About

Cody Wild is a Machine Learning Research Engineer at Google with over 7 years of experience in the field. He is a highly skilled professional with both production and research experience, and a persistent desire to understand new things and take on new challenges. Recently, he joined the Humans & Interaction group within Google Research. At Google, Cody is currently working on the Representation Learning for Imitation project, where he led the design of a modular PyTorch framework that facilitates easy implementation of different variants of (mostly contrastive) representation learning algorithms with minimal need for code duplication. He also co-led the implementation of the framework's initial complement of 10-12 algorithms, including CPC, SimCLR, and CEB. Additionally, he collaborated on experiment design for benchmarking these algorithms as pretraining for imitation. Cody is also a co-organizer of the 2021 BASALT NeurIPS Competition, where he collaborated on many areas of the competition, including training imitation learning baselines and designing user-friendly Mechanical Turk evaluation scripts to release for competitors. Before joining Google, Cody worked at the Center for Human-Compatible AI as a Machine Learning Research Engineer. There, he completed a multiprocessing-heavy refactor of an implementation of Deep Learning from Human Preferences and generalized the code to work on arbitrary Gym environments and use a more intuitive Wrapper design. Additionally, Cody led the design and testing of a modified adversarial training mechanism as a hardening strategy for trained policies, which led to co-authorship on ICLR 2020 paper "Adversarial Policies: Attacking Deep Reinforcement Learning." Cody's previous experience also includes working as a Senior Data Scientist at Sophos, where he authored and gave an accepted talk at CAMLIS on a novel clustered loss function used to integrate asymmetric costs of misclassification in a multi-class setting. He also engaged in experimentation to improve a business-critical, production model and was a proponent of two of the changes driving the highest performance boost, with a particular focus on the productive use of auxiliary loss functions to guide and improve learning. Cody holds a Master of Science degree in Analytics from the University of San Francisco and a Bachelor of Science degree in Mathematical Economics & Middle Eastern Studies from Tulane University. He has a strong technical background in Research Scientist, ML, Software Engineer, Deep Learning, Reinforcement Learning, Pytorch, Python, Data Scientist, and Neural Networks.