Automated machine learning
How model selection and hyperparameter search can use limited computational resources more carefully, including when a search should stop.
Ahsan HabibResearch
Curiosity, made rigorous.
I am interested in what makes a machine-learning result useful and trustworthy: the question being asked, the data behind the comparison and the resources spent to reach it. Earlier work in education and cultural heritage still shapes these questions.
Questions I work on.
Evaluation and evidence
How data leakage, related observations and comparison design affect the conclusions we draw from an experiment.
Learning and representation
How an idea is represented, whether in code, a physical learning activity or a 3D object, and what that representation makes possible.
The question I am pursuing now
When should a search stop?
Automated machine learning can try hundreds of model configurations. Improvements often slow well before the budget is spent, yet the search keeps consuming time, compute and energy. My doctoral research studies search-level stopping, deciding when enough has been learned, and how to compare such methods without leakage flattering the result.
- Stop after
- 40 of 200 trials
- Compute saved
- 80%
- Best score kept
- 96.7%
Published work.
- 2021
Journal article
Analyzing students’ experience in programming with computational thinking through competitive, physical, and tactile games: the quadrilateral method approach
Turkish Journal of Electrical Engineering and Computer Sciences, 29(5), 2280–2297.
Read the publication - 2017
Conference paper
Enhancing understanding of programming concepts through physical games
Learning Innovation and Teaching Enhancement Research Conference (UM-LiTeR), Kuala Lumpur, 15–16 March 2017, pp. 39–40.
Read the publication - 2016
Conference paper
ShonaBondhu: a cloud based system to handle flash flood
2016 International Conference on Networking Systems and Security (NSysS), pp. 1–6.
Read the publication
Research in progress.
These are ongoing research directions, separate from the published work above.
Doctoral research · In progress
My current PhD work examines search-level stopping and the validity of machine-learning comparisons, with attention to leakage controls and resource accounting.
A continuing practical question
I am also interested in how people learn to use AI thoughtfully: checking outputs, keeping opportunities for independent practice and deciding when human judgement should lead.
A conversation starts here
Let’s compare notes.
Whether it is a research idea, a system that needs untangling, a class to teach or a question you are still shaping, I would be glad to hear from you.
ahsan@pearlhx.com
- Malaysia
- +60 11 5117 1332
- Bangladesh
- +880 1717 560450