Spiking Neural Networks (SNNs) have been attached great importance due to the distinctive properties of low power consumption, biological plausibility, and adversarial robustness. The most effective way to train deep SNNs is through ANN-to-SNN conversion, which have yielded the best performance in deep network structure and large-scale datasets. However, there is a trade-off between accuracy and latency. In order to achieve high precision as original ANNs, a long simulation time is needed to match the firing rate of a spiking neuron with the activation value of an analog neuron, which impedes the practical application of SNN. In this paper, we aim to achieve high-performance converted SNNs with extremely low latency (fewer than 32 time-steps). We start by theoretically analyzing ANN-to-SNN conversion and show that scaling the thresholds does play a similar role as weight normalization. Instead of introducing constraints that facilitate ANN-to-SNN conversion at the cost of model capacity, we applied a more direct way by optimizing the initial membrane potential to reduce the conversion loss in each layer. Besides, we demonstrate that optimal initialization of membrane potentials can implement expected error-free ANN-to-SNN conversion. We evaluate our algorithm on the CIFAR-10, CIFAR-100 and ImageNet datasets and achieve state-of-the-art accuracy, using fewer time-steps. For example, we reach top-1 accuracy of 93.38\% on CIFAR-10 with 16 time-steps. Moreover, our method can be applied to other ANN-SNN conversion methodologies and remarkably promote performance when the time-steps is small.
Deep learning on point clouds has received increased attention thanks to its wide applications in AR/VR and autonomous driving. These applications require low latency and high accuracy to provide real-time user experience and ensure user safety. Unlike conventional dense workloads, the sparse and irregular nature of point clouds poses severe challenges to running sparse CNNs efficiently on the general-purpose hardware. Furthermore, existing sparse acceleration techniques for 2D images do not translate to 3D point clouds. In this paper, we introduce TorchSparse, a high-performance point cloud inference engine that accelerates the sparse convolution computation on GPUs. TorchSparse directly optimizes the two bottlenecks of sparse convolution: irregular computation and data movement. It applies adaptive matrix multiplication grouping to trade computation for better regularity, achieving 1.4-1.5x speedup for matrix multiplication. It also optimizes the data movement by adopting vectorized, quantized and fused locality-aware memory access, reducing the memory movement cost by 2.7x. Evaluated on seven representative models across three benchmark datasets, TorchSparse achieves 1.6x and 1.5x measured end-to-end speedup over the state-of-the-art MinkowskiEngine and SpConv, respectively.
Currently, most single image dehazing models cannot run an ultra-high-resolution (UHD) image with a single GPU shader in real-time. To address the problem, we introduce the principle of infinite approximation of Taylor's theorem with the Laplace pyramid pattern to build a model which is capable of handling 4K hazy images in real-time. The N branch networks of the pyramid network correspond to the N constraint terms in Taylor's theorem. Low-order polynomials reconstruct the low-frequency information of the image (e.g. color, illumination). High-order polynomials regress the high-frequency information of the image (e.g. texture). In addition, we propose a Tucker reconstruction-based regularization term that acts on each branch network of the pyramid model. It further constrains the generation of anomalous signals in the feature space. Extensive experimental results demonstrate that our approach can not only run 4K images with haze in real-time on a single GPU (80FPS) but also has unparalleled interpretability. The developed method achieves state-of-the-art (SOTA) performance on two benchmarks (O/I-HAZE) and our updated 4KID dataset while providing the reliable groundwork for subsequent optimization schemes.
In order to improve performance of photovoltaic/thermal (or PV/T for simplicity) collectors, this paper firstly validated a previous computational thermal model and then introduced an improved computational thermal model to investigate the effects of the major control parameters on the thermal performance of PV/T collectors, including solar cell temperature, back surface temperature, and outlet water temperature. Besides, a computational electrical model of PV/T system was also introduced to elaborate the relationship of voltage, current and power of a PV module (MSX60 polycrystalline solar cell) used in an experiment in the literature. Simulation results agree with the experimental data very well. The effects of the time-steps from 1 hour to minute, which is closed to the real time, were also reported. At last, several suggestions to improve the efficiency of PV/T system were illustrated.
In distributed or federated optimization and learning, communication between the different computing units is often the bottleneck, and gradient compression is a widely used technique for reducing the number of bits sent within each communication round of iterative methods. There are two classes of compression operators and separate algorithms making use of them. In the case of unbiased random compressors with bounded variance (e.g., rand-k), the DIANA algorithm of Mishchenko et al. [2019], which implements a variance reduction technique for handling the variance introduced by compression, is the current state of the art. In the case of biased and contractive compressors (e.g., top-k), the EF21 algorithm of Richt\'arik et al. [2021], which implements an error-feedback mechanism for handling the error introduced by compression, is the current state of the art. These two classes of compression schemes and algorithms are distinct, with different analyses and proof techniques. In this paper, we unify them into a single framework and propose a new algorithm, recovering DIANA and EF21 as particular cases. We prove linear convergence under certain conditions. Our general approach works with a new, larger class of compressors, which includes unbiased and biased compressors as particular cases, and has two parameters, the bias and the variance. These gives a finer control and allows us to inherit the best of the two worlds: biased compressors, whose good performance in practice is recognized, can be used. And independent randomness at the compressors allows to mitigate the effects of compression, with the convergence rate improving when the number of parallel workers is large. This is the first time that an algorithm with all these features is proposed. Our approach takes a step towards better understanding of two so-far distinct worlds of communication-efficient distributed learning.
