BrainWaves Newsletter 011

Curated insights and tools for curious minds.

📅 Date: 21 April 2025

⏱️Read time: 5 minutes

✨ Resource: Reinforcement Learning-Based Oscillation Dampening: Scaling Up Single-Agent Reinforcement Learning Algorithms to a 100-Autonomous-Vehicle Highway Field Operational Test
💡 Why it matters: This work demonstrates the real-world scalability of reinforcement learning for traffic flow control by deploying a single-agent RL algorithm across 100 autonomous vehicles without inter-agent communication. It shows how RL can reduce traffic oscillations and improve safety and reliability in highway systems.

💬 Vision Fine-tuning on GPT-4o for Visual Question Answering: Image classification through convolutional neural networks are long-time solved abundant in the literature. The novelty of this generative algorithm is that it can learn from visuals such as book cover images and answer prompt questions such as ‘what is the title of this book’ or ‘what genre is this book’. It turned out that fine-tuning yielded higher precision and better generalization predictions compared to non-fine-tuned GPT-4o. This particular algorithm is for smarter defect detection models. Full code is given.

📣 X Developer : If you are looking for a data to train your LLM, it worths to check X (formerly Twitter) Developer program. It offers API access to read and write posts on X. It makes me wonder if they count automated posts same as the genuine human ones as rate changes for number of posts you like to write and read and the rates are certainly not for everyone. Filtered stream API lets you to set a rule and select the preferred topics. Flip-side, it is a great opportunity to access real-time, comprehensive and organised datasets.

🔧 𝗙𝗥𝗘𝗘 AI YouTube channels: Andreas Horn shared list of ai channels including two-minute paper recaps, how ai mathematics work in background, interview podcasts, tutorials and more. Golden resources.