ML researcher, teacher at a university
telegram: https://t.me/pozdnyakov_vitaliy
email: pozdnyakov.vitaliy@yandex.ru
7 years in IT, 3 years in research, 3 years in education
2021 – now, AIRI (Moscow, research), Research scientist
Research and development of industrial artificial intelligence methods
2020 – now, HSE (Moscow, education), Instructor
Conducting practical classes on Network Science
2020 – 2022, MADE (Moscow, education), Instructor
Conducting practical classes on Machine Learning on Graphs
2020 – 2021, LAMBDA at HSE (Moscow, research), Research intern
Research and development of industrial machine learning methods
2016 – 2019, 1С (Moscow, software), Backend developer
Design and development of the 1C:ERP system
2014 – 2016, Wikimart (Moscow, e-commerce), Backend developer
Design and development of ERP systems
2012 – 2014, Partner LLC (Lipetsk, consulting), Programmer
Customer support, helpdesk
2019 – 2021, Moscow, Higher School of Economics
Data Science, Master’s degree
2007 – 2012, Lipetsk State Pedagogical University
Information Systems and Technologies, Specialist’s degree
2022, Fault diagnosis methods for chemical processes, AIRI
We have trained transformer-based model using self-supervised learning and deep clustering. Our method outperforms existing approaches in unsupervised setting on chemical sensor data. TPR for fixed FPR 0.05 (SOTA/ours): 0.64/0.87.
2021, Probabilistic forecasting using deep generative models, HSE & BCG Gamma
We have shown that existing deep probabilistic forecasting methods can be represented as a framework on the type of neural networks and the type of generative models. We have proposed forecasting approach based on GAN and TCN that shows promising results for future research.
2021, Digital twins for production of superconductors, LAMBDA at HSE
We have designed a digital twin of the tool for producing superconductors based on deep probabilistic forecasting model DeepAR. It is used to find optimal operating modes and adjust parameters of production.
Kovalenko, A., Pozdnyakov, V. and Makarov, I., 2022. Graph Neural Networks with Trainable Adjacency Matrices for Fault Diagnosis on Multivariate Sensor Data.International Conference on Industrial Information Integration. arXiv preprint arXiv:2210.11164. arXiv preprint
Golyadkin, M., Pozdnyakov, V., Zhukov, L. and Makarov, I., 2022. SensorSCAN: Self- Supervised Learning and Deep Clustering for Fault Diagnosis in Chemical Processes. arXiv preprint arXiv:2208.08879. arXiv preprint
Graph Neural Networks, Deep Generative Models, Time Series Forecasting, Python (numpy, pandas, scikit-learn, pytorch, networkx, DGL), SQL, Git