CV
Igor Sadalski
Summary
Machine Learning Scientist at Somite.ai. Working on multi-modal foundational models for life sciences.
Education
- MSc Computer Science (Machine Learning)2024-09Imperial College London
Courses: Deep Learning, Reinforcement Learning, Graph-Based Learning, Software Engineering for ML Systems, Natural Language Processing - BEng Mechanical Engineering (Robotics)2023-06Imperial College London
Work Experience
- Machine Learning Scientist2024-09 -Somite.ai
Orchestrated large-scale distributed training and published papers working on multimodal foundational models for life sciences.
Research Experience
- Research Assistant2024-04 - 2024-09Harvard University (PI: Prof. Stephanie Gil)
Investigated techniques for performing sub-quadratic attention; based on this work, designed a long-context time series transformer-based model. The model was later used to proactively route shuttle buses around Harvard campus. - Research Assistant2023-06 - 2023-08Caltech (PI: Prof. Aaron Ames)
Designed conditional variational autoencoders and diffusion models to obtain data-driven error estimates from flight data. These models were later used to improve drone flight performance and safety. - Research Assistant2022-06 - 2022-08Caltech (PI: Prof. Aaron Ames)Sped up model predictive controller running on robotic hardware by rewriting matrix operations using optimized C++ library.
Publications
- Measurement noise scaling laws for cellular representation learningNature Biotechnology (in preparation)Co-first author with Gokul Gowri. In preparation for Nature Biotechnology.
- Encoding Strategies for Single-Cell Foundational Models: A Large-Scale Benchmark of Gene and Expression Tokenization
- Scaling up Measurement Noise Scaling Laws2025ICML Workshop on Multi-modal Foundational Models for Life Sciences '25Presented at ICML Workshop on Multi-modal Foundational Models for Life Sciences '25.
- Generative Modelling of Residuals for Real-Time Risk-Sensitive Safety with Discrete-Time Control Barrier Functions2024