Actively Learning Costly Reward Functions for Reinforcement Learning

Transfer of recent advances in deep reinforcement learning to real-world applications is hindered by high data demands and thus low efficiency and scalability. Through independent improvements of components such as replay buffers or more stable learning algorithms, and through massively distributed systems, training time could be reduced from several days to several hours for standard benchmark tasks. However, while rewards in simulated environments are well-defined and easy to compute, reward evaluation becomes the bottleneck in many real-world environments, e.g., in molecular optimization tasks, where computationally demanding simulations or even experiments are required to evaluate states and to quantify rewards. Therefore, training might become prohibitively expensive without an extensive amount of computational resources and time. We propose to alleviate this problem by replacing costly ground-truth rewards with rewards modeled by neural networks, counteracting non-stationarity of state and reward distributions during training with an active learning component. We demonstrate that using our proposed ACRL method (Actively learning Costly rewards for Reinforcement Learning), it is possible to train agents in complex real-world environments orders of magnitudes faster. By enabling the application of reinforcement learning methods to new domains, we show that we can find interesting and non-trivial solutions to real-world optimization problems in chemistry, materials science and engineering.

Machine learning for rapid discovery of laminar flow channel wall modifications that enhance heat transfer

The calculation of heat transfer in fluid flow in simple flat channels is a relatively easy task for various simulations methods. However, once the channel geometry becomes more complex, numerical simulations become a bottleneck in optimizing wall geometries. We present a combination of accurate numerical simulations of arbitrary, non-flat channels and machine learning models predicting drag coefficient and Stanton number. We show that convolutional neural networks can accurately predict the target properties at a fraction of the time of numerical simulations. We use the CNN models in a virtual high-throughput screening approach to explore a large number of possible, randomly generated wall architectures. We find that S-shaped channel geometries are Pareto-optimal, a result which seems intuitive, but was not obvious before analysing the data. The general approach is not only applicable to simple flow setups as presented here, but can be extended to more complex tasks, such as multiphase or even reactive unit operations in chemical engineering.