SINGPRO project – Combining Data Analytics with Process Optimization

Date: December 02, 2019

Abstract: The aim of the SINGPRO project is to merge the Big Data platforms, machine learning and data analytics methodologies with process planning and scheduling optimization technologies. It applies these technologies to provide online, reactive and anticipative tools for more sustainable and efficient operations. The currently employed classical mathematical optimization models are limited by fixed parameter sets, which are commonly updated off-line and represent only statistical averages. Such parameters could be estimated much more precisely in an on-line fashion using Big Data technologies. By creating such collaboration interfaces between scheduling optimization, big data analytics and machine learning, the process related decision-making will become much more agile, self-aware and flexible. With the sophisticated data analytics methods, one can embed to the overall key performance indicators (KPI) also all information about the process, e.g., tracking abnormal situations (anomaly detection), individual process equipment performance degradations (predictive maintenance), anticipated process timings (prediction of process behavior) and scenario simulation (e.g., artificial intelligence AI planning). This is joint work with Iiro Harjunkoski from Aalto/CHEM.

Bio: Keijo Heljanko is a Full Professor of Computer Science at the University of Helsinki. He is a vice director of the HiDATA, Helsinki Center for Data Science. His research interests include distributed and parallel computing, Big Data, data science, and data science applications.

Speaker: Keijo Heljanko
Affiliation: Professor of Computer and information Sciences, Helsinki University

Place of Seminar: Lecture Hall T5, Konemiehentie 2, Aalto University

Machine learning in drone remote sensing

Date: November 18, 2019

Abstract: Utilization of unmanned aerial vehicles, drones, has increased explosively during the past couple of years. Integrating analytics and actuation either utilizing individual drones, drone teams or drone swarms will enable robotization of various environmental operations in the future. This presentation focuses on drone based remote sensing in agricultural and forestry applications. Drones become intelligent when coupled with sensors and analytics. Imaging sensors (color, hyperspectral and multispectral cameras) enable exploration of objects’ 3D structure, and reflection and absorption characteristics. Essential data preprocessing steps are georeferencing, 3D object reconstruction and radiometric calibration. Machine learning plays a fundamental role in building the analytics process. The physical observations measured in field or in laboratory are related to the remote sensing observations and models are built to extract information. Then in analytics applications remote sensing data are fed to the analyzer and the information are extracted utilizing the models. This presentation deals with remote sensing sensor, data calibration and machine learning and presents selected case studies in forest inventory, tree health analysis and grass quality and quantity assessment.

Bio: Dr. Eija Honkavaara is Research Manager at the Finnish Geospatial Research Institute in the National Land Survey (FGI) and Docent at the Aalto University in the field of Automatic Systems in Photogrammetry. Her current research focuses on drone based geoinformation and AI technologies for environmental remote sensing. The DroneFinland-team led by Dr Honkavaara has developed several generations of high-performance drone remote sensing systems with cameras, hyperspectral sensors, analytics applications as well as inhouse data processing chains. She has a wide international research network, is a leader of working group “Hyperspectral Imaging Applications” in the International Society for Photogrammetry and Remote Sensing (ISPRS) and has led several national and international projects on UAV remote sensing.

Speaker: Eija Honkavaara
Affiliation: Finnish Geospatial Research Institute, FGI

Place of Seminar: Lecture Hall T5, Konemiehentie 2, Aalto University

FCAI Research Programme on Interactive AI

Date: November 11, 2019

Abstract: How to create machines that understand people and can work with them as partners? This talk provides an overview of FCAI’s research on interactive AI and AI-aided design. The research program builds on simulation-based AI, which offers an exciting way to combine first principles-based and data-driven methods. I will review the methodology, present applications, and discuss open challenges.

Bio: Antti Oulasvirta leads the User Interfaces research group at Aalto University and the FCAI Research Programme on Interactive AI. Prior to joining Aalto, he was a Senior Researcher at the Max Planck Institute for Informatics and the Cluster of Excellence on Multimodal Computing and Interaction at Saarland University. He received his doctorate in Cognitive Science from the University of Helsinki in 2006, after which he was a Fulbright Scholar at the School of Information in the University of California-Berkeley in 2007-2008 and a Senior Researcher at Helsinki Institute for Information Technology HIIT in 2008-2011. He was awarded the ERC Starting Grant (2015-2020) for research on computational design of user interfaces. Dr. Oulasvirta serves as an associate editor for ACM TOCHI and frequently participates in the paper committees of HCI conferences, including the ACM SIGCHI Conference on Human Factors in Computing Systems (CHI). His work has been awarded the Best Paper Award and Best Paper Honorable Mention at CHI twelve times between 2008 and 2019. He has held keynote talks on computational user interface design at NordiCHI’14, CoDIT’14, EICS’16, IHCI’17, ICWE’19, and Chinese CHI ’19. In 2019, he was invited to the Finnish Academy of Science and Letters.

