IEEE International Workshop on
Environmental Acoustic Data Mining (EAD)


held in conjunction with
2015 IEEE International Conference on Data Mining (ICDM'2015),
Atlantic City, USA, November 14, 2015

WORKSHOP EAD

This workshop aims to bring together researchers and professionals from worldwide academia and industry for showcasing, discussing, and reviewing the whole spectrum of scientific and technological opportunities, challenges, solutions, and emerging applications in environmental acoustic data mining. We also encourage original work based on interdisciplinary research, such as computer science and ecology, where quantitative evidence is available demonstrating the mutual advantage of such an approach.

These data mining methods now become the key among others for recommendation, alerting, management of natural resources. Providing analysis, such as tracing, propagation, visualization or simulation, new computational approaches focus on representing, analyzing, and extracting useful pattern from them.

SCOPE

This workshop focuses on environmental acoustics data mining that has advanced significantly these years due to the prevalence of the biodiversity researches in long term autonomous sound recording all over the world, in deep forests, undersea, in lakes etc., as well as in habitats with different levels of anthropization, including agricultural lands and urban areas.



During the four sessions of the workshop, the large scale data mining methods will be investigated for soundscape or bioacoustic pattern detection and classification, at either low and high signal to noise ratio. Thus, the objectives are two folds : (a) to produce fast indexing with supervised or unsupervised data mining of complex bioacoustic patterns, (b) to propose summary / abstraction / overview of soundscape structures as theoritical network or flow analyses which is required for biodiversity monitoring.

We will focus on methods scaled to environmental survey using passive acoustics, and on the design of methods for accurate mid-level or high level features detection/classification based on advanced signal decomposition, compressed sensing for large scale analyses, Deep Neural Net for accurate classification, as well as methods for real-time spatial tracking.

Illustrations will be given ranging from cetaceans to birds songs, bats to dolphins biosonars and other animals from deep forest and abysses. Biodiversity analysis and environmental protection projects are some of the direct outcomes of these algorithms.



Paper submissions are welcome on the topics below.
Moreover we run an original data mining challenge on sound recordings from two Natural Reserves in Italy. This challenge will be a demonstration of a concret data mining paradigm, it aims to help in developping standards that are required to define the methodologies to be used in this rapidly expanding, but still unexplored, data science area that brings together very different disciplines. The challenge will have a specific session for presentation of keynote papers.

TOPICS

Topics of Interest (but not limited to) on Environmental Acoustic Data mining (EAD) :

CHALLENGE

A challenge is distributed in EAD ICML workshop on the topic about categorization of terrestrial soundscapes to be applied on PetaBytes of continuous recordings. Demonstration will be given during the workshop on a sample from a 5 TB of these recordings of several months in Sabiod project with CIBRA. Each challenger will be enabled to submit one run to be published on a web page with its own scores and the weekly scores of the other challengers and the max pooler. The final evaluation is based on mean average precision metrics computed on the similarity scores on the whole set with 11 categories. The script and the metrics used for the evaluation will be distributed.

Definition of the classes

Format of the run

For each category and each file, the participant has to produce a table with similarity score between 0 and 1, (1 for high similarity); the starting time and the cumulative estimated duration of the category. Thus each prediction item (i.e. each line of the file of 2000*11 lines * 5 columns) has to respect the following .csv format:
where the probability of the event of the ClassId to be true (a real value between 0 and 1). If the probability of the event = 0, then the 'start time' and 'cumulative duration' = 0.

The final ranking of the challengers will be based on the Mean Average Precision on the 11 classes.
Each submitted run will be completed with a short readme text file with few words depicting : 'signal representation for each class', 'detector definition for each class', 'CPU time for each class with precision of the machine type'.
Challengers are invited to send a working note (4 pages) describing their method to be published in the IEEE proceedings.

CALL FOR PAPER / SUBMISSION

Paper on any of these topics are welcome. The paper review will be consistent with standard double-blind practice with rigorous peer reviewing by at least 3 peer reviewers. Papers will be selected based on their originality, significance, relevance, technical contents, and clarity of presentation.

Full paper submissions should be limited to a maximum of 10 pages, and follow the IEEE ICDM format.

Keynote paper submissions (on the challenge) should be limited to a maximum of 8 pages, and follow the IEEE ICDM format.

More detailed informations are available in the IEEE ICDM 2015 Submission Instructions.

Please submit your manuscript through the DMS 2015 submission site.

All accepted papers or keynote papers will be included in the ICDM'15 Workshop Proceedings published by the IEEE Computer Society Press. Therefore, papers must not have been accepted for publication elsewhere or be under review for another workshop, conferences or journals.

