Data science approaches to conservation decision making
Carl Boettiger, Professor
Environmental Science, Policy and Management
Applications for Fall 2024 are closed for this project.
Become familiar with the mathematical, statistical, and computational tools used in the group and learn how to apply these methods to answer questions in ecological research and conservation decision making. Emphasis on the use of deep reinforcement learning and best practices in data science software development.
Role: We will work to train and evaluate cutting edge and emerging deep reinforcement learning algorithms on challenging ecological decision making problems. The ideal outcome will be a suite of trained agents on a set of example conservation challenges, similar to existing benchmarks in reinforcement learning that rely on benchmarks based in robotics and arcade games, e.g. https://gym.openai.com
Qualifications: Familiarity with two or more of the following areas will make an applicant both more competitive and more likely to enjoy a successful and productive research experience: Introductory statistics and probability, dynamical systems/differential equations, programming (particularly programming and data analysis in R or python), familiarity with git/GitHub, courses and/or research experience in ecology.
Programming experience in python, and particularly experience with popular reinforcement learning frameworks such as OpenAI gym, tensorflow/agents or stable-baselinesv3 would be ideal.
Hours: to be negotiated
Off-Campus Research Site: All work can be conducted remotely via GitHub. During the COVID-19 pandemic there will be no need to perform any of this work on the Berkeley campus.
Related website: https://boettiger-lab.github.io/conservation-gym/
Related website: https://stable-baselines3.readthedocs.io/