About#
I’m a Mathematics and Systems Sciences student at Aalto University and an AI/ML developer at AI4Value, where I work on probabilistic machine-learning tooling for hard, individualised problems. Originally from Oulu, based in Helsinki / Espoo.
Research statement#
My work reads as one program rather than a scatter of projects — and the academic record, read alongside it, is the same program told from a second angle. The shape is N-of-1 optimisation under uncertainty: each individual treated as their own dynamical system, the data sparse and noisy, the action consequential.
The recurring method is amortised / simulation-based Bayesian inference over a mechanistic (“digital shadow”) forward model — encode the structure you understand into a simulator, then learn the otherwise-intractable inverse. EASE Health and StimIQ are that one idea applied to two different organs; QSVT4CRA is the same statistical machinery moved into quantitative finance.
Underneath sits a disposition toward epistemic honesty — a preference for models that know their own domain of validity and hand off to a human when they leave it. The Lorenz attractor on the homepage is the fitting signature: a fully deterministic system that is nonetheless practically unpredictable — uncertainty placed front and centre.
Methods#
Amortised / simulation-based Bayesian inference (SNPE, the sbi library)
Mechanistic forward models / “digital shadows”
Dynamical systems, stochastic processes, time-series analysis
Optimisation & dynamic optimisation (incl. Pareto multi-objective)
Active learning & information-gain–driven experimental design
Copulas & dependence modelling
Statistical & probabilistic modelling
Languages: Python, Scala, C++, MATLAB, R. Tools: PyTorch, sbi, TensorFlow, pandas/NumPy, Docker, CI/CD; cloud (Azure, Verda) and HPC (LUMI).
Contact#
GitHub: PlayerPlanet
LinkedIn: https://linkedin.com/in/konsta-kiirikki