Extropic’s hardware massively accelerates certain kinds of probabilistic inference. Our ML team works on the science of training models in the thermodynamic paradigm, and we are looking for senior research and engineering talent to derive probabilistic ML theory, empirically demonstrate its scaling properties, and deploy performant models. Senior hires will be leading their own research direction and are therefore expected to quickly become experts across our abstraction stack, including the hardware, software, physics, and math.

Responsibilities

Collaborate with senior researchers, residents, engineers, and physicists to derive the theory of new probabilistic models and their learning rules, including energy-based models and diffusion models

Scale up experimentation infrastructure and optimize over the design space of models

Implement, visualize, and evaluate new architectures, training algorithms, and benchmarks

Publish papers, contribute to open source, and communicate design insights to our hardware team

Create production models for domain experts using customer data

Required Qualifications

Experience in scientific Python and at least one deep learning framework (PyTorch, JAX, TensorFlow, Keras)

Extremely strong foundations in probability and linear algebra

Familiarity with deep learning theory and literature, including theory of over-parameterization and scaling laws

Publications in top ML conferences (NeurIPS, ICML, ICLR, CVPR)

Experience training high-performance models, including familiarity with infrastructure (Slurm, Ray, Weights & Biases)

Experience deploying models, including familiarity with infrastructure (Ray, AWS, ONNX)

Preferred Qualifications

Experience designing probabilistic graphical models (PGM)

Experience training energy-based models (EBMs) or diffusion models

Experience with numerical methods in diffeq solvers

Experience with message passing or training graph neural networks (GNNs)

Strong theoretical background in information geometry

Strong theoretical background in random matrix theory

Strong grasp of computational Bayesian methods, including MCMC sampling methods and variational inference

This position will require access to information subject to control under U.S. export control laws and regulations, including the Export Administration Regulations (“EAR”). Please note that any offer for employment will be conditioned on authorization to receive controlled items.