A5103201662- Advanced Data Storage Technologies
- Parallel Computing and Optimization Techniques
- Algorithms and Data Compression
- Advanced Data Compression Techniques
- Advanced Neural Network Applications
- Domain Adaptation and Few-Shot Learning
- Distributed and Parallel Computing Systems
- Computer Graphics and Visualization Techniques
- Brain Tumor Detection and Classification
- Machine Learning and Data Classification
- Adversarial Robustness in Machine Learning
- Generative Adversarial Networks and Image Synthesis
- Advanced Vision and Imaging
- Image and Signal Denoising Methods
- Time Series Analysis and Forecasting
- Neural Networks and Applications
- Privacy-Preserving Technologies in Data
- Scientific Computing and Data Management
- Advanced Image Processing Techniques
- Distributed systems and fault tolerance
- Stochastic Gradient Optimization Techniques
- Numerical Methods and Algorithms
- Data Management and Algorithms
- Machine Learning and Algorithms
- Multimodal Machine Learning Applications
Temple University
2023-2025
Indiana University Bloomington
2022-2024
Temple College
2024
Indiana University
2023
Washington State University
2020-2022
Illinois Institute of Technology
2021
University of California, Riverside
2021
University of South Carolina Upstate
2021
University of Houston
2021
Argonne National Laboratory
2021
Deep neural networks (DNNs) have achieved remarkable success in many fields. However, large-scale DNNs also bring storage costs when storing snapshots for preventing clusters' frequent failures or incur significant communication overheads transmitting the Federated Learning (FL). Recently, several approaches, such as Delta-DNN and LC-Checkpoint, aim to reduce size of DNNs' snapshot by compressing difference between two neighboring versions (a.k.a., delta). we observe that existing applying...
To help understand our universe better, researchers and scientists currently run extreme-scale cosmology simulations on leadership supercomputers. However, such can generate large amounts of scientific data, which often result in expensive costs data associated with movement storage. Lossy compression techniques have become attractive because they significantly reduce size maintain high fidelity for post-analysis. In this paper, we propose to use GPU-based lossy cosmological simulations. Our...
Error-bounded lossy compression is becoming an indispensable technique for the success of today’s scientific projects with vast volumes data produced during simulations or instrument acquisitions. Not only can it significantly reduce size, but also control errors based on user-specified error bounds. Autoencoder (AE) models have been widely used in image compression, few AE-based approaches support error-bounding features, which are highly required by applications. To address this issue, we...
Error-bounded lossy compression has been identified as a promising solution for significantly reducing scientific data volumes upon users' requirements on distortion. For the existing error-bounded compressors, some of them (such SPERR and FAZ) can reach fairly high ratios others SZx, SZ, ZFP) feature speeds, but they rarely exhibit both ratio speed meanwhile. In this paper, we propose HPEZ with newly-designed interpolations quality-metric-driven auto-tuning, which features improved quality...
Today's deep neural networks (DNNs) are becoming deeper and wider because of increasing demand on the analysis quality more complex applications to resolve. The wide DNNs, however, require large amounts resources (such as memory, storage, I/O), significantly restricting their utilization resource-constrained platforms. Although some DNN simplification methods weight quantization) have been proposed address this issue, they suffer from either low compression ratios or high errors, which may...
Error-bounded lossy compression is critical to the success of extreme-scale scientific research because ever-increasing volumes data produced by today's high-performance computing (HPC) applications. Not only can error-controlled compressors significantly reduce I/O and storage burden but they retain high fidelity for post analysis. Existing state-of-the-art compressors, however, generally suffer from relatively low decompression throughput (up hundreds megabytes per second on a single CPU...
Error-bounded lossy compression is one of the most effective techniques for reducing scientific data sizes. However, traditional trial-and-error approach used to configure compressors finding optimal trade-off between reconstructed quality and ratio prohibitively expensive. To resolve this issue, we develop a general-purpose analytical ratio-quality model based on prediction-based framework, which can effectively foresee reduced ratio, as well impact compressed post-hoc analysis quality. Our...
