PyTorchKR Papers

최신 논문

총 716편
태그·기간 등 상세 필터

카테고리 탐색

로그인 후 이 태그 구독

arXiv:2208.01641공식 코드 · Papers with Code

Streaming-capable High-performance Architecture of Learned Image Compression Codecs

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.

arXiv:2208.00001공식 코드 · Papers with Code

FastGeodis: Fast Generalised Geodesic Distance Transform

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

arXiv:2207.12547공식 코드 · Papers with Code

The BUTTER Zone: An Empirical Study of Training Dynamics in Fully Connected Neural Networks

We present an empirical dataset surveying the deep learning phenomenon on fully-connected feed-forward multilayer perceptron neural networks. The dataset, which is now freely available online, records the per-epoch training and generalization performance of 483 thousand distinct hyperparameter choices of architectures, tasks, depths, network sizes (number of parameters), learning rates, batch sizes, and regularization penalties. Repeating each experiment an average of 24 times resulted in 11 million total training runs and 40 billion epochs recorded. Accumulating this 1.7 TB dataset utilized 11 thousand CPU core-years, 72.3 GPU-years, and 163 node-years. In surveying the dataset, we observe durable patterns persisting across tasks and topologies. We aim to spark scientific study of machine learning techniques as a catalyst for the theoretical discoveries needed to progress the field beyond energy-intensive and heuristic practices.

arXiv:2207.12620공식 코드 · Papers with Code

Large-displacement 3D Object Tracking with Hybrid Non-local Optimization

Optimization-based 3D object tracking is known to be precise and fast, but sensitive to large inter-frame displacements. In this paper we propose a fast and effective non-local 3D tracking method. Based on the observation that erroneous local minimum are mostly due to the out-of-plane rotation, we propose a hybrid approach combining non-local and local optimizations for different parameters, resulting in efficient non-local search in the 6D pose space. In addition, a precomputed robust contour-based tracking method is proposed for the pose optimization. By using long search lines with multiple candidate correspondences, it can adapt to different frame displacements without the need of coarse-to-fine search. After the pre-computation, pose updates can be conducted very fast, enabling the non-local optimization to run in real time. Our method outperforms all previous methods for both small and large displacements. For large displacements, the accuracy is greatly improved ($81.7\% \;\text{v.s.}\; 19.4\%$). At the same time, real-time speed ($>$50fps) can be achieved with only CPU. The source code is available at \url{https://github.com/cvbubbles/nonlocal-3dtracking}.

arXiv:2207.11512공식 코드 · Papers with Code

Combining Self-Training and Hybrid Architecture for Semi-supervised Abdominal Organ Segmentation

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.

arXiv:2207.10482공식 코드 · Papers with Code

LPYOLO: Low Precision YOLO for Face Detection on FPGA

In recent years, number of edge computing devices and artificial intelligence applications on them have advanced excessively. In edge computing, decision making processes and computations are moved from servers to edge devices. Hence, cheap and low power devices are required. FPGAs are very low power, inclined to do parallel operations and deeply suitable devices for running Convolutional Neural Networks (CNN) which are the fundamental unit of an artificial intelligence application. Face detection on surveillance systems is the most expected application on the security market. In this work, TinyYolov3 architecture is redesigned and deployed for face detection. It is a CNN based object detection method and developed for embedded systems. PYNQ-Z2 is selected as a target board which has low-end Xilinx Zynq 7020 System-on-Chip (SoC) on it. Redesigned TinyYolov3 model is defined in numerous bit width precisions with Brevitas library which brings fundamental CNN layers and activations in integer quantized form. Then, the model is trained in a quantized structure with WiderFace dataset. In order to decrease latency and power consumption, onchip memory of the FPGA is configured as a storage of whole network parameters and the last activation function is modified as rescaled HardTanh instead of Sigmoid. Also, high degree of parallelism is applied to logical resources of the FPGA. The model is converted to an HLS based application with using FINN framework and FINN-HLS library which includes the layer definitions in C++. Later, the model is synthesized and deployed. CPU of the SoC is employed with multithreading mechanism and responsible for preprocessing, postprocessing and TCP/IP streaming operations. Consequently, 2.4 Watt total board power consumption, 18 Frames-Per-Second (FPS) throughput and 0.757 mAP accuracy rate on Easy category of the WiderFace are achieved with 4 bits precision model.