Speaker: Antti Oulasvirta
Affiliation: Associate Professor, User interfaces, Department of Communications and Networking, Aalto University

Place of Seminar: Lecture Hall Exactum D122, University of Helsinki

FCAI Simulator-based inference: AI-assisted vaccine development

Date: November 04, 2019

Abstract: Several recent studies suggest that negative frequency-dependent selection (NFDS) acting through accessory genome loci is a strong force influencing the population evolution of major human bacterial pathogens. Such models can be fitted using likelihood- free inference for simulator-based statistical models, for example using the ELFI software (elfi.ai). Here we review some of these results and combine them with machine learning to introduce a practical method for optimizing vaccines by jointly using data from genomic population surveys and epidemiological studies. The results may lead to largely improved workflows to design new vaccines.

Bio: Jukka Corander is a professor at the Institute of Basic Medical Sciences, Faculty of Medicine, University of Oslo, Norway, also a professor at the Department of Mathematics and Statistics, University of Helsinki, Finland, and an honorary faculty member at the Wellcome Sanger Institute, Cambridge, UK. He has won two ERC grants, an ERC StG (2009-2014) from the mathematics and statistics panel and an ERC AdG (2017-2022) from the infection and immunity panel. He has published nearly 250 scientific peer-reviewed articles and his main research interests are microbial evolution and transmission modeling, statistical machine learning, population genomics, and inference algorithms. The statistical methods introduced by his research group have led to numerous discoveries on the evolution, resistance, virulence, and transmission of pathogenic bacteria and viruses.

Speaker: Jukka Corander

Affiliation: Professor of Statistics Department of Mathematics and Statistics, University of Helsinki

Place of Seminar: Lecture Hall T5, Konemiehentie 2, Aalto University

FCAI Highlight on easy modeling tools: The case of prior predictive distribution

Date: October 28, 2019

Abstract: One of the goals for FCAI is to make the development of probabilistic AI solutions easier and faster, by combining fundamental research on modeling principles with open-source tools that are being used by the industry and academia. This talk provides an overview to the activities related to this, and in particular drills into recent research on how the prior predictive distribution can be used for eliciting expert knowledge and for automatic parts of the modeling process.

Bio: Arto Klami is an assistant professor of computer science at University of Helsinki. He leads the Multi-source Probabilistic Inference group and is coordinating the FCAI Highlight on Easy and privacy-preserving modeling tools. He received his PhD from Aalto University in 2008 and worked as Academy Research Fellow in 2013-2019. He has published more than 60 scientific articles and his main research area is statistical machine learning, with contributions for example in data integration and approximate inference. He has worked on wide range of applications from computational neuroscience to modeling human activity, and is currently focusing on AI-driven ultrasonic cleaning and hyperspectral imaging.

Speaker: Arto Klami
Affiliation: Assistant Professor , Department of Computer Science Helsinki Institute for Information Technology

Place of Seminar: Lecture Hall Exactum D122, University of Helsinki

FCAI COMMUNITY EVENT

Date: October 21, 2019

FCAI organizes a community event to share what we do and to welcome new researchers from Aalto University, University of Helsinki, and VTT to join the FCAI community. We invite AI researchers as well as researchers from different fields to join the event!

Time: 9 – 11 AM
Venue: T1, CS Building, Konemiehentie 2, AALTO UNIVERSITY

FCAI Privacy-preserving and secure AI

Date: October 14, 2019

Abstract: I will introduce privacy work of FCAI research program in Privacy-preserving and secure AI. Our goal is to develop methods that allow learning from data under strong differential privacy guarantees. These enable setting strict limits on how much an adversary can learn about an individual contributing to a data set from the outcome of a computation using that data set. After a general overview, I will discuss differentially private data anonymisation.