Histopathological images of tumors contain abundant information about how tumors grow and how they interact with their micro-environment. Characterizing and improving our understanding of phenotypes could reveal factors related to tumor progression and their underpinning biological processes, ultimately improving diagnosis and treatment. In recent years, the field of histological deep learning applications has seen great progress, yet most of these applications focus on a supervised approach, relating tissue and associated sample annotations. Supervised approaches have their impact limited by two factors. Firstly, high-quality labels are expensive in time and effort, which makes them not easily scalable. Secondly, these methods focus on predicting annotations from histological images, fundamentally restricting the discovery of new tissue phenotypes. These limitations emphasize the importance of using new methods that can characterize tissue by the features enclosed in the image, without pre-defined annotation or supervision. We present Phenotype Representation Learning (PRL), a methodology to extract histomorphological phenotypes through self-supervised learning and community detection. PRL creates phenotype clusters by identifying tissue patterns that share common morphological and cellular features, allowing to describe whole slide images through compositional representations of cluster contributions. We used this framework to analyze histopathology slides of LUAD and LUSC lung cancer subtypes from TCGA and NYU cohorts. We show that PRL achieves a robust lung subtype prediction providing statistically relevant phenotypes for each lung subtype. We further demonstrate the significance of these phenotypes in lung adenocarcinoma overall and recurrence free survival, relating clusters with patient outcomes, cell types, grown patterns, and omic-based immune signatures.
In this paper, we demonstrate a method for training speaker embedding extractors using weak annotation. More specifically, we are using the full VoxCeleb recordings and the name of the celebrities appearing on each video without knowledge of the time intervals the celebrities appear in the video. We show that by combining a baseline speaker diarization algorithm that requires no training or parameter tuning, a modified loss with aggregation over segments, and a two-stage training approach, we are able to train a competitive ResNet-based embedding extractor. Finally, we experiment with two different aggregation functions and analyze their behaviour in terms of their gradients.
List-mode positron emission tomography (PET) image reconstruction is an important tool for PET scanners with many lines-of-response (LORs) and additional information such as time-of-flight and depth-of-interaction. Deep learning is one possible solution to enhance the quality of PET image reconstruction. However, the application of deep learning techniques to list-mode PET image reconstruction have not been progressed because list data is a sequence of bit codes and unsuitable for processing by convolutional neural networks (CNN). In this study, we propose a novel list-mode PET image reconstruction method using an unsupervised CNN called deep image prior (DIP) and a framework of alternating direction method of multipliers. The proposed list-mode DIP reconstruction (LM-DIPRecon) method alternatively iterates regularized list-mode dynamic row action maximum likelihood algorithm (LM-DRAMA) and magnetic resonance imaging conditioned DIP (MR-DIP). We evaluated LM-DIPRecon using both simulation and clinical data, and it achieved sharper images and better tradeoff curves between contrast and noise than the LM-DRAMA and MR-DIP. These results indicated that the LM-DIPRecon is useful for quantitative PET imaging with limited events. In addition, as list data has finer temporal information than dynamic sinograms, list-mode deep image prior reconstruction is expected to be useful for 4D PET imaging and motion correction.
We propose an end-to-end real time framework to generate high resolution graphics grade textured 3D map of urban environment. The generated detailed map finds its application in the precise localization and navigation of autonomous vehicles. It can also serve as a virtual test bed for various vision and planning algorithms as well as a background map in the computer games. In this paper, we focus on two important issues: (i) incrementally generating a map with coherent 3D surface, in real time and (ii) preserving the quality of color texture. To handle the above issues, firstly, we perform a pose-refinement procedure which leverages camera image information, Delaunay triangulation and existing scan matching techniques to produce high resolution 3D map from the sparse input LIDAR scan. This 3D map is then texturized and accumulated by using a novel technique of ray-filtering which handles occlusion and inconsistencies in pose-refinement. Further, inspired by human fovea, we introduce foveal-processing which significantly reduces the computation time and also assists ray-filtering to maintain consistency in color texture and coherency in 3D surface of the output map. Moreover, we also introduce texture error (TE) and mean texture mapping error (MTME), which provides quantitative measure of texturing and overall quality of the textured maps.
Conventional breast cancer imaging techniques are nowadays based on the use of ionising radiations or ultrasound waves for the inspection of breast areas. Nevertheless, these conventional techniques present some drawbacks related to patient safety, processing time and resolution issues. In this framework, microwave imaging can represent a valid alternative or a complementary technique compared to other conventional medical imaging modalities since it is safe (using non-ionising radiations), relatively cheap and more comfortable from patient point of view. Unfortunately, it is slow and computationally expensive, which strongly limit its use in clinical scenarios. In this paper, an artificial neural network for effective and almost real-time breast imaging is proposed. First, a realistic breast-like phantom generator was developed for training the network. Subsequently, numerical analyses have been conducted for the optimisation and the performance evaluation of the approach. The results seem very promising in terms of recovery performance as well as for the computation burden.