Error-bounded lossy compression is a critical technique for significantly reducing scientific data volumes. With ever-emerging heterogeneous high-performance computing (HPC) architecture, GPU-accelerated error-bounded compressors (such as CUSZ and cuZFP) have been developed. However, they suffer from either low performance or ratios. To this end, we propose CUSZ+ to target both high ratios throughputs. We identify that sparsity smoothness are key factors Our contributions in work fourfold:...
Error-bounded lossy compression is a state-of-the-art data reduction technique for HPC applications because it not only significantly reduces storage overhead but also can retain high fidelity postanalysis. Because supercomputers and are becoming heterogeneous using accelerator-based architectures, in particular GPUs, several development teams have recently released GPU versions of their compressors. However, existing GPU-based compressors suffer from either low decompression throughput or...
Lossy compression and asynchronous I/O are two of the most effective solutions for reducing storage overhead enhancing performance in large-scale high-performance computing (HPC) applications. However, current approaches have limitations that prevent them from fully leveraging lossy compression, they may also result task collisions, which restrict overall HPC To address these issues, we propose an optimization approach scheduling problem encompasses computation, I/O. Our algorithm adaptively...
Error-bounded lossy compression has been effective in significantly reducing the data storage/transfer burden while preserving reconstructed fidelity very well. Many error-bounded compressors have developed for a wide range of parallel and distributed use cases years. These are designed with distinct models design principles, such that each them features particular pros cons. In this paper we provide comprehensive survey emerging techniques different involving big to process. The key...
Deep neural networks (DNNs) are becoming increasingly deeper, wider, and non-linear due to the growing demands on prediction accuracy analysis quality. Training wide deep require large amounts of storage resources such as memory because intermediate activation data must be saved in during forward propagation then restored for backward propagation. However, state-of-the-art accelerators GPUs only equipped with very limited capacities hardware design constraints, which significantly limits...
Extreme-scale cosmological simulations have been widely used by today's researchers and scientists on leadership supercomputers. A new generation of error-bounded lossy compressors has in workflows to reduce storage requirements minimize the impact throughput limitations while saving large snapshots high-fidelity data for post-hoc analysis. In this paper, we propose adaptively provide compression configurations compute partitions with newly designed post-analysis aware rate-quality modeling....
Convolutional neural networks (CNNs) are becoming increasingly deeper, wider, and non-linear because of the growing demand on prediction accuracy analysis quality. The wide deep CNNs, however, require a large amount computing resources processing time. Many previous works have studied model pruning to improve inference performance, but little work has been done for effectively reducing training cost. In this paper, we propose ClickTrain: an efficient accurate end-to-end framework CNNs....
As supercomputers advance towards exascale capabilities, computational intensity increases significantly, and the volume of data requiring storage transmission experiences exponential growth. Adaptive Mesh Refinement (AMR) has emerged as an effective solution to address these two challenges. Concurrently, error-bounded lossy compression is recognized one most efficient approaches tackle latter issue. Despite their respective advantages, few attempts have been made investigate how AMR can...
Vast volumes of data are produced by today's scientific simulations and advanced instruments. These cannot be stored transferred efficiently because limited I/O bandwidth, network speed, storage capacity. Error-bounded lossy compression can an effective method for addressing these issues: not only it significantly reduce size, but also control the distortion based on user-defined error bounds. In practice, many applications have specific requirements or constraints compression, in order to...
Deep neural networks (DNNs) have gained considerable attention in various real-world applications due to the strong performance on representation learning. However, a DNN needs be trained many epochs for pursuing higher inference accuracy, which requires storing sequential versions of DNNs and releasing updated users. As result, large amounts storage network resources are required, significantly hampering utilization resource-constrained platforms (e.g., IoT, mobile phone).
DNNs are becoming increasingly deeper, wider, and nonlinear due to the growing demands on prediction accuracy analysis quality. When training a DNN model, intermediate activation data must be saved in memory during forward propagation then restored for backward propagation. Traditional saving techniques such as recomputation migration either suffers from high performance overhead or is constrained by specific interconnect technology limited bandwidth. In this paper, we propose novel...