Seminars take place at UPF, Campus Poblenou, Roc Boronat, 138, Barcelona and will only be streamed/recorded if the speaker has granted permission. Rooms 55.309 / 55.410 streaming / Auditorium streaming
EiTIC members: If you are interested in giving a Research Seminar or you would like to invite a speaker, please fill in the following form RSDetails Form.
| Past Research Seminars | |||
| MARCH | |||
|
March, 27th
10:30 h
Room 55.309 |
Invited Research Seminar Design for effective orchestration and pedagogical interventions aligned with learning analytics in technology-enhanced learning ecosystems: Notes of a sabbatical yearBy Yannis Dimitriadis Abstract Technology-enhanced learning ecosystems are becoming quite complex, especially when non-conventional approaches, such as collaborative or inquiry learning are employed. On the other hand, the recent advances in the learning analytics field have been very promising, for purposes of understanding and optimizing learning and the environments in which it occurs. However, the alignment between design for learning and learning analytics has been recently shown to be a pending, albeit essential, issue that would allow for effective and efficient pedagogical interventions and orchestration. This informal seminar – colloquium will discuss some recent work during and after the sabbatical year of Prof. Dimitriadis (2017-2018) at University of Edinburgh, EPFL Lausanne and UC Berkeley. Some topics include: - Design for orchestration and analytics (theoretical frameworks and principles) - Examples of on-going work, as e.g. inquiry science learning using WISE units in K-12, coaching in Agile Software Engineering course in Higher Education - Reflections on research mobility and its effects on personal and community trajectories Biography Dr. Yannis Dimitriadis (https://www.gsic.uva.es/members/yannis and https://scholar.google.es/citations?user=xrzS-v4AAAAJ&hl=es&oi=ao) is Full Professor of Telematics Engineering and ex Dean of the School of Doctoral Studies of the University of Valladolid, Spain. He is also the coordinator of the GSIC/EMIC research group, an inter-disciplinary group, integrating 20 researchers and practitioners from the field of Information and Communications Technologies (ICT) and Pedagogy. His research interests include learning design, design patterns and the conceptual and technological support to the orchestration of computer-supported collaborative learning processes. He has participated in more than 55 competitive research projects on technology-enhanced learning, such as the Kaleidoscope Network of Excellence in technology enhanced learning, Sharetec, Metis, EEE or Reset, being the PI in 24 of them. Dr. Dimitriadis has co-authored more than 90 journal papers (52 indexed in ISI-JCR), 190 conference papers, and 24 book chapters. Host: Davinia Hernandez-Leo |
||
|
March, 28th
15:30 h
Room 55.309 |
Invited Research Seminar Learning of musical structures for automatic improvisation systemsBy Ken Déguernel Abstract This work is part of the "Creative Dynamics of Improvised Interaction" project (DYCI2) that aims at creating digital musical agents able to listen, learn and interact with human performers to generate creative improvisations. The aim of this work is to create a system able to learn the dependencies between several dimensions (e.g. pitch, harmony, rhythm…), take into account the form of music upon several levels of organisation (e.g. beat, measure, section…), and use this information to generate more creative and more artistically credible improvisations. Biography
Ken Deguernel is a doctor in Computer Music. His research interests include music informatics, formal language theory, semigroup theory, probabilistic models and musicology. In particular, his work focuses on the understanding, the analysis and the modelling of order and disorder parts of creative processes employed by musicians during improvisation and composition in different contexts and different styles.
Host: Xavier Serra
|
||
| APRIL | |||
|
April, 3
15:30 h
Room 55.309 |
PhD Research Seminar Data processing under uncertainty and tractability limitationsBy Santiago Mazuelas Abstract
General data-driven problems can be seen as decision problems in which actions are chosen with the aid of data. For instance, supervised classification uses training data to choose a classification rule; and sequential inference uses time series to sequentially choose a target variable.
