Activity Update from the MLSP TC

Contributed by Kenneth E. Hild II (TC Liaison to eNews)

The Machine Learning and Signal Processing Technical Committee (MLSP TC) would like to take this opportunity to share with you some recent news and work related to our TC activities.

2014 MLSP Workshop

First of all, we are excited to announce that next year's MLSP Workshop will be in Rheims (also spelled "Reims"), France. This city was very prominent during the height of the Roman Empire and was, for many years, the site of the coronation of French kings. It is also the location of the signing of the unconditional German surrender to end World War II. Given that it is the largest city in the region, many regard Rheims as the effective capital of the province of Champagne. Rheims is the home of the Holy Ampulla, Place Royal, Place Cardinal-Lucon, Rue de Vesle, Porte de Mars, Palace of Tao, and the Cathedral of Notre-Dame de Reims. We currently expect that the workshop will be held at or near the end of September, 2104, and the paper deadline will be somewhere between the middle of March and early April, 2014.

MLSP TC meeting at ICASSP

Each year, the MLSP TC meets at the MLSP Workshop and at ICASSP. Many topics were discussed during the most recent TC meeting at ICASSP. The most notable topics included: the 2013 MLSP Workshop, Reproducible Research, the International Conference on Latent Variable Analysis (LVA), and changes to several MLSP TC committees.

Saeid Sanei reported on the MLSP 2013 Workshop, which took place on September 22-25, 2013, in Southampton, United Kingdom. Overall, the workshop was a great success. A total of 102 papers were accepted for publication. The plenaries were given by Sergios Theodoridis (University of Athens, Greece), Christian Jutten (University Joseph Fourier of Grenoble, France), and Simon Godsill (University of Cambridge, United Kingdom). The workshop hosted two tutorials, which were presented by Cédric Richard (University of Nice Sophia-Antipolis, France) and Lieven De Lathauwer (Katholieke Universiteit Leuven, Belgium). Also, the LVA Conference was held as a special track of the MLSP Workshop.

Mark Van Hulle led a discussion on Reproducible Research, which is something the MLSP TC wishes to foster.

THE MLSP TC also discussed our desire to and the possibility of continuing our collaboration with the LVA community. Specifically, we hope to continue to hold the MLSP Workshop and the LVA Conference together.

There have been a few changes to the subcommittees within the MLSP TC. The Election Subcommittee now consists of: Cedric Fevotte and Vincent Tan. Likewise, the Awards Subcommittee now consists of Asoke Nandi, Bhaskar Rao, and Ali Sayed.

MLSP nomination for new members

We recently held an election for adding new members to the MLSP TC. There were a total of 18 candidates and 10 positions to be filled. Unlike in past years, this time we used Single Transferable Vote, which is a standard multi-winner voting procedure based on ranking. The outcome of the voting will be presented later since the outcome was not available at the time this article was written.

2013 Data Analysis Competition

The MLSP TC hosts an annual Data Analysis Competition, which takes place during each workshop. The last competition, which finished recently, involved bird classification. Participants were asked to develop classification algorithms to reliably identify the set of bird species in real-world audio data collected in an acoustic monitoring scenario. An MLSP record number of 79 teams entered the contest. The Data Analysis Competition Committee wrote one short summary paper and they invited and subsequently accepted for publication (in the MLSP 2013 conference proceedings) companion papers from the top two teams.

Decision tree ensembles were the most popular type of classifier used and these methods generally achieved high accuracy. More interesting variations were displayed in the feature design part of the problem. Based on results of the top-ranking teams, some of the most useful features are: the histogram of segments provided in the baseline, template matching features, signal descriptors not based on segmentation, and location. Alternatively, convolutional neural nets also achieved excellent results, but do not need extensive feature engineering.

In addition to a record number of participating teams, we received highly effective solutions. Specifically, a total of 45 teams achieved AUC (Area Under the ROC Curve) greater than the benchmark performance AUC of 0.85 provided by the researcher who originally collected the data. The methods applied by the top two teams achieved AUCs of over 0.95. The first place team consists of Gabor Fodor from Budapest University of Technology and Economics in Budapest, Hungary. The second place team consists of Hong Wei Ng and Thi Ngoc Tho Nguyen from Advanced Digital Sciences Center, University of Illinois at Urbana-Champaign, and Singapore, respectively.

The prizes for the first and second place teams were $700 ($500 cash and $200 book gifts) and $600 ($450 cash and $150 book gifts), respectively, and included registration fee waivers. The first place team, the second place team, and the 7th place team had poster presentations at the 2013 MLSP Workshop.

Awards/Appointments

Jerónimo Anrenas-Garcia was recently appointed as an Associate Editor for the IEEE Signal Processing Letters. He is also the President of the IEEE Spanish Joint Signal Processing and Communications Chapter and, in this capacity, he helped to organize visits from two Designated Lecturers this year: Tulay Adali and John V Mathews.

Andrzej Cichocki was recently appointed as an Associate Editor for the IEEE Transactions on Signal Processing. He is also an Associate Editor for the Journal of Neuroscience Methods and the Journal of Computational Intelligence and Neuroscience. Andrzej was a Guest Editor of a special issue of IEEE Signal Processing Magazine, which was devoted to Source Separation, and a key note speaker for SISA 2013, where his talk was entitled, "Era of Big Data Processing: A New Approach via Tensor Networks and Tensor Factorizations". Finally, Andrzej was recently elevated to IEEE Fellow status.

Weifeng Liu was appointed as an Associate Editor for the IEEE Signal Processing Letters and the IEEE Transactions on Neural Networks and Learning Systems.

Tulay Adali gave a tutorial at the 18th Scandinavian Conference on Image Analysis, at Espoo, Finland and she gave the keynote speech at the 2013 Conference on Signal Processing and its Applications, which took place at Kyrenia, Cyprus. Both talks were on ICA and its recent generalization to multiple datasets, which is known as Independent Vector Analysis (IVA). Tulay also co-organized, along with Klaus Muller and Mark Cohen, a workshop on Multimodal Imaging at the Institute For Pure and Applied Mathematics, located at UCLA.

Bhaskar Rao was selected as a Distinguished Lecturer for 2014.

Atsushi Nakamura is now the IEEE SPS chair of the Kansai Chapter.

Ignacio Santamaria is the co-chair of the 3rd Workshop on Physics-Inspired Paradigms in Wireless Communications and Networks (PhysComnet 2014), which will take place in Hammamet, Tunisia in May, 2014.

Ali Sayed received a Technical Achievement Award, SPS Best Paper, and gave plenary lectures at both ICASSP 2013 and EUSIPCO 2013.

Paris Smaragdis gave a keynote lecture at the IEEE Workshop on Applications of Signal Processing to Audio and Acoustics (IEEE WASPAA) on using machine learning for audio signal processing.

We are also pleased to see that many of the Guest Editors of special issues in the IEEE Signal Processing Magazine are past or present members of the MLSP TC. For example, the Guest Editors for the IEEE Signal Processing Magazine Special Issue on Advances in Kernel-based Learning for Signal Processing includes: Klaus-Robert Müller, Tulay Adali, Kenji Fukumizu, Jose C. Principe, and Sergios Theodoridis. As a second example, the Guest Editors for the IEEE Signal Processing Magazine Special Issue on Source Separation and Applications includes: Tulay Adali, Andrzej Cichocki, Christian Jutten, Eric Moreau, and Arie Yeredor.