Escola Avançada em Big Data Analysis
Global pooling in deep convNets: why, where and how
I will discuss in this talk the importance of the global pooling operation in deep convolutional architectures. Several recent approaches only differ in the way (how) and place (where) it is achieved inside the network. I will detail several block combinations in deep architectures to achieve global pooling, and compare different pooling functions. Results and evaluations on different datasets for visual classification tasks will support (or not) our statements.
Vagner Figueredo de Santana
IBM Master Inventor, researcher at IBM, and associated collaborator at Federal University of ABC (UFABC). PhD and MSc degrees in Computer Science from University of Campinas (UNICAMP), 2012 and 2009, respectively. BSc degree in Computer Science from Presbyterian University Mackenzie (2006).
Was webmaster of Folha Online (www.folha.com.br) from 2002 to 2007 and visiting professor at Presbyterian University Mackenzie from 2009 to 2015.
Researches topics on Human-Computer Interaction since 2006.
Acts as a reviewer on the following conferences and journals:
- Reviews full papers for CHI, EICS, IHC, IHCI, INTERACT, MobileHCI, and CBIE conferences.
- Reviews papers for IJHCS (International Journal on Human Computer Studies) and UAIS (Universal Access in the Information Society) journals.
- Committee member of the IHC, SBSC and HCII conferences.
Main areas of interest are: Evaluation of User Interfaces, Interaction Log Analysis, Visual Analytics, Web Usage Mining, and Software Engineering.
Biometric performance evaluation with novel visualization
Biometric authentication verifies the identity of individuals based on what they are. However, biometric authentication systems are error prone and can reject genuine individuals or accept impostors. Researchers on biometric authentication quantify the quality of their algorithm by benchmarking it several databases. However, although the standard evaluation metrics state the performance of a system, they are not able to explain the reasons of these errors.
After presenting the existing evaluation procedures of biometric authentication systems as well as visualisation properties, this talk presents a novel visual evaluation of the results of a biometric authentication system which helps to find which individuals or samples are sources of errors and could help to fix the algorithms. Two variants are proposed: one where the individuals of the database are modelled as a firected graph and another one where the biometric database of scores is modelled as a partitioned power-graph where nodes represent biometric samples and power-nodes represent individuals. A novel recursive edge bundling method is also applied to reduce clutter. This proposal has been successfully applied on several biometric databases and proved its interest.
Romain Giot
Dr. Romain Giot obtained his PhD from University of Caen under the supervision of Pr. Bernadette Dorizzi and Pr. Christophe Rosenberger for his works in keystroke dynamics. He is now associate professor at the University of Bordeaux and he belongs to the LABRI research lab. He is particularly interested in template update for biometric systems and visualization algorithms for very large graphs.
Albert Bifet
Albert Bifet is Associate Professor at Telecom ParisTech and Honorary Research Associate at the WEKA Machine Learning Group at University of Waikato. Previously he worked at Huawei Noah's Ark Lab in Hong Kong, Yahoo Labs in Barcelona, University of Waikato and UPC BarcelonaTech. He is the author of a book on Adaptive Stream Mining and Pattern Learning and Mining from Evolving Data Streams. He is one of the leaders of MOA and Apache SAMOA software environments for implementing algorithms and running experiments for online learning from evolving data streams. He was serving as Co-Chair of the Industrial track of IEEE MDM 2016, ECML PKDD 2015, and as Co-Chair of KDD BigMine (2017, 2016, 2015, 2014, 2013, 2012), and ACM SAC Data Streams Track (2017, 2016, 2015, 2014, 2013, 2012).
Emílio Vital Brazil
Possui mestrado e Doutorado em Matemática com especialização em Computação Gráfica pelo IMPA (Instituto Nacional de Matemática Pura e Aplicada). No pós-doutorado trabalhou em Teoria dos Grafos e Computação Gráfica na COPPE/UFRJ. Também trabalhou como pós-doutorando por quatro anos no grupo de pesquisa em Interactive Reservoir Modeling, Visualization and Analytics na University of Calgary. Neste período trabalhou em cooperação com profissionais da indústria de petróleo no desenvolvimento de technologias interativas de computação visual em vários problemas de goe-ciências. Seus interesses incluem aplicações de computação gráfica e visualização, bem como aplicações de Computação gráfica que envolvem Aprendizado de Máquina. Atualmente é cientista pesquisador da IBM -Research Lab Brazil, no grupo Visual Analytics and Comprehension.
Bruno Gonçalves
Bruno Gonçalves is a Data Science fellow at NYU's Center for Data Science while on leave from a tenured faculty position at Aix-Marseille Université. He has a strong expertise in using large scale datasets for the analysis of human behavior. Since 2008 he has been pursuing the use of Data Science and Machine Learning to study human behavior. By processing and analyzing large datasets from Twitter, Wikipedia, web access logs, and Yahoo! Meme he studied how we can observe both large scale and individual human behavior in an obtrusive and widespread manner. The main applications have been to the study of Computational Linguistics, Information Diffusion, Behavioral Change and Epidemic Spreading. He is the author of 60+ publications with over 4800+ Google Scholar citations and an h-index of 28. In 2015 he was awarded the Complex Systems Society's 2015 Junior Scientific Award for "outstanding contributions in Complex Systems Science" and he is the editor of the book Social Phenomena: From Data Analysis to Models (Springer, 2015) and the author of the forthcoming book "Twitterology: The Social Science of Twitter" (Springer, 2018)".
Matthieu Cord
Matthieu Cord is Full Professor at the Computer Science Department LIP6, at UPMC-Sorbonne University/Paris/France. In 2009, he was nominated at the IUF (French Research Institute) for a 5 years delegation position. He is currently CNRS scientific advisor for INS2I. His research interests include Computer Vision, Pattern Recognition and Machine Learning. He developed several interactive learning systems for content-based image and video retrieval. He is now focusing on Machine Learning for Multimedia Processing, Deep Learning for visual data recognition, and Computational cooking. M. Cord has published a hundred scientific publications, including several recently published on deep learning (NIPS, ECCV, ICCV, CVPR). He is involved in several French (ANR, CNRS) and international projects (European IP and NoE, Singapore, Brazil, Canada) on those topics.
Eduardo Valle
Eduardo Valle is professor at the Department of Computer Engineering and Industrial Automation — DCA of the School of Electrical and Computer Engineering — FEEC at the State University of Campinas — UNICAMP.
He was also a faculty member of the RECOD Lab (REasoning for COmplex Data). He got a Ph.D in Computer Sciences at the University of Cergy-Pontoise, in 2008. He got a M.Sc. and a B.Sc., also in C.S., at the Federal University of Minas Gerais, in 2003 and 2001 respectively. He works with a talented team of researchers and students on smart services for education and health, and large-scale machine learning.