skip to main content

This content will become publicly available on October 31, 2024

Title: Survey: Exploiting Data Redundancy for Optimization of Deep Learning
Data redundancy is ubiquitous in the inputs and intermediate results of Deep Neural Networks (DNN) . It offers many significant opportunities for improving DNN performance and efficiency and has been explored in a large body of work. These studies have scattered in many venues across several years. The targets they focus on range from images to videos and texts, and the techniques they use to detect and exploit data redundancy also vary in many aspects. There is not yet a systematic examination and summary of the many efforts, making it difficult for researchers to get a comprehensive view of the prior work, the state of the art, differences and shared principles, and the areas and directions yet to explore. This article tries to fill the void. It surveys hundreds of recent papers on the topic, introduces a novel taxonomy to put the various techniques into a single categorization framework, offers a comprehensive description of the main methods used for exploiting data redundancy in improving multiple kinds of DNNs on data, and points out a set of research opportunities for future exploration.  more » « less
Award ID(s):
2146873 2047516
Author(s) / Creator(s):
; ; ; ;
Date Published:
Journal Name:
ACM Computing Surveys
Page Range / eLocation ID:
1 to 38
Medium: X
Sponsoring Org:
National Science Foundation
More Like this
  1. Abstract

    The process of tip‐enhanced Raman scattering (TERS) depends critically on the morphology near the apex of the tip used in the experiment. Many tip designs have focused on optimization of electromagnetic enhancement in the near‐field, which is controlled to a large extent by subtle details at the nanoscale that remain difficult to reproduce in the tip fabrication process. The use of focused ion beams (FIB) permit modification of larger features on the tip in a reproducible manner, yet this approach cannot produce sub‐20‐nm structures important for optimum near‐field enhancement. Nonetheless, FIB milling offers excellent opportunities for improving the far‐field radiation properties of the tip‐antenna, a feature that has received relatively little attention in the TERS research community. In this work, we use finite‐difference time‐domain (FDTD) simulations to study both the near‐field and far‐field radiation efficiency of several tip‐antenna systems that can be constructed with FIB techniques in a feasible manner. Starting from blunt etched tips, we find that excellent overall enhancement of the TERS signal can be obtained with pillar‐type tips. Furthermore, by applying vertical grooves on the tip's shaft, the overall efficiency can be improved even more, producing TERS signals that are up to 10‐fold stronger than signals obtained from an ideal (unmodified) sharp tip of 10‐nm radius. The proposed designs constitute a feasible route toward a tip fabrication process that not only yields more reproducible tips but also promises much stronger TERS signals.

