Explore a groundbreaking approach to interactive programming assistance in this 22-minute conference talk from the ACM Symposium on User Interface Software and Technology. Dive into the development of CodeMend, an innovative system that bridges the gap between natural language and code using bimodal embedding. Learn how this tool leverages neural embedding models to help novice and intermediate programmers overcome language barriers and expertise limitations when searching for and integrating code solutions. Discover the mixed-initiative interface that translates user goals into relevant API functions and proposed code changes. Examine the system's utility and accuracy through lab and simulation studies, and gain insights into potential future directions for enhancing programming assistance tools.
CodeMend - Assisting Interactive Programming with Bimodal Embedding
Association for Computing Machinery (ACM) via YouTube
Overview
Syllabus
Intro
Programming is not so easy
What you've seen in the video
Traditional Solutions
Three Scenarios
Deep Learning is a Good Fit
How does this work?
Opening the black box
Map Code to Neural Net
Mapping Code to Neural Net
Interface Design - Inspiration
How well does it work?
Experiment 1: How well does the model work?
Experiment 1: Example Model Output
Experiment 1: Results
Experiment 2. How well does the system work?
Experiment 2 - Results
Future Directions
Summary
Taught by
ACM SIGCHI
