#AIDayFinland FINNISH CENTER FOR ARTIFICIAL INTELLIGENCE Samuel Kaski & the FCAI preparation team http://fcai.fi
2
EXPONENTIAL GROWTH STARTS SLOWLY BUT THEN
ARTIFICIAL INTELLIGENCE Recent breakthroughs in AI largely due to advances in machine learning, methods that learn from data, made possible by Algorithmic improvements (especially in deep neural networks) Increased computation capacity (GPU) Larger data sets (ImageNet, web-scale text data, ) Easy-to-use software, backed up by web-scale companies (TensorFlow etc) AI as technology Will bring disruptive transformations to all societies in a very short-term future, opening both new opportunities for growth and a need for solutions mitigating the effects of the societal changes The international competition of the new economy is tough, and the division to leaders and followers will take place within the next few years The gap between AI research and products is short compared to most other sciences, and fundamental and applied research often overlap
8 KEYS TO AI Finland s AI program at tekoalyaika.fi/ 1. Enhancement of business competitiveness through the use of AI 2. Effective utilisation of data in all sectors 3. Ensure AI can be adopted more quickly and easily 4. Ensure top-level expertise and attract top experts 5. Make bold decisions and investments 6. Build the world s best public services 7. Establish new models for collaboration 8. Make Finland a frontrunner in AI Much more detailed recommendations at tekoalyaika.fi/
FCAI IN BRIEF Aalto and University of Helsinki, 15+ companies Partnership program for others (international and national) 29 AI professors, 30+ professors of other fields Funding: Universities (professors), companies, and competitive funding Builds on the strong and long record of AI research in Finland Partners with existing ecosystems and helps create new ones to realize the potential for AI-led economic growth Goal: next-generation artificial intelligence, which is interactive, dependable, and dataefficient. Impact: Education broadly; results into use in partner companies and new companies, attractiveness through visibility Leading artificial intelligence center in the northern Europe. In AI publications at the level of many world s 20- best universities 7 http://fcai.fi
RESEARCH FOCUS: BOTTLENECKS 1. Data scarcity. Most of the value in big data is in the enormous number of small questions it could answer. At this resolution data becomes a scarce resource. We will make AI applicable to a significantly wider scope of questions by several means: data-efficiency, prior knowledge, privacy-preserving sharing of data. 2. Dependability. Many AI techniques cannot be trusted because they are vulnerable to manipulation and information stealing, and reliability is not known. We will develop the required privacy-preserving, secure and resilient AI. Societal trust stemming from dependable AI enables wide applicability. 3. AI does not understand the user. We will give AI the capability to understand the user, which enables for making AI understandable. Outcome: AIs that are able to augment human capabilities.
Finnish Center for Artificial Intelligence (FCAI) Head and deputy head of flagship candidate Prof. Samuel Kaski, Aalto University Prof. Petri Myllymäki, University of Helsinki Host organizations of candidate flagship Aalto University (coord) University of Helsinki RESEARCH AND IMPACT PLAN Fig. 1: Graphical table of contents of this document and the programs of FCAI.
HIGHLIGHTS AGILE PROBABILISTIC AI PROBABILISTIC PROGRAMMING Reliable inference from limited data Uncertainty quantification for (autonomous) decisionmaking Democratizes AI: Replace difficult mathematical derivations with programming tasks Core developers of Stan with 10,000+ users [mc-stan.org] Leading textbook in Bayesian statistics, 20,000+ citations
HIGHLIGHTS PRIVACY- PRESERVING AND SECURE AI PRIVACY-PRESERVING LEARNING Building trust between humans and AI, to make sharing and use of personal data possible Theoretical guarantees that personal data used for training AI cannot leak Web-scale competitors invest heavily in solving the challenge for large populations. This is not enough for e.g. rare diseases, MyData, local IoT sensoring etc. Our highlights: Privacy retained in drug sensitivity analysis already in populations below 1000 Privacy in distributed computation Security in AI
HIGHLIGHTS SIMULATOR- BASED INFERENCE Learning parameters of complex simulators from observed data enables next generation AI in health, human-computer interaction, process industry etc. ELFI: Engine for Likelihood-Free Inference Optimization of graphical user interfaces using psychological models and reinforcement learning. Breakthroughs in the design of GUIs, visualizations, input methods. Simulator model for spreading of infectious diceases, modeling transmission, mutation and recuperation.
HIGHLIGHTS
HIGHLIGHTS INTERACTIVE AI Inverse modelling of complex interactive behavior How: Cognitive user models, reinforcement learning and ABC Enables understanding the user and taking her goals and constraints into account Kangasrääsiö et al., CHI 201
H-index/citations: Median for AI faculty AI output: Publications in top-level AI forums in 2014-2017 Ranking: Shanghai University Ranking 2017 COMPETENCE Critical mass in Digitalization @Aalto: 80 professors directly in the field + 80 in key applications UH adds a significant number in CS FCAI is on par with many global top-20 universities known for their AI output: H-index Citations AI output Ranking FCAI (UH+Aalto) 32 5800 164 56 (UH) UCL (UK) 31 6100 117 16 Cornell (US) 38 8200 140 14 Stanford (US) 47 7000 337 2 Aalto CS: The US News ranking (computer science) 2018 #47 2017 #60 2016 #83 2015 #98 Academy professor, ERC Advanced grant, multiple starting grants, 6 Academy research fellows, 100+ patents,
PARTNERS 15+ company partners, including: Academic partnership program: More information will come late
WHAT S IN IT FOR ME? Work with experts: Company Membership Program (also for governmental organizations) Note: In this field, distance between fundamental research and appications is often very short. That is why we are not making a distinction. Hire an expert: All who have recently tried know that hiring real AI or machine learning experts is almost impossible. Networking within FCAI and working with students helps Get free labor to work on your problem: Contribute data, contribute research problems for students Become an expert: Teemu Roos will tell shortly about education Collaborate and contribute: AI research programs; AI and X program; Academic partnership program Common theme: Focus on substance! Additionally: Enjoy the better tools, improved decision making, wealth and prosperity AI brings to Finland!
CORE PERSONNEL: RESEARCH Privacy-preserving & secure AI Interactive AI Agile probabilistic AI Simulator-based inference Deep Learning Antti Honkela N. Asokan Antti Oulasvirta Giulio Jacucci Aki Vehtari Arto Klami Samuel Kaski Jaakko Lehtinen Harri Valpola Personalized medicine AI for Health AI and data science AI and X program Jukka Corander Samuli Ripatti Simo Särkkä Harri Lähdesmäki Hannu Toivonen Aristides Gionis Petri Myllymäki Kai Puolamäki
CORE PERSONNEL: ADMIN AND IMPACT