I have worked in Skoltech since 2017, starting as a junior research scientist and holding an assistant professor position now. My lab focuses on deep learning models for sequential data and robust machine learning. I am curious about engineering applications and fundamental questions in deep learning, looking for interesting problems that can lay a new foundation in deep learning and other sciences that benefit from it. These developments result in top-level publications, impactful industrial projects, empowering courses, and strong student theses. While the text below provides a nice overview of my achievements and research interests, you may also visit my lab’s website.
My lab focuses on developing new models for sequential data and the robustness of deep learning models. This research is tightly connected to probabilistic machine learning and representation learning for various data modalities. Our joint publications with leading companies like Sberbank and Saudi Aramco pursue the desire to develop solutions that revolutionize the industry. On the other hand, we continue to publish at top deep learning venues, including KDD, AISTATS, and ACMM, and in top journals like IEEE Geoscience and Remote Sensing Letters and Journal of Petroleum Science and Engineering.
Industrial projects and further education
During the last few years, I have completed numerous industrial projects that earned tens of millions of dollars for significant banks, oil and gas companies, and telecom. As a lead in these projects, I collaborated with the industry on developing specifically tailored technologies based on machine learning and deep learning: production models are accurate and efficient. Their performance lasts long and delivers benefits to companies all this time. Another focus of my work is teaching others to work in this way. I developed modern courses for students and industrial partners at all levels, from engineers and experienced data scientists to top managers.
My alma mater was the Department of Control and Applied Mathematics of MIPT, which I completed in 2012. My master’s thesis proposed the Bayesian approach for linear regression that allows an automated feature selection. In my Ph.D. thesis, I obtained a new result on the effectiveness of Bayesian procedures for Gaussian process regression. These developments laid a theoretical justification for the selection of the design of experiments for variable fidelity models and proved minimax errors for Gaussian process regression. The publication venues for these results were several top journals and conferences, including the AISTATS conference. Working in a start-up, DATADVANCE, during my PhD, I took part in developing the pSeven library dedicated to data analysis for engineers. The pioneering tool for data fusion solved a regression problem for the case of data with more than one fidelity. Other applications of my work are now works in companies such as AREVA, TOTAL, and Airbus. In 2018, these results were awarded the Moscow government award for young scientists, “Development of predictive analytics methods for processing industrial, biomedical and economic data,” in joint with E. Burnaev and M. Panov.
Banks and related companies:
In 2018 Alexey was awarded the Moscow government award for young scientists “Development of predictive analytics methods for processing industrial, biomedical and economic data” joint with E. Burnaev and M. Panov.
Present students (with the year of the planned defense)