Reinforcement Learning, Spring 2026, Homework 1
Instructor: Prof. Iddo Drori

1. Supervised learning

Sign up for the free gemini-3 pro plan for students.

a. Generate pairs: Choose 5 papers from alphaxiv.org and put each one in a separate notebooklm.google.com project.
In each project generate an infographic, slide deck, and data table.

b. Prompt: Upload the 5 pairs of (paper, infographic) into gemini-3-pro-preview and ask it to generate a prompt that would generate each infographic from each paper using gemini. Repeat for the (paper, slide deck) and (paper, data table) pairs, to generate three base prompts: paper2inforgraphic, paper2slides, and paper2table.

c. Test on unseen example: Chose another paper and upload it to gemini-3-pro-preview, and use each of the three base prompts to directly generate an infographic, slide deck, and data table for the paper. Modify the base prompts to generate results in your own style.


2. Multi-armed bandits

a. Install, build, and run Claude Code with plugins:
curl -fsSL https://claude.ai/install.sh | bash
claude update
claude
/plugin install frontend-design@claude-plugins-official
/plugin install superpowers@claude-plugins-official
/plugin install code-simplifier@claude-plugins-official
..
Install the chrome plugin.
/init

b. Prompt Claude Code to implement and test four independent Bernoulli multi-armed bandit algorithms in Python: greedy, epsilon-greedy, upper-confidence bound (UCB), Thompson posterior sampling. 

c. Prompt Claude Code to build a webapp using the frontend-design plugin that is an interactive simulation which allows a user to compete with the four multi-armed bandit algorithms it implemented and tested in multiple rounds and visualize the results, and use the superpowers and code-simplifier plugins to generate and clean the code.

Submission:
Submit all deliverables (data, code, prompts, and app) on the shared class Google Drive
under the subdirectory with your name, e.g. Homeworks/1/Your Name/
Put everything for this homework inside your folder.
Include a short README.md (or README.txt) at the top of your folder describing:
what you did, where each deliverable is located, how to run your code/webapp (commands, dependencies).
Make sure any notebooks/code run without needing your local paths (use relative paths where possible).
Use an LLM (Gemini or GPT) to review your work before submission.

Out: 1/26, due: 2/5, 10 points.