arXiv:2207.09765공식 코드 · Papers with Code

ApHMM: Accelerating Profile Hidden Markov Models for Fast and Energy-Efficient Genome Analysis

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.

arXiv:2207.07886공식 코드 · Papers with Code

An Experimental Evaluation of Machine Learning Training on a Real Processing-in-Memory System

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.

arXiv:2207.03206공식 코드 · Papers with Code

Leveraging Log Instructions in Log-based Anomaly Detection

Artificial Intelligence for IT Operations (AIOps) describes the process of maintaining and operating large IT systems using diverse AI-enabled methods and tools for, e.g., anomaly detection and root cause analysis, to support the remediation, optimization, and automatic initiation of self-stabilizing IT activities. The core step of any AIOps workflow is anomaly detection, typically performed on high-volume heterogeneous data such as log messages (logs), metrics (e.g., CPU utilization), and distributed traces. In this paper, we propose a method for reliable and practical anomaly detection from system logs. It overcomes the common disadvantage of related works, i.e., the need for a large amount of manually labeled training data, by building an anomaly detection model with log instructions from the source code of 1000+ GitHub projects. The instructions from diverse systems contain rich and heterogenous information about many different normal and abnormal IT events and serve as a foundation for anomaly detection. The proposed method, named ADLILog, combines the log instructions and the data from the system of interest (target system) to learn a deep neural network model through a two-phase learning procedure. The experimental results show that ADLILog outperforms the related approaches by up to 60% on the F1 score while satisfying core non-functional requirements for industrial deployments such as unsupervised design, efficient model updates, and small model sizes.

arXiv:2207.00032공식 코드 · Papers with Code

DeepSpeed Inference: Enabling Efficient Inference of Transformer Models at Unprecedented Scale

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).

arXiv:2206.08107공식 코드 · Papers with Code

Closed-Form Diffeomorphic Transformations for Time Series Alignment

Time series alignment methods call for highly expressive, differentiable and invertible warping functions which preserve temporal topology, i.e diffeomorphisms. Diffeomorphic warping functions can be generated from the integration of velocity fields governed by an ordinary differential equation (ODE). Gradient-based optimization frameworks containing diffeomorphic transformations require to calculate derivatives to the differential equation's solution with respect to the model parameters, i.e. sensitivity analysis. Unfortunately, deep learning frameworks typically lack automatic-differentiation-compatible sensitivity analysis methods; and implicit functions, such as the solution of ODE, require particular care. Current solutions appeal to adjoint sensitivity methods, ad-hoc numerical solvers or ResNet's Eulerian discretization. In this work, we present a closed-form expression for the ODE solution and its gradient under continuous piecewise-affine (CPA) velocity functions. We present a highly optimized implementation of the results on CPU and GPU. Furthermore, we conduct extensive experiments on several datasets to validate the generalization ability of our model to unseen data for time-series joint alignment. Results show significant improvements both in terms of efficiency and accuracy.

arXiv:2206.06079공식 코드 · Papers with Code

OHM: GPU Based Occupancy Map Generation

Occupancy grid maps (OGMs) are fundamental to most systems for autonomous robotic navigation. However, CPU-based implementations struggle to keep up with data rates from modern 3D lidar sensors, and provide little capacity for modern extensions which maintain richer voxel representations. This paper presents OHM, our open source, GPU-based OGM framework. We show how the algorithms can be mapped to GPU resources, resolving difficulties with contention to obtain a successful implementation. The implementation supports many modern OGM algorithms including NDT-OM, NDT-TM, decay-rate and TSDF. A thorough performance evaluation is presented based on tracked and quadruped UGV platforms and UAVs, and data sets from both outdoor and subterranean environments. The results demonstrate excellent performance improvements both offline, and for online processing in embedded platforms. Finally, we describe how OHM was a key enabler for the UGV navigation solution for our entry in the DARPA Subterranean Challenge, which placed second at the Final Event.