IMPORTANT DATES

WORKSHOP ORGANIZATION

Workshop Chair

Program Committee

Program

This workshop is planned to be a half-day event, including a keynote and two oral sessions. The tentative technical program is as follows:

SHORT BIO. OF CHAIR MEMBERS

Prof. Hervé Glotin

Hervé Glotin is a Professor at the Insitut Universitaire de France and Univ. of Toulon, in the Systems & Information Sciences CNRS lab. He is leading the Scaled Acoustic Biodiversity Project for the Big Data French National Research Concil (http://sabiod.org). He received a diploma in computer science from University Pierre et Marie Curie-Paris and carried out his PhD at the Inst. of Perceptual Artificial Intelligence (IDIAP), CH and Inst. of Spoken Communication - Perception Team Grenoble on "Robust adaptive multi-stream automatic speech recognition using voicing and localization cues". In 2000 he was involved as an expert at the Johns Hopkins CSLP lab with the IBM human language team in audiovisual Large Vocabulary Speech Recognition. After two years as a research engineer at CNRS lab on phonology and Semantic analysis, he became an assistant professor at the University of Toulon in 2003.

His research focuses on multimodal pattern analysis and retrieval systems, audiovisual indexing, cognitive models and bioacoustics. He is the co-author of one hundred of international refereed articles, and of an international (US, CANADA...) patent on a real-time bio-acoustic indexing algorithm. He was the general chair of ICML 2013, NIPS2013 and ICML2014 worshops on machine learning for bioacoustics ( http://sabiod.org/events.html ).
Recent publications :

Prof. Gianni Pavan

Professor of Ecology at the University of Architecture of Venice (1994-2006), Professor of Bioacoustics at the University of Pavia (2006 – now), President of the Bioacoustic and Environmental Research Interdisciplinary Center of the University of Pavia. Started working on bioacoustics in 1980, expert in computational bioacoustics, develops software and equipment for acoustic monitoring of terrestrial and marine habitats; after many years of work on marine mammals, he is now mainly involved with the study of terrestrial soundscapes.
Recent publications :

Dr. Peter Dugan

Peter Dugan is currently the PI on the National Oceanic Partnership (NOPP) Grant focusing on detection, classification and localization of marine mammals. He received his PhD in Electrical Engineering and Combined behavioral biology from Binghamton University in NY. Prior to Cornell University he held positions in the industry in companies such as Hughes Link Flight Simulation and Lockheed Martin. He also has an extensive publication and patent portfolio showcasing advanced methodologies in machine learning for marine mammal vocalizations. His interests and motivations include the research and development of computationally intelligent systems, by combining traditional "shallow systems" with "deep learning systems" for object detection and classification in order to enhance system accuracy. The NOPP grant has been awarded 1M$ for the years 2012-2015. As the PI, his goal is to investigate new approaches and deliever comparative studies working on integrated teams representing Science, Technology, Engineering and Mathematics (STEM).

Dr. Dugan is the head of the data analytic system at the Cornell Bioacoustics Research Program (BRP).  His current work focuses on using parallel and distributed computer networks for processing large quantities of continuous sound data using advanced detection-classification algorithms. This project funded through ONR, combines the application of high-performance-computing system called the acoustic data accelerator (HPC-ADA) to explore the spatio-temporal dynamics for a suite of acoustically active marine mammals (fin, humpback, minke, and right whales). Mechanics of the HPC-ADA will be discussed along with how distributed processing is tackling large datasets and high sample rates (200 kHz). The results yield insights into acoustic behavior for marine mammals with a goal to better help understand marine ecology for large cetaceans.
It has been published into :

Dr. Zhong-Qiu Zhao

Dr. Zhong-Qiu Zhao is an associate professor at Hefei University of Technology, China. He obtained the Master’s degree in Pattern Recognition & Intelligent System at Institute of Intelligent Machines, Chinese Academy of Sciences, Hefei, China, in 2004, and the PhD degree in Pattern Recognition & Intelligent System at University of Science and Technology, China, in 2007. From April 2008 to November 2009, he held a postdoctoral position in image processing in CNRS UMR6168 Lab Sciences de l’Information et des Systmes, France. From Jan. 2012 to Dec. 2014, he held a ‘Hongkong Scholar’ research position in pattern recognition at the Department of Computer Science of Hongkong Baptist University, Hongkong, China. Now he works in Laboratory of Data Mining and Intelligent Computing, Hefei University of Technology, China. His research is about pattern recognition, For multimodal data, including environmental data.
Recent publications :

LINKS

Past Workshops

CONTACT

Hervé GLOTIN
Université de Toulon - France
Tel. (+33) 04 94 14 28 24
Email: glotin@univ-tln.fr