    more » « less
  2. Deep Neural Networks (DNNs) are pervasively applied in many artificial intelligence (AI) applications. The high performance of DNNs comes at the cost of larger size and higher compute complexity. Recent studies show that DNNs have much redundancy, such as the zero-value parameters and excessive numerical precision. To reduce computing complexity, many redundancy reduction techniques have been proposed, including pruning and data quantization. In this paper, we demonstrate our cooptimization of the DNN algorithm and hardware which exploits the model redundancy to accelerate DNNs. 
    more » « less
  3. Model compression is an important technique to facilitate efficient embedded and hardware implementations of deep neural networks (DNNs), a number of prior works are dedicated to model compression techniques. The target is to simultaneously reduce the model storage size and accelerate the computation, with minor effect on accuracy. Two important categories of DNN model compression techniques are weight pruning and weight quantization. The former leverages the redundancy in the number of weights, whereas the latter leverages the redundancy in bit representation of weights. These two sources of redundancy can be combined, thereby leading to a higher degree of DNN model compression. However, a systematic framework of joint weight pruning and quantization of DNNs is lacking, thereby limiting the available model compression ratio. Moreover, the computation reduction, energy efficiency improvement, and hardware performance overhead need to be accounted besides simply model size reduction, and the hardware performance overhead resulted from weight pruning method needs to be taken into consideration. To address these limitations, we present ADMM-NN, the first algorithm-hardware co-optimization framework of DNNs using Alternating Direction Method of Multipliers (ADMM), a powerful technique to solve non-convex optimization problems with possibly combinatorial constraints. The first part of ADMM-NN is a systematic, joint framework of DNN weight pruning and quantization using ADMM. It can be understood as a smart regularization technique with regularization target dynamically updated in each ADMM iteration, thereby resulting in higher performance in model compression than the state-of-the-art. The second part is hardware-aware DNN optimizations to facilitate hardware-level implementations. We perform ADMM-based weight pruning and quantization considering (i) the computation reduction and energy efficiency improvement, and (ii) the hardware performance overhead due to irregular sparsity. The first requirement prioritizes the convolutional layer compression over fully-connected layers, while the latter requires a concept of the break-even pruning ratio, defined as the minimum pruning ratio of a specific layer that results in no hardware performance degradation. Without accuracy loss, ADMM-NN achieves 85× and 24× pruning on LeNet-5 and AlexNet models, respectively, --- significantly higher than the state-of-the-art. The improvements become more significant when focusing on computation reduction. Combining weight pruning and quantization, we achieve 1,910× and 231× reductions in overall model size on these two benchmarks, when focusing on data storage. Highly promising results are also observed on other representative DNNs such as VGGNet and ResNet-50. We release codes and models at 
    more » « less
  4. null (Ed.)
    Training Deep Neural Networks (DNNs) is resource-intensive and time-consuming. While prior research has explored many different ways of reducing DNN training time, the impact of input data pipeline , i.e., fetching raw data items from storage and performing data pre-processing in memory, has been relatively unexplored. This paper makes the following contributions: (1) We present the first comprehensive analysis of how the input data pipeline affects the training time of widely-used computer vision and audio Deep Neural Networks (DNNs), that typically involve complex data pre-processing. We analyze nine different models across three tasks and four datasets while varying factors such as the amount of memory, number of CPU threads, storage device, GPU generation etc on servers that are a part of a large production cluster at Microsoft. We find that in many cases, DNN training time is dominated by data stall time : time spent waiting for data to be fetched and pre-processed. (2) We build a tool, DS-Analyzer to precisely measure data stalls using a differential technique, and perform predictive what-if analysis on data stalls. (3) Finally, based on the insights from our analysis, we design and implement three simple but effective techniques in a data-loading library, CoorDL, to mitigate data stalls. Our experiments on a range of DNN tasks, models, datasets, and hardware configs show that when PyTorch uses CoorDL instead of the state-of-the-art DALI data loading library, DNN training time is reduced significantly (by as much as 5X on a single server). 
    more » « less
  5. Undergraduate instructional biology laboratories are typically taught within two paradigms. Some labs focus on protocols and techniques delivered in “cookbook” format with defined experimental outcomes. There is increasing momentum to alternatively employ student-driven, open-ended, and discovery-based strategies, oftenviacourse-based undergraduate research experiences (CUREs) using crowd-sourcing initiatives. A fraction of students also participate in funded research in faculty research labs, where they have opportunities to work on projects designed to expand the frontiers of human knowledge. These experiences are widely recognized as valuable but are not scalable, as most institutions have many more undergraduates than research lab positions. We sought to address this gap through our department’s curriculum by creating an opportunity for students to participate in the real-world research process within a laboratory course. We conceived, developed, and delivered an authentic, guided research experience to students in an upper-level molecular biology laboratory course. We refer to this model as a “research program-linked CURE.” The research questions come directly from a faculty member’s research lab and evolve along with that research program. Students study post-transcriptional regulation in mycobacteria. We use current molecular biology methodologies to test hypotheses like “UTRs affect RNA and protein expression levels,” “there is functional redundancy among RNA helicases,” and “carbon starvation alters mRNA 5′ end chemistries.” We conducted standard assessments and developed a customized “Skills and Concepts Inventory” survey to gauge how well the course met our student learning outcomes. We report the results of our assessments and describe challenges addressed during development and execution of the course, including organizing activities to fit within an instructional lab, balancing breadth with depth, and maintaining authenticity while giving students the experience of obtaining interpretable and novel results. Our data suggest student learning was enhanced through this truly authentic research approach. Further, students were able to perceive they were participants and contributors within an active research paradigm. Students reported increases in their self-identification as scientists, and a positive impact on their career trajectories. An additional benefit was reciprocation back to the funded research laboratory, by funneling course alumni, results, materials, and protocols.

    more » « less