arXiv:2206.00065공식 코드 · Papers with Code

FELARE: Fair Scheduling of Machine Learning Tasks on Heterogeneous Edge Systems

Edge computing enables smart IoT-based systems via concurrent and continuous execution of latency-sensitive machine learning (ML) applications. These edge-based machine learning systems are often battery-powered (i.e., energy-limited). They use heterogeneous resources with diverse computing performance (e.g., CPU, GPU, and/or FPGAs) to fulfill the latency constraints of ML applications. The challenge is to allocate user requests for different ML applications on the Heterogeneous Edge Computing Systems (HEC) with respect to both the energy and latency constraints of these systems. To this end, we study and analyze resource allocation solutions that can increase the on-time task completion rate while considering the energy constraint. Importantly, we investigate edge-friendly (lightweight) multi-objective mapping heuristics that do not become biased toward a particular application type to achieve the objectives; instead, the heuristics consider "fairness" across the concurrent ML applications in their mapping decisions. Performance evaluations demonstrate that the proposed heuristic outperforms widely-used heuristics in heterogeneous systems in terms of the latency and energy objectives, particularly, at low to moderate request arrival rates. We observed 8.9% improvement in on-time task completion rate and 12.6% in energy-saving without imposing any significant overhead on the edge system.

arXiv:2205.15795공식 코드 · Papers with Code

A Meta Reinforcement Learning Approach for Predictive Autoscaling in the Cloud

Predictive autoscaling (autoscaling with workload forecasting) is an important mechanism that supports autonomous adjustment of computing resources in accordance with fluctuating workload demands in the Cloud. In recent works, Reinforcement Learning (RL) has been introduced as a promising approach to learn the resource management policies to guide the scaling actions under the dynamic and uncertain cloud environment. However, RL methods face the following challenges in steering predictive autoscaling, such as lack of accuracy in decision-making, inefficient sampling and significant variability in workload patterns that may cause policies to fail at test time. To this end, we propose an end-to-end predictive meta model-based RL algorithm, aiming to optimally allocate resource to maintain a stable CPU utilization level, which incorporates a specially-designed deep periodic workload prediction model as the input and embeds the Neural Process to guide the learning of the optimal scaling actions over numerous application services in the Cloud. Our algorithm not only ensures the predictability and accuracy of the scaling strategy, but also enables the scaling decisions to adapt to the changing workloads with high sample efficiency. Our method has achieved significant performance improvement compared to the existing algorithms and has been deployed online at Alipay, supporting the autoscaling of applications for the world-leading payment platform.

arXiv:2205.14756공식 코드 · Papers with Code

EfficientViT: Multi-Scale Linear Attention for High-Resolution Dense Prediction

High-resolution dense prediction enables many appealing real-world applications, such as computational photography, autonomous driving, etc. However, the vast computational cost makes deploying state-of-the-art high-resolution dense prediction models on hardware devices difficult. This work presents EfficientViT, a new family of high-resolution vision models with novel multi-scale linear attention. Unlike prior high-resolution dense prediction models that rely on heavy softmax attention, hardware-inefficient large-kernel convolution, or complicated topology structure to obtain good performances, our multi-scale linear attention achieves the global receptive field and multi-scale learning (two desirable features for high-resolution dense prediction) with only lightweight and hardware-efficient operations. As such, EfficientViT delivers remarkable performance gains over previous state-of-the-art models with significant speedup on diverse hardware platforms, including mobile CPU, edge GPU, and cloud GPU. Without performance loss on Cityscapes, our EfficientViT provides up to 13.9$\times$ and 6.2$\times$ GPU latency reduction over SegFormer and SegNeXt, respectively. For super-resolution, EfficientViT delivers up to 6.4x speedup over Restormer while providing 0.11dB gain in PSNR. For Segment Anything, EfficientViT delivers 48.9x higher throughput on A100 GPU while achieving slightly better zero-shot instance segmentation performance on COCO.

