[CS] Announcement of Chengcheng Wan's Candidacy Exam

[nitayack at uchicago.edu]

This is an announcement of Chengcheng Wan's Candidacy Exam
===============================================
Date: Friday, July 2, 2021

Time: 2 pm CST

Location: https://zoom.us/j/8759657011?pwd=Q0ttZkpIQ01UcWNQd2E1ZFNiaDZpdz09 Meeting ID: 875 965 7011
Passcode: PxN4Mt

Candidacy Exam Candidate: Chengcheng Wan

Title:  Accurate anytime learning for energy and timeliness in software systems

Abstract: Machine learning (ML) provides efficient solutions for a number of problems that were difficult to solve with traditional computing techniques. Deep neural networks (DNNs) have become a key workload for many computing systems due to their high inference accuracy. This accuracy, however, comes at a cost of long latency, high energy usage, or both. It also introduces many ML-specific bugs into software systems.
There exist many challenges in integrating ML components into traditional software systems in an accurate, efficient, and robust way. In this proposal, we are tackling this challenge one three different levels. We will also talk about our ongoing effort about optimizing ML software behaviors with run-time diagnose.
To improve the flexibility of neural networks, we have proposed a novel neural network architecture and a customized optimizer that support anytime prediction. This design allows one neural network to generate a series of increasingly accurate outputs over time without sacrificing accuracy for flexibility. To tackle system level deployment problems, we have designed a runtime scheduler ALERT, which holistically configures neural networks and system resources together to meet application-specific accuracy, performance, and energy-consumption constraints. To understand the problems in ML software, we conducted the first comprehensive study about how real-world applications are using machine learning APIs. Guided by this study, we proposed Keeper, a new testing framework for software systems that use machine learning APIs. It automatically generates test inputs to thoroughly test whether machine learning APIs are properly used in software.

Advisors: Shan Lu

Committee Members: Hank Hoffmann, Michael Maire, and Shan Lu