Matching Items (2)
Filtering by
- All Subjects: Model Compression
- All Subjects: deduplication
- Creators: Yang, Yingzhen
Description
Machine learning models are increasingly employed by smart devices on the edge to support important applications such as real-time virtual assistants and privacy-preserving healthcare. However, deploying state-of-the-art (SOTA) deep learning models on devices faces multiple serious challenges. First, it is infeasible to deploy large models on resource-constrained edge devices whereas small models cannot achieve the SOTA accuracy. Second, it is difficult to customize the models according to diverse application requirements in accuracy and speed and diverse capabilities of edge devices. This study proposes several novel solutions to comprehensively address the above challenges through automated and improved model compression. First, it introduces Automatic Attention Pruning (AAP), an adaptive, attention-based pruning approach to automatically reduce model parameters while meeting diverse user objectives in model size, speed, and accuracy. AAP achieves an impressive 92.72% parameter reduction in ResNet-101 on Tiny-ImageNet without causing any accuracy loss. Second, it presents Self-Supervised Quantization-Aware Knowledge Distillation (SQAKD), a framework for reducing model precision without supervision from labeled training data. For example, it quantizes VGG-8 to 2 bits on CIFAR-10 without any accuracy loss. Finally, the study explores two more works, Contrastive Knowledge Distillation Framework (CKDF) and Log-Curriculum based Module Replacing (LCMR), for further improving the performance of small models. All the works proposed in this study are designed to address real-world challenges, and have been successfully deployed on diverse hardware platforms, including cloud instances and edge devices, catalyzing AI for the edge.
ContributorsZhao, Kaiqi (Author) / Zhao, Ming (Thesis advisor) / Li, Baoxin (Committee member) / Zou, Jia (Committee member) / Yang, Yingzhen (Committee member) / Arizona State University (Publisher)
Created2024
Description
The meteoric rise of Deep Neural Networks (DNN) has led to the development of various Machine Learning (ML) frameworks (e.g., Tensorflow, PyTorch). Every ML framework has a different way of handling DNN models, data types, operations involved, and the internal representations stored on disk or memory. There have been initiatives such as the Open Neural Network Exchange (ONNX) for a more standardized approach to machine learning for better interoperability between the various popular ML frameworks. Model Serving Platforms (MSP) (e.g., Tensorflow Serving, Clipper) are used for serving DNN models to applications and edge devices. These platforms have gained widespread use for their flexibility in serving DNN models created by various ML frameworks. They also have additional capabilities such as caching, automatic ensembling, and scheduling. However, few of these frameworks focus on optimizing the storage of these DNN models, some of which may take up to ∼130GB storage space(“Turing-NLG: A 17-billion-parameter language model by Microsoft” 2020). These MSPs leave it to the ML frameworks for optimizing the DNN model with various model compression techniques, such as quantization and pruning. This thesis investigates the viability of automatic cross-model compression using traditional deduplication techniques and storage optimizations. Scenarios are identified where different DNN models have shareable model weight parameters. “Chunking” a model into smaller pieces is explored as an approach for deduplication. This thesis also proposes a design for storage in a Relational Database Management System (RDBMS) that allows for automatic cross-model deduplication.
ContributorsDas, Amitabh (Author) / Zou, Jia (Thesis advisor) / Zhao, Ming (Thesis advisor) / Yang, Yingzhen (Committee member) / Arizona State University (Publisher)
Created2021