LambdaZero – Exascale Search of Molecules: Multiple Job Openings
Science:
Graph Learning Fellowship
Responsibilities:
- Participate in brainstorms; Propose solutions to the current limitation of graph-learning methods on protein-protein and drug graphs; in particular the information bottleneck
- Implement solution in PyTorch, and torch geometric
- Collaborate with Will Hamilton, Jian Tang, Michael Bronstein and Yoshua Bengio and students
About you, preferred qualifications:
- 1-2 years experience with pytorch
- Experience with TorchGeometric or other graph learning library
- PhD student or recent graduate
Semi-supervised Learning Fellowship
Responsibilities:
- Participate in brainstorms; Propose solutions to the current limitation of representation-learning methods for gene-expression data. Some of the important problems are hidden confounder variables
- Implement solution in PyTorch
- Collaborate with Will Hamilton, Jian Tang, Brooks Paige, Charlie Roberts, and Yoshua Bengio and students
About you, preferred qualifications:
- 1-2 years experience with pytorch
- Interest in proteins/biology is highly valued
- Experience with Torch and implementation of VAE, GANs
- Interest in causal models
- PhD student or recent graduate
Bioinformatics Fellowship
Responsibilities:
- Participate in brainstorms; provide bioinformatics expertise to the discussion. Screen relevant datasets. Propose solutions to combine different data sources in a biological sensible way
- Implement solution with python, pandas, numpy
- Closely collaborate with Relation Therapeutics
About you, preferred qualifications:
- 1-2 years experience with python and pandas
- Experience with Gene expression data, drug synergy scores, Protein Interaction Graphs
- Some basic understanding of Machine Learning is highly valued
- PhD student or recent graduate
Reinforcement Learning Fellowship
Responsibilities:
- Participate in brainstorms; Identify limitations of model-based RL algorithms when applied to discrete optimization problems and propose solutions
- Learn/understand baseline methods in non-convex discrete optimization during the project
- Learn/understand common graph neural networks (GNN)
- Implement solutions in code
- Collaborate with Doina Precup, Pierre-Luc Bacon, and Yoshua Bengio
About you, preferred qualifications:
- Previous experience 1-2 years in RL
- Familiarity with Ray-RlLib
- PhD student or recent graduate
Uncertainty Learning Fellowship
Responsibilities:
- Design/optimize techniques to acquire data from oracles of different fidelities; oracles estimate energy of molecules; good oracles could cost 7 orders of magnitude more compared to cheap oracles
- Be willing to learn/understand biophysics & biological assays we perform
- Participate in discussions propose and implement new ideas
- Collaborate with Brooks Paige, Jose Miguel Hernandez-Lobato, Yoshua Bengio
About you: preferred qualifications
- Familiarity with models of uncertainty IE: ensemble of models, Bayesian Ridge Regression, MC Dropout, etc.
- Familiarity with acquisition functions Batch-BALD, UCT, Thompson Sampling, Max Value Entropy Search, etc.
- PhD student or recent graduate
Mathematics Fellowship
Responsibilities
- Support mathematics requirements across a wide range of LambdaZero projects. Sometimes it could be deriving the loss function; sometimes it could be understanding the gradient explosion and deriving the normalization coefficients; sometimes it could be about if a certain kind of MPNN could solve 1-WL test.
- Collaborate with all professors and student on the project
About you, preferred qualifications:
- Undergraduate degree in mathematics, statistics, or other relevant field.
- Emphasis on academic excellence (solid GPA and/or published works)
- Critical thinking ability that could allow us to switch between parts of the project and quickly understand the problem and clearly communicate the solution.
- Ability to code examples in python/jupyter-notebook
- PhD student or recent graduate
Engineering:
Distributed Computing Fellowship
Responsibilities
- Develop and optimize computational speed/performance of large-scale ML models executed in parallel on multiple GPUs including: model-based RL algorithms, hyperparameter search algorithms, bayesian acquisition function on large datasets
- Further develop our our automated hyperparameter search techniques
- Document your code rigorously
- Review pull requests
About you, preferred qualifications:
- Experience with multi-GPU models
- Experience with Project Ray
- Practical experience in Facebook/Deepmind/etc/AI startup
Cloud Computing Fellowship
Responsibilities
- Continue the development of our existing auto scaling AWS cluster API to support requests from RL algorithms running in real time on another server
- Develop distributed S3 instance data storage API
About you, preferred qualifications:
- Experience with AWS/Azure or other cloud environment
Algorithms Fellowship
Responsibilities:
- Implement CUDA kernels, in particular for graphs
- Solve algorithmic problems such as fast shortest path on graph, or implement policy-based diffusion on graphs
About you, preferred qualifications
- Past contributions to C++ code to one of the graph libraries such as torch geometric or similar
Operations:
Business Development Management Fellowship
Responsibilities:
- Identify partnership opportunities
- Develop relationships with pharmaceutical companies. Attend scientific conferences and events. Prepare, modify according to feedback, and close collaboration agreements.
- Ability to sometimes work on weekends
Preferred Qualifications
- An undergraduate/graduate degree from a top-tier university could be a plus
Apply
The initial duration of a fellowship is 1 semester or 4 months. We might be able to offer a limited number of 1-year extensions depending on success of the project and personal contributions.
We maximize the value of our team as a whole rather than per individual. Therefore we consciously/actively seek for team additions from diverse backgrounds.