arXiv:2205.11854공식 코드 · Papers with Code

Multi-Agent Collaborative Inference via DNN Decoupling: Intermediate Feature Compression and Edge Learning

Recently, deploying deep neural network (DNN) models via collaborative inference, which splits a pre-trained model into two parts and executes them on user equipment (UE) and edge server respectively, becomes attractive. However, the large intermediate feature of DNN impedes flexible decoupling, and existing approaches either focus on the single UE scenario or simply define tasks considering the required CPU cycles, but ignore the indivisibility of a single DNN layer. In this paper, we study the multi-agent collaborative inference scenario, where a single edge server coordinates the inference of multiple UEs. Our goal is to achieve fast and energy-efficient inference for all UEs. To achieve this goal, we first design a lightweight autoencoder-based method to compress the large intermediate feature. Then we define tasks according to the inference overhead of DNNs and formulate the problem as a Markov decision process (MDP). Finally, we propose a multi-agent hybrid proximal policy optimization (MAHPPO) algorithm to solve the optimization problem with a hybrid action space. We conduct extensive experiments with different types of networks, and the results show that our method can reduce up to 56\% of inference latency and save up to 72\% of energy consumption.

arXiv:2205.11498공식 코드 · Papers with Code

Injecting Domain Adaptation with Learning-to-hash for Effective and Efficient Zero-shot Dense Retrieval

Dense retrieval overcome the lexical gap and has shown great success in ad-hoc information retrieval (IR). Despite their success, dense retrievers are expensive to serve across practical use cases. For use cases requiring to search from millions of documents, the dense index becomes bulky and requires high memory usage for storing the index. More recently, learning-to-hash (LTH) techniques, for e.g., BPR and JPQ, produce binary document vectors, thereby reducing the memory requirement to efficiently store the dense index. LTH techniques are supervised and finetune the retriever using a ranking loss. They outperform their counterparts, i.e., traditional out-of-the-box vector compression techniques such as PCA or PQ. A missing piece from prior work is that existing techniques have been evaluated only in-domain, i.e., on a single dataset such as MS MARCO. In our work, we evaluate LTH and vector compression techniques for improving the downstream zero-shot retrieval accuracy of the TAS-B dense retriever while maintaining efficiency at inference. Our results demonstrate that, unlike prior work, LTH strategies when applied naively can underperform the zero-shot TAS-B dense retriever on average by up to 14% nDCG@10 on the BEIR benchmark. To solve this limitation, in our work, we propose an easy yet effective solution of injecting domain adaptation with existing supervised LTH techniques. We experiment with two well-known unsupervised domain adaptation techniques: GenQ and GPL. Our domain adaptation injection technique can improve the downstream zero-shot retrieval effectiveness for both BPR and JPQ variants of the TAS-B model by on average 11.5% and 8.2% nDCG@10 while both maintaining 32$\times$ memory efficiency and 14$\times$ and 2$\times$ speedup respectively in CPU retrieval latency on BEIR. All our code, models, and data are publicly available at https://github.com/thakur-nandan/income.

arXiv:2205.10034공식 코드 · Papers with Code

MoESys: A Distributed and Efficient Mixture-of-Experts Training and Inference System for Internet Services