Speaker: Antti Honkela
Affiliation: Associate Professor, Department of Computer Science, University of Helsinki

Place of Seminar: Lecture Hall Exactum D122, University of Helsinki

FCAI Agile Probabilistic AI

Date: October 07, 2019

Abstract: I provide an overview of FCAI Agile probabilistic AI research program, which develops interactive and AI-assisted processes for building new AI models with practical probabilistic programming. I also discuss common themes in existing workflow patterns in Bayesian statistics and machine learning, and ideas how to assist a user through the complete workflow so that the user expertise is maximally exploited and made stronger.

Speaker: Aki Vehtari
Affiliation: Associate Professor in computational probabilistic modeling, Aalto University

Place of Seminar: Lecture Hall T6, Konemiehentie 2, Aalto University

Engineering and testing of AI systems

Date: September 30, 2019

Abstract: There is no AI without software. To build and deploy AI systems for real use far more is needed than machine learning models. In this talk, I will briefly describe the software engineering challenges in the development of AI systems. Surprisingly, the intersection of AI and SW engineering has only recently started to attract wider interest. I will, in particular, discuss the testing of AI systems, how it differs from classical SW testing, and what kind of new interesting problems arise. I will conclude with an example of our own work of fault injection to datasets for the testing of ML systems.

Bio: Jukka K. Nurminen started as a professor of computer science at the University of Helsinki in spring 2019. He has worked extensively on software research in industry at Nokia Research Center, in academia at Aalto University, and in applied research at VTT. His key research contributions are on energy-efficient software and mobile peer-to-peer and cloud solutions but his experience ranges widely from applied optimization to AI, from network planning tools to mobile apps, and from software project management to tens of patented inventions. He received his MSc degree in 1986 and PhD degree in 2003 from Helsinki University of Technology (now Aalto University). Currently his main interests are on tools and techniques for developing data-intensive software systems including testing of AI solutions, computational moral as well as software development for quantum computers.

Speaker: Jukka K. Nurminen

Affiliation: Professor of Computer Science, University of Helsinki

Place of Seminar: Lecture Hall Exactum D122, University of Helsinki

Can AI be responsible?

Date: September 23, 2019

Abstract: The question whether AI can be responsible can be understood in two ways: Either as “Can AI agents be responsible?” or “Can AI research be responsible?”. I will argue that only the latter is possible, because AI agents including robots are not autonomous in the sense required by fitness to be held morally responsible. The argument employs Alfred Mele’s history-sensitive account of autonomy and responsibility to the effect that even if robots were to have all the capacities required of moral agency, their history would deprive them from autonomy in a responsibility-undermining way.

The talk is based on joint work with Pekka Mäkelä, in particular the article: Raul Hakli & Pekka Mäkelä: Moral Responsibility of Robots and Hybrid Agents. The Monist, Volume 102, Issue 2, April 2019, Pages 259–275, https://doi.org/10.1093/monist/onz009

Speaker: Raul Hakli
Affiliations: University Lecturer, Practical Philosophy, University of Helsinki

Place of Seminar: Lecture Hall T6, Konemiehentie 2, Aalto University

Introducing AI in Society Program

Date: September 23, 2019

Speaker: Petri Ylikoski
Affiliations: Professor of Sociology, University of Helsinki

Place of Seminar: Lecture Hall T6, Konemiehentie 2, Aalto University

Real-world model-based reinforcement learning with deep neural networks

Date: September 16, 2019

Speaker: Harri Valpola
Affiliation: CEO, co-founder at Curious AI

Place of Seminar: Lecture Hall Exactum D122, University of Helsinki

Intelligent Mobile Robots

Date: September 09, 2019

Abstract: Dr. Saarinen is among the pioneers in the field of mobile robotics. Prior to establishing GIM Ltd, 2014 a leading mobile robotics R&D company in Finland, he was 15 years in academia. During this time he researched autonomous navigation with a special focus on lidar-based approaches. He is responsible for over 50 peer-reviewed research articles in scientific publications. His widely cited work includes contributions in 3D mapping and localization. Currently, Dr Saarinen is leading team of 30 highly talented robotics engineers with the aim of becoming the largest supplier of intelligent mobile robot solutions.

Speaker: Jari Saarinen
Affiliation: CEO and Founder of GIM Ltd, CTO and founder of Sensible 4

Place of Seminar: Lecture Hall T6, Konemiehentie 2, Aalto University