1. Data Science - Machine Learning for Scientific Applications

Machine learning is revolutionizing science. On the one hand, deep learning has supported major advances in computer vision and natural language processing, as recognized by the Turing award (Nobel award counterpart in computer science) in 2018. On the other, explainable ML methods can produce knowledge from scientific data. Our host groups, JSI-KT and UL-FRI, perform world-leading research in this area, together with some of the other partners in SMASH, having coordinated high-profile international projects and organized premier scientific events, e.g., ICML, ECML/PKDD (JSI-KT) and won the ’Excellent in Science’ prize for Computer and Information Science in 2019, 2020 (UL-FRI), and 2021 (both UL-FRI and JSI-KT) from ARRS. Some of the ongoing interdisciplinary research activities at the host partners are as follows:

Research sub-areas

1.1 Learning from complex data and computational scientific discovery

JSI-KT is developing methods for learning from complex scientific data, including temporal, spatial and relational data. This includes automated modelling of dynamical systems via symbolic regression, relational ensemble learning, and combining explainable machine learning with neural methods for representation learning.

Host institution: JSI KT

1.2 Deep learning and computer vision

UL-FRI is developing several computer-vision learning-based solutions such as anomaly detection, image segmen- tation, object detection and counting and visual tracking.

Host institution: UL FRI

1.3 Beyond supervised learning

UL-FRI and JSI-KT are developing methods for learning under different levels of supervision, including unsupervised, semi- supervised and self-supervised learning (pre-training), few-shot learning, weakly supervised learning, and active learning, to more efficiently explore data while minimizing the tedious and costly human annotations.

Host institution: JSI KT, UL FRI

1.4 Software and infrastructure for High-Performance Computing

IZUM is developing tools, services and infrastructure for the public sector and industry to make better use of supercomputing and large capacity storage resources.

Host institution: IZUM

SMASH research areas
1. Data Science - Machine Learning for Scientific Applications
2. Fundamental Physics - Machine learning for Particle Physics, Astrophysics and Cosmology
3. Linguistics - Computing for Human and Animal Communication
4. Climate - Machine Learning in Climate Research
5. Precision Medicine - Personalised Medicine and Life Sciences