Sina Sheikholeslami
Postdoc
Researcher
About me
I’m a Postdoc at the Division of Energy Systems, Department of Energy Technology at KTH, working with Francesco Fuso-Nerini and Ricardo Vinuesa on leveraging AI for Sustainability and Climate Action. I primarily work on the Beyond 2030 project, and I'm also affiliated with the KTH Climate Action Centre.
I did my PhD studies in the Distributed Computing Group of the Division of Software and Computer Systems (SCS), EECS School of KTH, where I was advised by Vladimir Vlassov, Amir Payberah, and Jim Dowling. During my PhD studies, I worked on systems for machine learning and deep learning. In particular, we developed and released the first framework for automated, parallel ablation studies for deep learning, called AutoAblation, as part of Maggy; and introduced a novel approach for dataset partitioning in data-parallel training of deep neural networks by considering the importance of dataset examples, named Importance-aware DPT, which won the Best Artefact Award of DAIS 2023.
Later, I became interested in the idea of "reusing" the computation results of one ML/DL pipeline stage in another stage. To that end, we showed that it makes sense to reuse "model weights" from the winning hyperparameter tuning trial, to initialize the model at the training stage. This work resulted in our paper, "Deep Neural Network Weight Initialization from Hyperparameter Tuning Trials", which is to appear in the proceedings of ICONIP 2024 but you can already find our author's version and our reproducible experiments here. My final work during my Ph.D. studies is about utilizing large language models (LLMs) for conducting ablation studies in ML/DL, and the corresponding paper is due to be published in the proceedings of EuroMLSys 2025.
Before starting my Ph.D. studies, I was an EIT Digital Master School scholar (2017-2019) and did my M.Sc. studies at Eindhoven University of Technology (first year) and KTH (second year) in Data Science (with a minor in Innovation and Entrepreneurship). I did my internship at Hopsworks AB and RISE SICS, which led to my thesis, "Ablation Programming for Machine Learning".
Prior to that, I was a Big Data R&D Engineer at Digikala.com. I did my Bachelor of Science in Computer Software Engineering at Amirkabir University of Technology (Tehran Polytechnic), where I was the President of CEIT's Students' Scientific Chapter from January 2014 to March 2015.
In addition to my studies, I was also a Steward of the PhD Chapter's Masters of Ceremonies Group. Before that, I was a Board Member and the Council Coordinator of KTH's PhD Chapter from January to July 2024. Prior to that, I was a member of EECS PhD Student Council (January 2020 - December 2023), where I was the Vice-chair (January 2021 - December 2022), and a representative in the New Faculty Appointment Committee, the Faculty Promotion Committee, the School Assembly of EECS, and the Council for Third-Cycle Education of EECS. I was also a member of the Nominating Committee of the PhD Chapter (May 2020 - December 2022).
Outside KTH, I have been Sweden's Local Representative for the EIT Digital Alumni Foundation since January 2020.
Courses
Data Mining (ID2222), assistant
Data Mining (FID3016), assistant
Data-Intensive Computing (ID2221), assistant, teacher
Operating Systems (ID1206), teacher