The development of data processing techniques to address such problems is hindered by the fact that actual probability distributions of state variables are often unknown or intractable in practice. For instance, in supervised classification the actual distribution of features-label pairs is often unknown, and in sequential inference the actual posterior distribution of target values is often intractable.
In this talk, I will describe approximation techniques to address uncertainty and tractability limitations based on optimization methods. In particular, I will present data processing techniques that address distributions' uncertainty in supervised classification and distributions' intractability in sequential inference.
Biography
Santiago Mazuelas received the Ph.D. in Mathematics and Ph.D. in Telecommunications Engineering from the University of Valladolid, Spain, in 2009 and 2011, respectively.
Since 2017 he has been Ramon y Cajal Researcher at the Basque Center for Applied Mathematics (BCAM). Prior to joining BCAM, he was a Staff Engineer at Qualcomm Corporate Research and Development from 2014 to 2017. He previously worked from 2009 to 2014 as Postdoctoral Fellow and Associate in the Wireless Information and Network Sciences Laboratory at the Massachusetts Institute of Technology (MIT). He is a frequent visitor of the Laboratory for Information and Decision Systems (LIDS) at the MIT, where he holds the Research Affiliate appointment. His general research interest is the application of mathematics to solve engineering problems, currently his work is primarily focused on statistical signal processing, machine learning, and data science.
Dr. Mazuelas is Associate Editor for the IEEE Communications Letters and served as Co-chair for the Symposiums on Wireless Communications at the 2014 IEEE Globecom and at the 2015 IEEE ICC. He has the Young Scientist Prize from the Union Radio-Scientifique Internationale (URSI) Symposium in 2007, and the Early Achievement Award from the IEEE ComSoc in 2018. His papers received the IEEE Communications Society Fred W. Ellersick Prize in 2012, and Best Paper Awards from the IEEE ICC in 2013, the IEEE ICUWB in 2011, and the IEEE Globecom in 2011.
|
||
|
April, 25
15:30 h
Room 55.309 |
Invited Research Seminar Compositionality and Automated Hierarchical Skill Discovery Using KL ControlBy Andrew Saxe Abstract Hierarchical architectures are critical to the scalability of reinforcement learning methods. Most current hierarchical frameworks execute actions serially, with macro-actions comprising sequences of primitive actions. We propose an alternative to these control hierarchies based on concurrent execution of many actions in parallel. Our scheme exploits the guaranteed concurrent compositionality provided by the linearly solvable Markov decision process (LMDP) framework, which naturally enables a learning agent to draw on several macro-actions simultaneously to solve new tasks. We introduce the Multitask LMDP module, which maintains a parallel distributed representation of tasks and may be stacked to form deep hierarchies abstracted in space and time. Next we turn to the problem of learning an appropriate hierarchical decomposition of a domain into subtasks. In the compositional setting afforded by the LMDP, the subtask discovery problem can be posed as finding an optimal low-rank approximation of the set of tasks the agent will face in a domain. We use non-negative matrix factorization to discover this minimal basis set of tasks, and show that the technique learns intuitive multilevel hierarchical decompositions in a variety of domains. Our method has several qualitatively desirable features: it is not limited to learning subtasks with single goal states, instead learning distributed patterns of preferred states; it learns qualitatively different hierarchical decompositions in the same domain depending on the ensemble of tasks the agent will face; and it may be straightforwardly iterated to obtain genuinely deep hierarchical decompositions. Biography Dr. Andrew Saxe is a Postdoctoral Research Associate in the Department of Experimental Psychology, University of Oxford working with Christopher Summerfield and Tim Behrens. He was previously a Swartz Fellow at Harvard University with Haim Sompolinsky. He completed his PhD in Electrical Engineering at Stanford University, advised by Jay McClelland, Surya Ganguli, Andrew Ng, and Christoph Schreiner. His dissertation received the Robert J. Glushko Dissertation Prize from the Cognitive Science Society. His research focuses on the theory of deep learning and its applications to phenomena in neuroscience and psychology. Host: Vicenç Gómez |
||