a crucial role in enhancing rate-distortion performance. However, existing global context modules rely on computationally intensive quadratic complexity computations to capture global correlations. This quadratic complexity imposes limitations on the potential of high-resolution image coding. Moreover, effectively capturing local, global, and channel-wise contexts with acceptable even linear complexity within a single entropy model remains a challenge. To address these limitations, we propose the Linear Complexity Multi-Reference Entropy Model (MEM++). MEM++ effectively captures the diverse range of correlations inherent in the latent representation. Specifically, the latent representation is first divided into multiple slices. When compressing a particular slice, the previously compressed slices serve as its channel-wise contexts. To capture local contexts without sacrificing performance, we introduce a novel checkerboard attention module. Additionally, to capture global contexts, we propose the linear complexity attention-based global correlations capturing by leveraging the decomposition of the softmax operation. The attention map of the previously decoded slice is implicitly computed and employed to predict global correlations in the current slice. Based on MEM++, we propose image compression model MLIC++. Extensive experimental evaluations demonstrate that our MLIC++ achieves state-of-the-art performance, reducing BD-rate by 13.39% on the Kodak dataset compared to VTM-17.0 in PSNR. Furthermore, MLIC++ exhibits linear GPU memory consumption with resolution, making it highly suitable for high-resolution image coding. Code and pre-trained models are available at https://github.com/JiangWeibeta/MLIC.
Neural operators, such as Fourier Neural Operators (FNO), form a principled approach for learning solution operators for PDEs and other mappings between function spaces. However, many real-world problems require high-resolution training data, and the training time and limited GPU memory pose big barriers. One solution is to train neural operators in mixed precision to reduce the memory requirement and increase training speed. However, existing mixed-precision training techniques are designed for standard neural networks, and we find that their direct application to FNO leads to numerical overflow and poor memory efficiency. Further, at first glance, it may appear that mixed precision in FNO will lead to drastic accuracy degradation since reducing the precision of the Fourier transform yields poor results in classical numerical solvers. We show that this is not the case; in fact, we prove that reducing the precision in FNO still guarantees a good approximation bound, when done in a targeted manner. Specifically, we build on the intuition that neural operator learning inherently induces an approximation error, arising from discretizing the infinite-dimensional ground-truth input function, implying that training in full precision is not needed. We formalize this intuition by rigorously characterizing the approximation and precision errors of FNO and bounding these errors for general input functions. We prove that the precision error is asymptotically comparable to the approximation error. Based on this, we design a simple method to optimize the memory-intensive half-precision tensor contractions by greedily finding the optimal contraction order. Through extensive experiments on different state-of-the-art neural operators, datasets, and GPUs, we demonstrate that our approach reduces GPU memory usage by up to 50% and improves throughput by 58% with little or no reduction in accuracy.
Performance Benchmarking of HPC systems is an ongoing effort that seeks to provide information that will allow for increased performance and improve the job schedulers that manage these systems. We develop a benchmarking tool that utilizes machine learning models and gathers performance data on GPU-accelerated nodes while they perform material segmentation analysis. The benchmark uses a ML model that has been converted from Caffe to PyTorch using the MMdnn toolkit and the MINC-2500 dataset. Performance data is gathered on two ERDC DSRC systems, Onyx and Vulcanite. The data reveals that while Vulcanite has faster model times in a large number of benchmarks, and it is also more subject to some environmental factors that can cause performances slower than Onyx. In contrast the model times from Onyx are consistent across benchmarks.
The goal of Online Domain Adaptation for semantic segmentation is to handle unforeseeable domain changes that occur during deployment, like sudden weather events. However, the high computational costs associated with brute-force adaptation make this paradigm unfeasible for real-world applications. In this paper we propose HAMLET, a Hardware-Aware Modular Least Expensive Training framework for real-time domain adaptation. Our approach includes a hardware-aware back-propagation orchestration agent (HAMT) and a dedicated domain-shift detector that enables active control over when and how the model is adapted (LT). Thanks to these advancements, our approach is capable of performing semantic segmentation while simultaneously adapting at more than 29FPS on a single consumer-grade GPU. Our framework's encouraging accuracy and speed trade-off is demonstrated on OnDA and SHIFT benchmarks through experimental results.
