July 18, 2020
BigHat founder Peyton Greenside and BigHat consultant Vivek Myers presented their work A Hierarchical Approach to Scaling Batch Active Search Over Structured Data at the ICML Workshop on Real World Experiment Design and Active Learning. The method enables efficient, large-batch search over a structured set of protein or DNA sequences by combining deep representation learning with a hierarchical search strategy. It balances broad exploration over unobserved sequence space with more greedy local search where promising sequences have been identified. These innovations were possible by taking advantage of the natural emergent structure in a large corpus of related biological sequences to help guide experimental design in a high-throughput setting.