Asma Atamna's Home Page

About Me

I am a postdoctoral researcher in the group of Prof. Tobias Glasmachers  at Ruhr University Bochum, Germany. My research lies at the intersection of numerical optimization and machine learning (ML). I work on designing adaptive hyperparameter update mechanisms for ML optimizers, such as Adam and SGD. While research has shown that optimal hyperparameter values vary over the course of optimization, most existing approaches to hyperparameter optimization (HPO) focus on selecting static values that remain fixed. In contrast, adaptive mechanisms dynamically adjust hyperparameters based on real-time signals gathered during the optimization process, thereby also allowing to recover from suboptimal initial values. 

In parallel, I contribute to applied ML research aimed at supporting small and medium-sized enterprises (SMEs), particularly those with limited in-house ML expertise. Through the ecoKI project, I help develop a user-friendly ML platform that assists SMEs in the process industries with their digitalization and AI adoption efforts, ultimately helping reduce energy consumption and optimize operational efficiency.

Working closely with industrial partners made me increasingly aware that standard academic benchmarks often fail to capture the complexity of real-world problems, highlighting the need for more realistic evaluation settings in ML research. This observation motivated the development of ContainerGym, a reinforcement learning (RL) benchmark inspired from a real-world sequential decision making problem.