The TWIML AI Podcast (formerly This Week in Machine Learning & Artificial Intelligence) Deep Reinforcement Learning for Logistics at Instadeep with Karim Beguir - #302
Sep 25, 2019
Karim Beguir, Co-founder and CEO of InstaDeep, shares his journey from a small Tunisian town to leading innovations in AI for logistics. He discusses how deep reinforcement learning is revolutionizing decision-making in logistics, improving efficiency and cost-effectiveness. The conversation touches on the use of synthetic datasets for model training and the complexities of enhancing passenger experiences in ride-sharing. Karim emphasizes the significance of adaptive reward functions and the balance between learning-based and heuristic approaches to optimize outcomes.
AI Snips
Chapters
Books
Transcript
Episode notes
AlphaZero's Chess Mastery
- AlphaZero, DeepMind's algorithm, mastered chess in four hours using distributed machine learning.
- It replaced hundreds of rules with just two concepts: searching and learning, outperforming Stockfish.
Beyond Games
- AlphaZero's success demonstrates the potential of search and learning beyond game playing.
- Real-world problems like ride-sharing and bin packing can benefit from similar solutions.
Zero-Data Training
- InstaDeep uses OpenStreetMap to model city environments for ride-sharing, but training can start with zero data.
- Building a realistic simulated environment is key, enabling learning by doing, similar to AlphaZero.
