Overview
HAIF unites research groups from diverse fields including computing, materials science, social sciences, law, humanities, and health sciences. The project’s unique, human-centric approach emerges from this interdisciplinary collaboration. HAIF’s objective is to provide training that promotes the safe, secure, legally, and ethically sustainable use of AI, with an emphasis on humans as developers, users, and decision-subjects affected by the technology.
In addition to individual and societal perspectives, HAIF’s research themes explore the technological properties that shape interactions between humans and AI systems, such as transparency, interpretability, reliability, and accountability. The global research collaborations and industry partnerships within the HAIF network bring internationality, cross-sectorality, and business relevance to the doctoral training project.
TurkuNLP’s Involvement
At TurkuNLP, our research group focuses on the algorithmic and computational aspects of natural language processing (NLP), a cornerstone of modern AI. Our work centers on machine learning for NLP, combined with large-scale corpora, regularly handling collections in the 100+ billion word range. Recent projects include training generative large language models in high-performance computing environments, and building core language technology for Finnish and numerous other languages. We extensively utilize national computing resources and have been selected to pilot the two most recent generations of GPU-accelerated supercomputers, including LUMI, Europe’s largest supercomputer.
Research Topics
Within HAIF, our group is pursuing the following research directions:
Investigating how to compress and transfer knowledge from large multilingual reasoning models into smaller, modular agents that retain reasoning abilities while being efficient for deployment in resource-constrained environments.
Designing multilingual LLM-based agent systems that collaborate across languages to solve complex tasks, dynamically delegating work based on linguistic capabilities and reasoning strengths.
Research on LLM-based systems that retrieve, reason over, and explain health information in an accurate, trustworthy manner adapted to patients' literacy levels.
Exploring how models integrate and reason across diverse input modalities including vision, language, audio, and biosignals with causal, spatial, and temporal reasoning.
Developing agentic systems that decompose multimodal reasoning into planning, tool selection, and execution.
Exploring whether program-based reasoning and structured representations outperform scaling for enhancing reasoning in small models.
Keywords: NLP, large language models, large corpora, deep learning, multilingual methods, reasoning agents, health applications of NLP, multimodal reasoning
Collaboration with ELLIS Institute Finland
Our team is also dynamic team with both TurkuNLP group and ELLIS Institute Finland, a world-class research hub in AI and machine learning, focusing on advancing natural language processing and machine learning in real-world applications.