Egor Danilov

In this position, I worked as a researcher in neural network-assisted inference and interpolation stochastic processes. Specifically, I worked with irregularly sampled time series modelled as Gaussian processes.

In the course of the work, I introduced a Neural-network based method to the inference of the power spectrum of irregularly sampled time series. This novelty significantly improved performance both in inference and interpolation. This work was published in one of the most famous Neural networks conferences, NeurIPS 2022, as well as at conferences of two world-known physics institutions: Fermilab and CERN.

Additionally, I participated in the research activity of the DeepSkies lab. I helped people with technical questions and paper writing. Moreover, I assisted interns of bachelor's and master's levels with their machine learning research.

In this position, I worked as a researcher in stochastic optimization. The problem was to infer a power spectrum of a non-linearly transformed 2d Gaussian process.

During the work, I developed a Jax-based auto-differentiable code for the data simulation. Then, I introduced a gradient descent-based pipeline for optimizing the parameters of the power spectrum of the transformed Gaussian process. At the moment, the results of my work are used in the studies of the recent JWST telescope images.

In this position, I worked as a researcher in applying generative neural networks to computer vision. The problem was to make an interpretable neural network-based model of an image. Given this model one could optimize its interpretable parameters for a given image and draw conclusions about the image from these parameters.

During the work, I developed a Variational autoencoder, that managed to resolve the main generative properties of single-color galaxy images. The results of my work were used to kick-start a Phd research in neural-network-based image deconvolution.

In this position, I worked as a backend software developer in cryptocurrency trading. I developed a web interface for trading on the Bitmex exchange in Haskell. Additionally, I helped with minor tasks like a consultation about the company site.