- This event has passed.
Huan Wang PhD Dissertation Defense
March 28, 2024 @ 11:00 am - 12:00 pm
Announcing:
PhD Dissertation Defense
Name:
Huan Wang
Title:
Towards Efficient Deep Learning in Computer Vision via Network Sparsity and Distillation
Date:
3/28/2024
Time:
11:00:00 AM
Committee Members:
Prof. Yun Fu (Advisor)
Prof. Octavia Camps
Prof. Zhiqiang Tao
Abstract:
AI, empowered by deep learning, has been profoundly transforming the world. However, the excessive size of these models remains a central obstacle that limits their broader utility. Modern neural networks commonly consist of millions of parameters, with foundation models extending to billions. The rapid expansion in model size introduces many challenges including training cost, sluggish inference speed, excessive energy consumption, and negative environmental implications such as increased CO2 emissions.
Addressing these challenges necessitates the adoption of efficient deep learning. The dissertation focuses on two overarching approaches, network pruning and knowledge distillation, to enhance the efficiency of deep learning models in the context of computer vision. Network pruning focuses on eliminating redundant parameters in a model while preserving the performance. Knowledge distillation aims to enhance the performance of the target model, referred to as the “student,” by leveraging guidance from a stronger model, known as the “teacher”. This approach leads to performance improvements in the target model without reducing its size.
In this defense presentation, I will start with the background and major challenges of leveraging these techniques to improve the efficiency of deep neural networks. Then, I shall present the proposed solutions for various vision tasks, including image classification, single-image super-resolution, novel view synthesis / neural rendering / NeRF / NeLF, text-to-image generation / diffusion models, and photorealistic head avatars. Extensive results and analyses will justify the efficacy of the proposed approaches, demonstrating that pruning and distillation make a generic and complete framework for efficient deep learning in various domains. Finally, a comprehensive summary (with takeaways) and outlook of the future work will conclude the presentation.