AI in the Physical World: Reinforcement Learning for Industrial Design and Optimisation

DeepAI London II An evening on reinforcement learning, industrial design, and the optimisation of physical systems. Most industrial systems still run on heuristics, legacy control logic, and the judgement of experienced operators. That approach works, but it leaves a lot on the table. Robotics, logistics, plant control, process scheduling, aerodynamic and structural design, energy management and materials selection all involve search spaces far larger than any engineering team can work through by hand. Reinforcement learning arrived at this problem from an unlikely direction. Ten years ago its best known results were in games. An agent learned to play Atari from raw pixels, and then AlphaGo beat the strongest Go players in the world. Then in 2016, DeepMind pointed the same methods at cooling in Google's data centres and cut the energy used for cooling by around 40 per cent. By 2018 the system was controlling live equipment directly. Games had become industrial control. Since then RL has been turning up in tokamak plasma control, chip floorplanning, turbine and grid operation, component design, and process control in factories and fabs. How well do these methods generalise once they leave simulation? What does it take to trust a learned policy with capital equipment and safety critical plant? Where does simulation stop being good enough and real world data have to take over? What direction might continual learning take? And which parts of industrial design and operation are next? These are some of the questions our panellists will debate. Join us at Shoreditch Treehouse for an evening of talks and debate with researchers and operators applying RL across a variety of industrial settings, followed by a panel debate, and drinks with the industrial RL community. Following our previous events hosted with Amazon, this is the third event in the DeepAI series. Agenda 6:30 – 7:00 — Doors open, arrivals & refreshments 7:00 – 7:20 — Guest talk I (speaker to be announced) 7:20 – 7:40 — Guest talk II (speaker to be announced) 7:40 – 8:00 — Guest talk III (speaker to be announced) 8:00 – 8:30 — Panel discussion 8:30 – 9:30 — Social hour Who this is for Researchers and engineers in reinforcement learning, foundation model training, and deep neural networks; anyone using RL in the real world. Venue Shoreditch Treehouse, London NOTE: No stiletto heels please!