Intelligent Convergence Lab (ICL)

Our name reflects what we aim for: intelligent convergence. We bring together multiple sources of intelligence, including data priors, existing models, human expertise, and real-time feedback, to develop learning systems that can continuously adapt, transfer, and converge toward optimal solutions as tasks and environments change.

We work on the fundamentals of machine learning and AI across theory, algorithms, and applications. Our future research will focus on these aspects in next-generation generative AI, including controllable and test-time adaptive generation, reasoning and self-verification, and the theory of compositionality, diversity, and sampling efficiency. We also explore how these ideas can advance AI for Science, integrating foundation models, domain knowledge, human expertise, and experimental feedback to accelerate scientific discovery.

Department of Computer Science, Aalto University · ELLIS Institute Finland

Research Themes

Generative Models

Diffusion and flow models, controllable and test-time adaptive generation, and the theory of generalization and memorization in generative models.

Learning to Adapt

Meta-learning, continual learning and test-time adaptation: systems that keep learning and improving as tasks and data change.

Trustworthy ML

Robustness under distribution shift, algorithmic fairness, privacy, and reliable learning from limited data.

Learning Theory

Information-theoretic and PAC-Bayesian generalization bounds that explain, and guide, transfer and adaptation.

Probabilistic ML

Bayesian optimization, Gaussian processes, neural processes and prior-data fitted networks: principled uncertainty for decisions under limited data.

AI for Science

Foundation-model-guided Bayesian optimization for molecular discovery, and machine learning for healthcare such as digital hearing health.

Join Us

We are recruiting PhD students and postdocs, including through the ELLIS Institute Finland call (deadline 21 September 2026) and the HIIT Postdoctoral Fellow call (deadline 4 October 2026).

News