While modern internet services, such as chatbots, search engines, and online advertising, demand the use of large-scale deep neural networks (DNNs), distributed training and inference over heterogeneous computing systems are desired to facilitate these DNN models. Mixture-of-Experts (MoE) is one the most common strategies to lower the cost of training subject to the overall size of models/data through gating and parallelism in a divide-and-conquer fashion. While DeepSpeed has made efforts in carrying out large-scale MoE training over heterogeneous infrastructures, the efficiency of training and inference could be further improved from several system aspects, including load balancing, communication/computation efficiency, and memory footprint limits. In this work, we present a novel MoESys that boosts efficiency in both large-scale training and inference. Specifically, in the training procedure, the proposed MoESys adopts an Elastic MoE training strategy with 2D prefetch and Fusion communication over Hierarchical storage, so as to enjoy efficient parallelisms. For scalable inference in a single node, especially when the model size is larger than GPU memory, MoESys builds the CPU-GPU memory jointly into a ring of sections to load the model, and executes the computation tasks across the memory sections in a round-robin manner for efficient inference. We carried out extensive experiments to evaluate MoESys, where MoESys successfully trains a Unified Feature Optimization (UFO) model with a Sparsely-Gated Mixture-of-Experts model of 12B parameters in 8 days on 48 A100 GPU cards. The comparison against the state-of-the-art shows that MoESys outperformed DeepSpeed with 33% higher throughput (tokens per second) in training and 13% higher throughput in inference in general. Particularly, under unbalanced MoE Tasks, e.g., UFO, MoESys achieved 64% higher throughput with 18% lower memory footprints.

arXiv:2205.09707공식 코드 · Papers with Code

PLAID: An Efficient Engine for Late Interaction Retrieval

Pre-trained language models are increasingly important components across multiple information retrieval (IR) paradigms. Late interaction, introduced with the ColBERT model and recently refined in ColBERTv2, is a popular paradigm that holds state-of-the-art status across many benchmarks. To dramatically speed up the search latency of late interaction, we introduce the Performance-optimized Late Interaction Driver (PLAID). Without impacting quality, PLAID swiftly eliminates low-scoring passages using a novel centroid interaction mechanism that treats every passage as a lightweight bag of centroids. PLAID uses centroid interaction as well as centroid pruning, a mechanism for sparsifying the bag of centroids, within a highly-optimized engine to reduce late interaction search latency by up to 7$\times$ on a GPU and 45$\times$ on a CPU against vanilla ColBERTv2, while continuing to deliver state-of-the-art retrieval quality. This allows the PLAID engine with ColBERTv2 to achieve latency of tens of milliseconds on a GPU and tens or just few hundreds of milliseconds on a CPU at large scale, even at the largest scales we evaluate with 140M passages.

arXiv:2205.09248공식 코드 · Papers with Code

MESH2IR: Neural Acoustic Impulse Response Generator for Complex 3D Scenes

We propose a mesh-based neural network (MESH2IR) to generate acoustic impulse responses (IRs) for indoor 3D scenes represented using a mesh. The IRs are used to create a high-quality sound experience in interactive applications and audio processing. Our method can handle input triangular meshes with arbitrary topologies (2K - 3M triangles). We present a novel training technique to train MESH2IR using energy decay relief and highlight its benefits. We also show that training MESH2IR on IRs preprocessed using our proposed technique significantly improves the accuracy of IR generation. We reduce the non-linearity in the mesh space by transforming 3D scene meshes to latent space using a graph convolution network. Our MESH2IR is more than 200 times faster than a geometric acoustic algorithm on a CPU and can generate more than 10,000 IRs per second on an NVIDIA GeForce RTX 2080 Ti GPU for a given furnished indoor 3D scene. The acoustic metrics are used to characterize the acoustic environment. We show that the acoustic metrics of the IRs predicted from our MESH2IR match the ground truth with less than 10% error. We also highlight the benefits of MESH2IR on audio and speech processing applications such as speech dereverberation and speech separation. To the best of our knowledge, ours is the first neural-network-based approach to predict IRs from a given 3D scene mesh in real-time.