ML Researcher - Differential Equations
Geometric machine learning | Graphs | Topology | Geometic modeling | Neural DEs | Biological systems | Scale to 100s of millions of parameters | OSS Contribution | Collaboration with leaders in the field
Fully remote: We hire the right people wherever they are | Top quartile salary | Stock options | 401K | Health, dental and vision insurance | OSS contribution | Collaboration with leaders in the field
Geometric machine learning provides a unique mechanism for understanding the dynamics of biological systems through the lens of dynamical systems theory. Syntensor maintains a single backbone model which is used to capture the dynamics and state-space of a very large-scale representation (2m nodes per cell type for ~2k cell types) of human physiology, and neural Differential Equation-based approaches (in combination with Graph Neural Networks) are at the core of this effort.
Join a team of world-leading experts in this domain and help drive the field forward as we tackle a problem space only recently made tractable by advances in ML and HPC. Work on edge-of-chaos dynamics, multiple-attractor states, stability and control mechanics, and more, all at large (computational) scale with models scaling to 100s of millions of parameters.
This role offers the potential for extraordinary impact in a well-established market/economy for deep learning applications.
- Find and implement recently published methods with the goal of inventing new architectures and methods for dynamical systems simulation in biology. To do this you will develop methodology and experiments to validate implementations and possible improvements by testing on smaller, academic problems.
- Get full support with publishing results (using toy/academic datasets) if effective methods are discovered.
- OSS contributions - Improve the Torchdyn library by integrating existing methods such as those from Julia’s DE Packages, Diffrax and TorchDiffEq and fill in missing gaps in the DiffEqML ecosystem.
- Collaborate with DiffEqML, MILA, UM, Stanford and other academic teams and get exposure to leading methods in deep learning and HPC.
- Competitive compensation: We offer a starting salary in the top quartile for role and level, based on local benchmarks.
- Stock options: You are joining an early stage startup. We want you to have ‘skin in the game’ and your options package will reflect this.
- 401K for US-based team members.
- Your health, dental and vision insurance will be fully covered, with partial coverage for family/partners as well.
- Self-manage: Enjoy the flexibility and opportunities that come with early stage startups. You’ll work closely with all areas of the business with a high degree of autonomy and the opportunity to progress as an individual contributor or manager. Syntensor is a distributed-first company, working from home and collaborating asynchronously from the East and West Coast US, Canada and Europe.
- Take on a grand challenge and share our purposeful mission: We’re building a biological systems simulator that will transform how medicines are developed and prescribed. You will be advancing the field while making a big impact on the wellbeing of millions of patients.
- 3 years experience at graduate+ level in Differential Equations (coursework/research in numerical methods, Ordinary Differential Equations, Partial Differential Equations, Nonlinear Dynamical Systems)
- Strong theoretical understanding of Dynamical Systems
- Publications in computational biology/physics/mathematics/chemistry or publications in ML
- Familiar with graphs, topology, geometric modeling
Nice to haves
- Open Source Software contributor
- Publication(s) in applications of Deep Learning towards computational biology/physics/mathematics/chemistry
- Successfully invented and implemented a unique use-case for applying ML + differential equations
- Experience working within the domain of biological systems
- PyTorch / Julia / JAX User
- Publications at a premier CS conference e.g. ICML, NeurIPS, ICLR, etc.
- Experience with early-stage startups
We are on a mission to provide access to more effective medicines for millions of patients. We’re building a model of human molecular physiology for research scientists and clinicians that can answer the fundamental question, "will it work?"
Every modernized field of engineering has a systems simulator to test complex interactions in bits rather than atoms. This doesn’t yet exist for biology. Without one, drug development is expensive because risk of failure is very high; 30-60% of prescribed medicines have no clinical benefit to patients and adverse reaction to treatment is the 4th leading cause of death in the US, ahead of pulmonary disease and diabetes. Syntensor is taking on this grand challenge, developing fundamental machine learning methods and applying them at scale to biological data so every individual patient can be prescribed the most effective, least toxic treatment possible.
What we do
We are productionizing and scaling up a generalizable machine learning platform that predicts efficacy and toxicity for any drug in any indication. We are using an extensive, heterogeneous biomedical graph, novel fundamental ML methods and advanced engineering infrastructure to generate and explain model outputs for users of our app. Currently, our users are research scientists involved in drug development.
We are a small team of people with diverse skills and a shared bias towards problem solving and execution. We are inventors and builders who believe in the scientific method; feedback and iteration is essential to our process and we share our work early and often. That said, we aim high. Our mission and the domain in which we operate demand that we take on some of the hardest problems researchers, scientists, engineers and designers face, and we are determined to build technology that solves them properly and usefully for users of our platform. We are looking for talented people who are motivated by the challenge of hard problems and who are already curious about the technological, scientific or cultural domains with which we engage.
We are a distributed-first team and very relaxed about where and when work happens, but come together as a whole team weekly to sync-up. We work with intrinsic curiosity and motivation towards well defined goals (even where there are unknown unknowns). Our diversity, great communication and respectful, supportive teamwork make us highly effective.