AA4 – Energy Transition



Using Mathematical Programming to Enhance Multiobjective Learning

Project Heads

Aswin Kannan

Project Members

Zijun Li

Project Duration

01.11.2022 − 31.12.2025

Located at

HU Berlin


The relatively newer space of Multiobjective machine learning has focused on Bayesian optimization, with tuning limited to the space of model hyperparameters. Examples of objectives include accuracy, inference time, and complexity. We aim to develop tractable optimization formulations to address the extensive tunable space of models, hyperparameters, model parameters, and feature engineering embodiments. We will also address questions related to multi-label classification, regression, and algorithmic bias (example: fairness). From an Implementation perspective, our focus will be on developing tailored versions of solvers like SGD (by exploitation of problem structure). Our main applications of interest include problems in energy markets and imaging.

Bachelors Theses 

(1) Thomas Dittmar (HU Berlin, summer 2023): Comparison of machine learning models on tabular weather data for the prediction of wind power production.

(2) Hasan Mert Goekalp (HU Berlin, summer 2023): On effectively handling metrics beyond accuracy in renewable energy based regression problems.

Related Publications

(1) Zijun Li and Aswin Kannan, Algorithm Switching for Multiobjective Predictions in Renewable Energy Markets, Under Review, December 2023.

(2) Aswin Kannan, Benefits of Multiobjective Learning in Solar Energy predictions, Proceedings of the AI2SE Workshop, AAAI, February 2023.

(3) Aswin Kannan Et. Al., HyperASPO: Fusion of Model and Hyper Parameter Optimization for Multi-objective Machine Learning, Proceedings of the IEEE Conference on Big Data, December 2021.

Related Pictures

Benefits of hybrid schemes and multiobjective learning in renewable energy predictions.