From Code Generation Towards Software Engineering: Advancing Code Intelligence with Language Models

Overview

Explore how large language models can advance beyond code generation to tackle complex software engineering tasks through enhanced symbolic and global reasoning capabilities.

Syllabus

- Introduction to Code Intelligence -- Overview of Language Models in AI -- Historical Context: From Rule-based Systems to Neural Networks - Fundamentals of Code Generation with Language Models -- Basics of Transformer Architecture -- Pre-training and Fine-tuning for Code Completion -- Limitations of Current Code Generation Techniques - Symbolic Reasoning in Language Models -- Introduction to Symbolic AI Concepts -- Integrating Symbolic Reasoning with Neural Models -- Case Studies: Symbolic Enhancement in Code Analysis - Global Reasoning for Software Engineering Tasks -- Understanding Software Engineering Complexity -- Role of Global Context in Code Comprehension -- Techniques for Capturing Global Code Structure - Advanced Applications in Software Engineering -- Automated Debugging and Error Correction -- Assistive Models for Software Design and Architecture -- Enhancing Model Performance on Legacy Codebases - Challenges and Future Directions -- Ethical Considerations in AI-driven Software Development -- Limitations of Current AI Techniques in SE -- Future Prospects: Unifying Symbolic and Neural Approaches - Practical Workshop: Building and Deploying a Code Intelligence Model -- Tools and Frameworks: PyTorch, TensorFlow, and Hugging Face -- Hands-on Project: Developing a Code Suggestion System - Conclusion and Next Steps -- Recap of Key Learnings -- Directions for Continued Learning and Research Opportunities

Taught by

Tags