Selected works and publications

Here’s a list of previous publications and works I am most fond of and most importantly influence my work as CTO of Levenlight AI.

Placeholder

Applications of sequential learning and stochastic calculus to trading and AI

My previous works on sequential learning can be divided into 3 different categories:

  1. Sequential predictions for optimal trading execution in finance, both by analyzing fundamental technical signals, and news releases;

  2. Sequential predictions to predict human behavior either on UGC platforms like Youtube or when playing Kinect-like games through MoveNet;

  3. Sequential predictions for accelerated denoising in Veo 3 and Genie 3.

This is where there is overlap with my other passion: data generation which I have used extensively in the past and keep leveraging heavily today to improve models:

  1. Data generation to simulate human behavior on UGC platforms like YouTube;

  2. Data generation to simulate humans as seen by a camera in varied real-life conditions with systems like Kubric;

  3. Today, similar generators based on old-fashioned procedural generation augmented with modern diffusion models to produce better embodied AI.