Start date: ASAP
Status: Full or Part-time employee, paid hourly
Location: The San Francisco Bay area
Reports to: Executive Director
Application: at the bottom of this page
As part of a collaboration with the Center for Human Compatible AI (CHAI) under UC Berkeley professor Stuart Russell, BERI is seeking to hire engineers to work on the projects listed below. If any of these projects interests you, we encourage you to apply:
Open-sourcing inverse reinforcement learning (IRL) packages currently under development at CHAI.
Developing a well-commented and modular Python implementation of Point Based Value Iteration for use by the CHAI research team.
Writing or adapting existing code to reproduce natural language processing (NLP) results from Stanford researchers, e.g.
- Naturalizing a Programming Language via Interactive Learning (Wang, Ginn, Liang, Manning; 2017)
- From Language to Programs: Bridging Reinforcement Learning and Maximum Marginal Likelihood (Guu, Pasupat, Liu, Liang; 2017)
- Learning Symmetric Collaborative Dialogue Agents with Dynamic Knowledge Graph Embeddings (He, Balakrishnan, Eric, Liang; 2017)
Developing algorithmic implementations of Negotiable Reinforcement Learning.
Measuring the combined efficacy of goal-inference algorithms (such as IRL) with planning algorithms (such as Point-Based Value Iteration) in mechanism design applications (e.g. an algorithm playing a game on behalf of two players who have different beliefs about the game)
Taking responsibility for the existence of an easily usable shared codebase for CHAI researchers. (This would be a larger project for an engineer interested in working with BERI/CHAI in the long term.)
Ability to meet in person is required, but much of the work could also be done remotely. Funds would be available to pay for any experiments that require additional resources, e.g., cloud computing, or polling human subjects via mechanical turk.
We are especially interested in applicants who can take initiative and will find ways to help our partners. This role involves figuring out what would be helpful for researchers and then doing it.
Total hours currently needed: 10 to 40 hours a week, depending on level of independence and quality of work (with more hours going to candidates with higher independence and initiative).