3D anomaly detection is an emerging and vital computer vision task in industrial manufacturing (IM). Recently many advanced algorithms have been published, but most of them cannot meet the needs of IM. There are several disadvantages: i) difficult to deploy on production lines since their algorithms heavily rely on large pre-trained models; ii) hugely increase storage overhead due to overuse of memory banks; iii) the inference speed cannot be achieved in real-time. To overcome these issues, we propose an easy and deployment-friendly network (called EasyNet) without using pre-trained models and memory banks: firstly, we design a multi-scale multi-modality feature encoder-decoder to accurately reconstruct the segmentation maps of anomalous regions and encourage the interaction between RGB images and depth images; secondly, we adopt a multi-modality anomaly segmentation network to achieve a precise anomaly map; thirdly, we propose an attention-based information entropy fusion module for feature fusion during inference, making it suitable for real-time deployment. Extensive experiments show that EasyNet achieves an anomaly detection AUROC of 92.6% without using pre-trained models and memory banks. In addition, EasyNet is faster than existing methods, with a high frame rate of 94.55 FPS on a Tesla V100 GPU.
We present YOLOBench, a benchmark comprised of 550+ YOLO-based object detection models on 4 different datasets and 4 different embedded hardware platforms (x86 CPU, ARM CPU, Nvidia GPU, NPU). We collect accuracy and latency numbers for a variety of YOLO-based one-stage detectors at different model scales by performing a fair, controlled comparison of these detectors with a fixed training environment (code and training hyperparameters). Pareto-optimality analysis of the collected data reveals that, if modern detection heads and training techniques are incorporated into the learning process, multiple architectures of the YOLO series achieve a good accuracy-latency trade-off, including older models like YOLOv3 and YOLOv4. We also evaluate training-free accuracy estimators used in neural architecture search on YOLOBench and demonstrate that, while most state-of-the-art zero-cost accuracy estimators are outperformed by a simple baseline like MAC count, some of them can be effectively used to predict Pareto-optimal detection models. We showcase that by using a zero-cost proxy to identify a YOLO architecture competitive against a state-of-the-art YOLOv8 model on a Raspberry Pi 4 CPU. The code and data are available at https://github.com/Deeplite/deeplite-torch-zoo
Neuromorphic processors have garnered considerable interest in recent years for their potential in energy-efficient and high-speed computing. The Locally Competitive Algorithm (LCA) has been utilized for power efficient sparse coding on neuromorphic processors, including the first Loihi processor. With the Loihi 2 processor enabling custom neuron models and graded spike communication, more complex implementations of LCA are possible. We present a new implementation of LCA designed for the Loihi 2 processor and perform an initial set of benchmarks comparing it to LCA on CPU and GPU devices. In these experiments LCA on Loihi 2 is orders of magnitude more efficient and faster for large sparsity penalties, while maintaining similar reconstruction quality. We find this performance improvement increases as the LCA parameters are tuned towards greater representation sparsity. Our study highlights the potential of neuromorphic processors, particularly Loihi 2, in enabling intelligent, autonomous, real-time processing on small robots, satellites where there are strict SWaP (small, lightweight, and low power) requirements. By demonstrating the superior performance of LCA on Loihi 2 compared to conventional computing device, our study suggests that Loihi 2 could be a valuable tool in advancing these types of applications. Overall, our study highlights the potential of neuromorphic processors for efficient and accurate data processing on resource-constrained devices.
Mesh-based numerical solvers are an important part in many design tool chains. However, accurate simulations like computational fluid dynamics are time and resource consuming which is why surrogate models are employed to speed-up the solution process. Machine Learning based surrogate models on the other hand are fast in predicting approximate solutions but often lack accuracy. Thus, the development of the predictor in a predictor-corrector approach is the focus here, where the surrogate model predicts a flow field and the numerical solver corrects it. This paper scales a state-of-the-art surrogate model from the domain of graph-based machine learning to industry-relevant mesh sizes of a numerical flow simulation. The approach partitions and distributes the flow domain to multiple GPUs and provides halo exchange between these partitions during training. The utilized graph neural network operates directly on the numerical mesh and is able to preserve complex geometries as well as all other properties of the mesh. The proposed surrogate model is evaluated with an application on a three dimensional turbomachinery setup and compared to a traditionally trained distributed model. The results show that the traditional approach produces superior predictions and outperforms the proposed surrogate model. Possible explanations, improvements and future directions are outlined.
The unprecedented accuracy of convolutional neural networks (CNNs) across a broad range of AI tasks has led to their widespread deployment in mobile and embedded settings. In a pursuit for high-performance and energy-efficient inference, significant research effort has been invested in the design of FPGA-based CNN accelerators. In this context, single computation engines constitute a popular approach to support diverse CNN modes without the overhead of fabric reconfiguration. Nevertheless, this flexibility often comes with significantly degraded performance on memory-bound layers and resource underutilisation due to the suboptimal mapping of certain layers on the engine's fixed configuration. In this work, we investigate the implications in terms of CNN engine design for a class of models that introduce a pre-convolution stage to decompress the weights at run time. We refer to these approaches as on-the-fly. This paper presents unzipFPGA, a novel CNN inference system that counteracts the limitations of existing CNN engines. The proposed framework comprises a novel CNN hardware architecture that introduces a weights generator module that enables the on-chip on-the-fly generation of weights, alleviating the negative impact of limited bandwidth on memory-bound layers. We further enhance unzipFPGA with an automated hardware-aware methodology that tailors the weights generation mechanism to the target CNN-device pair, leading to an improved accuracy-performance balance. Finally, we introduce an input selective processing element (PE) design that balances the load between PEs in suboptimally mapped layers. The proposed framework yields hardware designs that achieve an average of 2.57x performance efficiency gain over highly optimised GPU designs for the same power constraints and up to 3.94x higher performance density over a diverse range of state-of-the-art FPGA-based CNN accelerators.
