Join us Thursday, March 31 at 4 pm CET (10 am ET), to learn more about Artificial Intelligence!
Prof. Kalle Astrom will tell us more, LIVE from Sweden!
Humans and animals excel in their power-efficient use of sensor data and the processing of such data to decide on relevant actions. Although natural cognition is still poorly understood, the automatic analysis of sensor data, e g from cameras, microphones, radio receivers, and inertial measurement units, is developing rapidly and is finding applications and use in all parts of society. Within the health care sector, there is an increased need to automatically analyse medical data e g from imaging (PET, CT, MR, etc.) or from medical records, or from wearable devices. Within research, there is a need to automatically analyse, extract and visualize relevant data from e g MAX IV and ESS. Other application areas are for example self-driving cars and smart-phones. In this talk, I will discuss two aspects of perception: the structure from motion problem and machine learning, and also present our work on building a network for artificial intelligence research at Lund University. The structure from motion problem in computer vision is the problem of determining camera position and orientation as well as the 3D positions of scene features using the motion of image features only. The analogous problem for audio and radio is the problem of determining sender and receiver positions using the received audio or radio signal only. For both video, audio, and radio there are a number of challenges, e.g. feature detection, robust feature matching, and robust parameter estimation. The problem is challenging also because of the non-linear nature of the problem. Machine learning has always been a central part of computer vision research, but the methods and results have increased dramatically since 2010. Problems that were virtually impossible ten years ago are now straightforward to solve. The field is still developing fast with the invention of new computational architectures and novel ways to pose the problem that allows for fewer labeled data.
Biography: Kalle Åström is a professor at the Centre for Mathematical Sciences at Lund University. His research spans from basic research in algebraic geometry to applied mathematical research (machine learning, image analysis, computer vision and mathematical modeling) and to dissemination in industry and society as a whole. His master's thesis on autonomous vehicles was awarded first prize in Innovation Cup, 1991. The dissertation from 1996, was awarded the best Nordic dissertation in machine learning and image analysis at the Scandinavian Conference on Image Analysis in 1997. He is the founder of four companies, Neuromathic, Spiideo, Cognimatics and Decuma, where Decuma won the EU's "IST grand prize" in 2005. He has been co-editor for IEEE transaction on Pattern Analysis and Machine Intelligence and was the track chair for International Conference on Pattern Recognition in 2014. He has written about 230 reviewed scientific articles in scientific journals and conferences and has been the main or co-supervisor for 25 PhD students. Current research funding includes WASP (wasp-sweden.org), where he leads the Perception and Learning cluster, ELLIIT (elliit.se) and EU project Adacorsa (adacorsa.eu). Since 2018 he is coordinating the network for Artificial Intelligence and Machine Learning at Lund University, (ai.lu.se).