Pinned Loading

1. Literary-LLM-Knowledge-Data-Poisoning Literary-LLM-Knowledge-Data-Poisoning Public

   Data poisoning attacks on LLMs — corrupting Harry Potter knowledge via Tolkien-style fine-tuning, with quantitative analysis of knowledge degradation (Reichman University, 2025)

   Jupyter Notebook 1

2. FlappyBird-RL-Agent FlappyBird-RL-Agent Public

   Q-Learning vs SARSA for Flappy Bird — tabular RL with custom reward shaping, state space reduction, and cross-difficulty evaluation (Reichman University, 2025)

   Jupyter Notebook 1

3. RL-MiniGrid-PPO-Agent RL-MiniGrid-PPO-Agent Public

   PPO vs Dueling Double DQN for MiniGrid MultiRoom navigation — curriculum learning across 2/4/6-room environments with custom reward shaping (Reichman University, 2025)

   Jupyter Notebook 1

4. llm-hallucination-sketches llm-hallucination-sketches Public

   Streaming LLM hallucination detection using 5 sketch data structures (CMS, HLL++, AMS, Bloom, MinHash) — F1=0.77 in 256 bytes of memory (Reichman University, 2026)

   Jupyter Notebook

5. Multimodal-Harmful-Content-Detection-Social-Media-Memes- Multimodal-Harmful-Content-Detection-Social-Media-Memes- Public

   Multimodal harmful meme detection combining CLIP vision-language embeddings with OCR text extraction and neural classifiers (Reichman University, 2025)

   Jupyter Notebook 1

6. LunarLander-DQN-vs-PPO LunarLander-DQN-vs-PPO Public

   Reinforcement Learning: DQN vs PPO on LunarLander-v3 — built from scratch with PyTorch & Gymnasium

   Jupyter Notebook