C S 370

Candidate Topics

• The source code for Twitter's ranking algorithm was released on GitHub. What (if anything) can we learn by analyzing the code?
• How can we use a large language model to redline a legal contract? How can we use it to solve cryptic crossword puzzles and explain its reasoning?
• Canvas records timestamped keystroke traces for students taking a quiz. Visualize such traces as trajectories in space-time, where "space" refers to the questions in the quiz. Using anonymized collections of such traces, derive statistical analytics of pedagogical interest.
• Use the Manim animation engine to programmatically create animations of pedagogical relevance to C S 429.
• Use the Python librosa library to create a large data set of North Indian classical music, tagged by rāgas. Design and train a machine-learning model to characterize performance fragments.
• The complete mapping of the connectome of an insect brain was a milestone event in neuroscience. What can computer architects learn from this data?
• DNA is often referred to as "the code of life." How literally can this metaphor be taken? What structural and behavioral ideas from binary computer programs have analogs in DNA, and vice versa?
• Investigate the tradeoffs between memory locality and numerical convergence for algorithms that train neural networks.
• Computer architectures have evolved dramatically over the last decade. How has this affected performance results obtained in the past? Take some performance-critical operation, measure it on current-generation machines, and explain how and why these results differ (or not) from earlier results in the literature. This can be a starting point for a much more extensive research project.
• Perform a literature survey of some selected area of (non-quantum) computer architecture and write an annotated bibliography.
• Prove or disprove: every even natural number greater than 2 is the sum of two prime numbers.
• BYOT: Bring Your Own Topic.