About me
I am interested in programs that build their own predictive models of the world and use them in decision-making. In recent years neural networks are the best tool for this. Earlier, I worked with probabilistic methods for state estimation, often in the domain of robotics.
Outside AI, I view prediction markets as a promising social technology and have built some prototypes.
Skills
programming (python, js, ts, c++, matlab)
machine learning (pytorch, tensorflow, sklearn, pandas)
big data (spark, bigquery)
probability theory, reinforcement learning
web dev (fastapi, flask, docker, react, sql, nosql)
Google Cloud Platform
Projects
Troper - (alpha) web app recommending similar movies to the one specified by the user
EstyMate.co - web app for tracking & updating personal beliefs
Predict Virus (defunct) - dashboard for exploring covid datasets and models
Pommerman - built a reinforcement learner for a tournament of multiplayer Bomberman competition at NeurIPS
Hexplode - finalist in a multiplayer game playing AI competition organised by Man AHL
Robot maze navigation - implemented state estimation, control and planning for navigation
An old (2014) AI safety presentation
Underwater robotics - participant in EuRathlon robotics competition; underwater perception, control, navigation and communication
Research
Learning to track environment state via predictive autoencoding
PhD thesis on Predictive and reactive reinforcement learning from images
MSc dissertation on quantifying informativeness of sensors in a multi-target tracking network
Sensor management with regional statistics for the PHD filter
BEng dissertation on probabilistic multi-target trackingÂ