Vlad (Vladimir) Pyltsov
I am currently a PhD student in Mechanical Engineering at Columbia University in QSEL group under supervision of Professor Vijay Modi.
I received my BS in Mechanical Engineering from Boston University in 2023, where I worked with Professor Michael Gevelber and completed UROP under Professor Kamil Ekinci.
Research
My research broadly concerns modeling of energy systems with applications of modern frameworks such as machine learning and optimization. My recent work has been in building energy, forecasting and data-driven modeling, and techno-economic analysis.
Formal list of selected works can be found here: Selected Projects and Publications. I also write here about informal research (mostly Time-Series Forecasting but also other topics): Blog. If you find any of my work interesting or have questions, feel free to reach out via email.
News
Jan 2026 Paper published in "Solar Energy" journal. We analyze a solar-storage system in an urban setting and evaluate its economics given its ability to provide benefits to the distribution-level system.
Jul 2025 Placed 4-th at ICML CO-BUILD 2025 Competition. I applied a periodic subsampling strategy to train a tree-based architecture for temperature-sensor predictions on tabular data.
May 2025 Placed 2-nd in the IISE PG&E Energy Analytics Challenge 2025. We iterated various modern ML architectures with the periodic subsampling strategy to predict load in a tabular data setting.
Oct 2024 Joint paper published in "Energy and Building" journal. We provide results of the CO2 sensor placement experiments in large occupancy room and compare energy savings with that of camera-based occupancy detection.
Selected Projects & Publications
Blog
(Exp. Mar 2026) On the variance and bias trade-off of linear models in time series forecasting
I consider adding a mixer between bins in periodic sampled linear models to improve performance and encounter a typical bias-variance trade-off.
(Exp. Mar 2026) Modern approaches to loss functions in time-series forecasting
I review recent works on loss functions in time-series forecasting. I implement various proposed approaches and experiment with certain minor modifications.