ML Software Engineer
Berkeley Existential Risk Initiative
Amherst, MA, United States
Job TitleML Software Engineer
- Start date: January, 2022 (flexible)
- Length of engagement: One year, with the possibility of extension
- Time commitment: Full time, ~40 hours/week
- Compensation: Based on experience; $140,000 - $210,000
- Work location: Remote. Preference may be given to candidates who can meet in person in Amherst, MA once/week.
BERI is seeking a full-time machine learning software engineer as part of our collaboration with the Autonomous Learning Laboratory at UMass Amherst. As the lead developer on the project, this person will have significant independence and autonomy.
We’ve secured funding for this position for at least one year, which we believe will be enough time to implement the core functionality. Extensions will depend on our ability to secure more funding during the first year of work.
=== PURPOSE ===
Prof. Philip Thomas and his collaborators have designed a general and flexible framework for a new kind of machine learning algorithm called a Seldonian algorithm, which clarifies the three-way relationship between ML system designers, users, and the ML systems themselves. In particular, the Seldonian framework simplifies the problem of specifying and regulating undesirable behavior by involving end-users in the specification and oversight of system safety parameters. A high-level description of the Seldonian framework can be found here; the full Science paper can be found here.
We want to hire a software developer to create a software library that makes it easier for researchers and practitioners to apply and create Seldonian algorithms. This library would facilitate and advance academic research on safe machine learning and would provide a practical tool for corporations to responsibly apply machine learning to high-risk high-reward applications. The end goal is to get more safety constraints built into ML systems.
=== RESPONSIBILITIES ===
- Plan, design, and code a software library (or set of related libraries) that can be used by other software developers to more easily implement Seldonian algorithms in their projects.
- Deeply familiarize yourself with the existing literature on Seldonian algorithms
- Collaborate closely with Prof. Thomas and his collaborators on all aspects of the development process, from the discussion of high-level strategic goals (e.g. what programming language to use) to the design and implementation of individual library features
=== BENEFITS ===
- Become a world-leading expert on the practical implementation of Seldonian algorithms, encouraging the development and deployment of safer ML systems.
- More independence and autonomy than you’d have at a big tech company
- Time-off (paid vacation, holidays, sick, bereavement, and parental leave)
- Platinum health, dental, and vision insurance (98% of employee premiums covered by BERI)
- Life and short-term/long-term disability insurances (100% of employee premiums covered by BERI)
- Very flexible work schedule including hours, location, and unpaid vacation policies
- Telecommuting is OK
- No Agencies Please
- 3-7 years of relevant experience
- Deep knowledge of Python or C++
- Broad familiarity with commonly-used machine learning algorithms
- Experience leading and owning a software development project from beginning to end
- Excellent technical communication skills, the ability to elaborate complex technical concepts and collaborate effectively with fellow engineers and researchers
- Prior research or research engineering experience
If you feel you do not meet these qualifications but are still interested in applying, please err on the side of applying anyway.
About the Company
BERI is an independent 501(c)(3) public charity. Our mission is to improve human civilization’s long-term prospects for survival and flourishing. Currently, our main strategy is collaborating with university research groups working to reduce existential risk (“x-risk”), by providing them with free services and support.
The Autonomous Learning Laboratory (ALL) conducts foundational artificial intelligence (AI) research, with emphases on AI safety and reinforcement learning (RL), and particularly the intersection of these two areas.