Aleksei Rozanov

PhD student, Computer Science · University of Minnesota

I'm a PhD student in Computer Science at the University of Minnesota, advised by Prof. Vipin Kumar. I work on self-supervised learning, transfer learning, and spatiotemporal modeling for problems where supervision is sparse or expensive to obtain.

Currently I'm developing causal pretext tasks for multivariate spatiotemporal data and exploring Prior-Fitted Networks for black-box optimization in physical systems. More broadly, I'm interested in AI for weather and climate, energy systems, and sim-to-real transfer.

Interests

  • Self-supervised representation learning, transfer learning, and domain adaptation
  • Spatio-temporal and sequential modeling
  • Foundation models for heterogeneous observational data
  • Bayesian optimization

Experience

  • Research assistant, Data Mining Lab, Department of Computer Science and Engineering, University of Minnesota Minneapolis, MN · May 2025 – present

    Knowledge-guided deep learning for spatio-temporal modeling in zero- and few-shot settings; global gridded data products from multi-source spatial data; large-scale dataset pipelines, evaluation, and sensitivity analysis.

  • Research assistant, Real-time GeoInformation Systems (GEMS Informatics), University of Minnesota Saint Paul, MN · Aug 2024 – Aug 2025

    Satellite data acquisition and processing tooling, interactive environmental visualization on the web, ML for agriculture, analysis of real-time environmental sensor data.

  • Freelancer, Geo Data Science Consulting, UpWork Remote · Nov 2023 – Aug 2024

    End-to-end ML for time series, remote sensing detection, and regression; deforestation monitoring stack (Flask, SQL, Google Earth Engine) on Google Cloud; dataset work for LLM benchmarking.

  • Research assistant, Climate and Environmental Physics Laboratory, Ural Federal University Yekaterinburg, Russia · May 2021 – Aug 2024

    Carbon flux estimation with remote sensing, meteorology, and neural networks; publications and conference talks.

  • Project teaching assistant, Climatematch Academy · Computational Tools for Climate Science Remote · Jul 2023, Jul 2024

    Trained students to create Python-based climate data workflows (xarray, pandas, matplotlib).

Education

  • Ph.D. Computer Science · University of Minnesota · Minneapolis, MN · 2025 – 2028 (expected)
  • M.S. Geographic Information Science (MGIS) · University of Minnesota · Minneapolis, MN · 2024 – 2025
  • M.S. Big Data and Machine Learning · ITMO University · St. Petersburg, Russia · 2022 – 2024
  • B.S. Hydrometeorology · Ural Federal University · Yekaterinburg, Russia · 2018 – 2022

Publications

  • CarbonBench: A Global Benchmark for Upscaling of Carbon Fluxes Using Zero-Shot Learning
    Rozanov, A., Renganathan, A., Zhang, Y., & Kumar, V. · ACM SIGKDD (under review, 2026)
    arXiv
  • Task Aware Modulation using Representation Learning for Upscaling of Terrestrial Carbon Fluxes
    Rozanov, A., Renganathan, A., & Kumar, V. · AAAI 2026 Bridge Program on Knowledge-Guided Machine Learning
    arXiv
  • Knowledge-Guided Machine Learning Models to Upscale Evapotranspiration in the US Midwest
    Rozanov, A., Subedi, S., Sharma, V., & Runck, B. C. · arXiv preprint · 2025
    arXiv
  • All-Weather Drone Vision: Passive SWIR Imaging in Fog and Rain
    Bessonov, A., Rozanov, A., et al. · Drones (2025), 9(8), 553
    Paper
  • Estimates of Carbon Dioxide Flux into the Forest Ecosystem Based on Results of Ground-Based Hyperspectral Sounding of the Atmosphere and an Artificial Neural Network Model
    Rozanov, A. P., et al. · Atmospheric and Oceanic Optics (2024), 37(2), 199–204
    Paper
  • A Neural Network Model for Estimating Carbon Fluxes in Forest Ecosystems from Remote Sensing Data
    Rozanov, A. P., & Gribanov, K. G. · Atmospheric and Oceanic Optics (2023), 36(4), 323–328
    Paper