Learned cardinality estimation methods have achieved high precision compared to traditional methods. Among learned methods, query-driven approaches have faced the workload drift problem for a long time. Although both data-driven and hybrid methods are proposed to avoid this problem, most of them suffer from high training and estimation costs, limited scalability, instability, and long-tail distribution problems on high-dimensional tables, which seriously affects the practical application of learned cardinality estimators. In this paper, we prove that most of these problems are directly caused by the widely used progressive sampling. We solve this problem by introducing predicate information into the autoregressive model and propose Duet, a stable, efficient, and scalable hybrid method to estimate cardinality directly without sampling or any non-differentiable process, which can not only reduce the inference complexity from $O(n)$ to $O(1)$ compared to Naru and UAE but also achieve higher accuracy on high cardinality and high-dimensional tables. Experimental results show that Duet can achieve all the design goals above and be much more practical. Besides, Duet even has a lower inference cost on CPU than that of most learned methods on GPU.
The curse-of-dimensionality taxes computational resources heavily with exponentially increasing computational cost as the dimension increases. This poses great challenges in solving high-dimensional PDEs, as Richard E. Bellman first pointed out over 60 years ago. While there has been some recent success in solving numerically partial differential equations (PDEs) in high dimensions, such computations are prohibitively expensive, and true scaling of general nonlinear PDEs to high dimensions has never been achieved. We develop a new method of scaling up physics-informed neural networks (PINNs) to solve arbitrary high-dimensional PDEs. The new method, called Stochastic Dimension Gradient Descent (SDGD), decomposes a gradient of PDEs into pieces corresponding to different dimensions and randomly samples a subset of these dimensional pieces in each iteration of training PINNs. We prove theoretically the convergence and other desired properties of the proposed method. We demonstrate in various diverse tests that the proposed method can solve many notoriously hard high-dimensional PDEs, including the Hamilton-Jacobi-Bellman (HJB) and the Schr\"{o}dinger equations in tens of thousands of dimensions very fast on a single GPU using the PINNs mesh-free approach. Notably, we solve nonlinear PDEs with nontrivial, anisotropic, and inseparable solutions in 100,000 effective dimensions in 12 hours on a single GPU using SDGD with PINNs. Since SDGD is a general training methodology of PINNs, it can be applied to any current and future variants of PINNs to scale them up for arbitrary high-dimensional PDEs.
This paper presents a low-cost network architecture for training large language models (LLMs) at hyperscale. We study the optimal parallelization strategy of LLMs and propose a novel datacenter network design tailored to LLM's unique communication pattern. We show that LLM training generates sparse communication patterns in the network and, therefore, does not require any-to-any full-bisection network to complete efficiently. As a result, our design eliminates the spine layer in traditional GPU clusters. We name this design a Rail-only network and demonstrate that it achieves the same training performance while reducing the network cost by 38% to 77% and network power consumption by 37% to 75% compared to a conventional GPU datacenter. Our architecture also supports Mixture-of-Expert (MoE) models with all-to-all communication through forwarding, with only 8.2% to 11.2% completion time overhead for all-to-all traffic. We study the failure robustness of Rail-only networks and provide insights into the performance impact of different network and training parameters.
Reinforcement learning is time-consuming for complex tasks due to the need for large amounts of training data. Recent advances in GPU-based simulation, such as Isaac Gym, have sped up data collection thousands of times on a commodity GPU. Most prior works used on-policy methods like PPO due to their simplicity and ease of scaling. Off-policy methods are more data efficient but challenging to scale, resulting in a longer wall-clock training time. This paper presents a Parallel $Q$-Learning (PQL) scheme that outperforms PPO in wall-clock time while maintaining superior sample efficiency of off-policy learning. PQL achieves this by parallelizing data collection, policy learning, and value learning. Different from prior works on distributed off-policy learning, such as Apex, our scheme is designed specifically for massively parallel GPU-based simulation and optimized to work on a single workstation. In experiments, we demonstrate that $Q$-learning can be scaled to \textit{tens of thousands of parallel environments} and investigate important factors affecting learning speed. The code is available at https://github.com/Improbable-AI/pql.