Prof. Carlo Regazzoni
University of Genova, Italy
Unsupervised clustering of generalized coordinates for incremental generative models in autonomous systems
Learning incrementally from past experiences is a process at the basis of self awareness in autonomous systems. A self aware system must be able to analyze proprioreceptive and exteroreceptive sensor data collected during past experiences or by observing external objects doing experiences on their own and to extract from them generative models. Such models should allow it to predict behaviors when facing new experiences and to eveluate if such new experiences are governed by the same statistical rules as past experiences or follow different rules, i.e. can be considered anomalies wrt previous experiences. Discrepancies detected, (i.e. generalized errors) should make the agent capable to learn new generative models so allowing the agent to remain updated wrt unstationary conditions of the external world and of itself. In this incremental process a key step for the agent is being capable to estimate new generative models from generalized errors. In this talk we will discuss autonomous vehicles and cognitive Uavs applications to look how the process of estimating new generative models requires using unsupervised clustering techniques capable to find clusters of generalized errors characterized by good predictive properties Examples will be shown in which high dimensional (eg lidar and video) are used together with low dimensional sensor data (eg odometry or proprioreceptive data) to obtain clusters of generalized errors at the latent space of Bayesian filters used as generative models.
Prof. Carlo S. Regazzoni (Senior Member, IEEE) received the M.S. and Ph.D. degrees from the University of Genova, Genoa, Italy, in 1987 and 1992, respectively. Since 2005, he has been a Full Professor of Cognitive Telecommunications Systems with DITEN, University of Genova. He has been coordinating international interactive and cognitive environment Ph.D. courses with UNIGE, Geneva, Switzerland, since 2008. He has authored peer-reviewed papers on more than 100 international journals and on international conferences proceedings (350). His research interests include cognitive dynamic systems, adaptive and self-aware multimodal signal processing, Bayesian machine learning, and cognitive radio. Prof. Regazzoni served as a Vice President Conferences IEEE Signal Processing Society from 2015 to 2017, and as the general chair, the technical program chair and other roles in several international conferences in its research field. He is/has been an Associate/Guest Editor of several international journals, including Proceedings of the IEEE, IEEE Transactions on Image Processing, IEEE Transactions on Circuits and Systems for Video Technology, and IEEE Signal Processing Magazine.