Schedule for Fall 2015 - Kyoto University Informatics Seminar

Schedule for Fall 2015

Official course name: Perspectives in Informatics 5, Graduate School of Informatics

Unless noted otherwise, all talks Thursday 14:45 -16:15, Joho 2, Research Bldg. No.7 (総合7), Main Campus

    • Natasha Devroye(University of Illinois at Chicago), A touch of network information theory , hosted by D. Avis
    • Jakob Eriksson (University of Illinois at Chicago), Tracking your every move - today and tomorrow, hosted by D. Avis
    • Follows Monday schedule, no seminar today
    • Jean-François Aujol (Institut de Mathématiques de Bordeaux - CNRS), Mathematical modelling of textures, hosted by M. Cuturi
  • October 26 irregular schedule, talk on Monday
    • Klaus-Robert Müller (T.U. Berlin), Machine Learning & Big Data, hosted by M. Cuturi
    • Jean-Luc Rouas (Bordeaux Computer Science Research Laboratory, LABRI), Introduction to automatic speech processing and its application to language studies, hosted by M. Cuturi
  • November 05 irregular time slot, 17:00 - 18:30
    • Zonghua Zhang (Institut Mines-Télécom/TELECOM Lille), Towards Cyber Attacks Mitigation: from Cost-effective Security Hardening to Autonomic Cyberdefense, hosted by X. Liang
    • Endre Boros (Rutgers University), NIM, Co-NIM, and hypergraph-NIM, hosted by K. Iwama
  • November 26 irregular time slot, 16:30 - 18:00
    • Francois Le Gall (University of Tokyo), Overview of the Recent Progress on Matrix Multiplication, hosted by D. Avis
    • Rajeev Raman (University of Leicester), Encodings = (Data Structures) - (Data), hosted by K. Iwama
    • Ichiro Fujinaga (McGill University), The Research Program of the Distributed Digital Music Archives and Libraries Laboratory at McGill University, hosted by A. Jatowt

Detailed Information and Abstracts

October 1

14:45 -16:15, Joho 2, Research Bldg. No.7 (総合7), Main Campus
Natasha Devroye, University of Illinois at Chicago
Title: A touch of network information theory
Abstract: Information theory is a branch of applied mathematics, electrical engineering, and computer science involving the quantification of information. We provide a brief introduction to network information theory — the quantification of information transmitted in networks — and the types of questions posed and solved, with a particular focus on the capacity of communication networks. We outline some recent topics of interest to the presenter, including the capacity of interference, cognitive, two-way, and relay networks.

October 8

14:45 -16:15, Joho 2, Research Bldg. No.7 (総合7), Main Campus
Jakob Erikkson, University of Illinois at Chicago
Title: Tracking your every move - today and tomorrow
Abstract: Not too long ago, tracking the movements of individuals was an obscure activity largely reserved for detective novels and the occasional creepy stalker. Lately, however, massive-scale continuous location surveillance has quietly become a fact of life, pursued by organizations as diverse as Google, Amazon, the U.S. Drug Enforcement Agency, and the U.S. Department of Transportation, not to mention cyber-criminals, jealous spouses and helicopter parents. An equally wide range of technologies is used for this virtual stakeout job, including spyware on your laptop and mobile devices, roadside radio receivers (Wi-Fi, Bluetooth and more), license plate reading devices, face-recognizing surveillance cameras, RFID tags and readers, and more.

In this talk, we will review some of the more pervasive people-tracking methods in use today, together with related techniques and more (or less) well-known uses. We'll then put on a pair of decidedly rose-colored glasses, and try to see what good our Orwellian future may bring, and what challenges lie ahead, beyond the quaint notion of protecting your location privacy.

October 22

14:45 -16:15, Joho 2, Research Bldg. No.7 (総合7), Main Campus
Jean-François Aujol, (IMB, Université Bordeaux)
Title: Mathematical modelling of textures

Abstract: In this lecture, we will consider the modelling of textures from a mathematical point of view. Starting with the pioneer work of Y. Meyer in 2001, numerous functional analysis spaces have been proposed to model textures. We will revisit these spaces, and relate them to classical modelling of textures (periodic textures, oscillations, textons, mathematical morphology, …). The problem of image decomposition into a geometrical part and a textural part will serve as a basic application to test the different models.

October 26

IRREGULAR SCHEDULE : 16:30 -18:00, Joho 1, Research Bldg. No.7 (総合7), Main Campus
Klaus-Robert Mueller, (TU Berlin)
Title: Machine Learning & Big Data

Machine Learning has been instrumental in analysing data. This talk will first p rovide a brief introduction to machine learning and discuss the scaling challeng es for kernel methods and deep neural network approaches. The subsequent part wi ll touch selected ML applications in the Neurosciences ( in particular Brain Computer Interfacing), and if time permits, Physics etc..

Brain Computer Interfacing (BCI) - discussed in part II - aims at making use of brain signals for e.g. the control of objects, spelling, gaming and so on. Here BCI is seen from a machine learning and signal processing perspective; the wealth, the complexity and the difficulties of the data available pose a truely enormous challenge. In real-time a multi-variate very noise contaminated data stream is to be processed and classified. I report in particular about the Berlin Brain Computer (BBCI) Interface that is based on EEG signals and take the audience all the way from the measured signal, the preprocessing and filtering, the classification to the respective application. BCI communication is discussed in a clinical setting and for gaming.

Part III reports on recent work, where ML is applied to the exploration of chemical compound space and materials. Here the focus will be placed on the quest for better representations of molecules and solids.

October 29

14:45 -16:15, Joho 2, Research Bldg. No.7 (総合7), Main Campus
Jean-Luc Rouas, Bordeaux Computer Science Research Laboratory, LABRI, France
Title: Introduction to automatic speech processing and its application to language studies

Abstract: In this talk I will present what is automatic speech processing and give general information about the main applications. At first, I will explain what are the physiological specificities of the spoken voice, what are the possible levels of analysis and the difficulties that can be encountered according to the language of interest.

I will then detail how most speech processing systems are built around common parametrisation and supervised modelling schemes. I will illustrate this with some example of systems such as speech activity detection, speaker verification and language identification. I will then further detail how a speech recognition system works, starting with a brief historical review and giving some insights on recent advances.

At last, I will present some problems related with my on-going research and my main point of interest which is using speech technologies to improve linguistic and para-linguistic knowledge, more particularly on the prosodic aspects of speech.

November 05

17:00 -18:30, Joho 2, Research Bldg. No.7 (総合7), Main Campus
Zonghua Zhang, Institut Mines-Télécom, France
Title: Towards Cyber Attacks Mitigation: from Cost-effective Security Hardening to Autonomic Cyberdefense
Abstract: Instead of preventing attacks from occurrence, which is largely recognized to be mission impossible in practice, attack mitigation generally aims at lessening the impact of, if not completely ceasing, an inevitable attack based on its early detection. In other words, attack mitigation mechanisms need to ensure that the performance of network functions or services will be maintained at a satisfactory level despite the ongoing attacks. This talk is intended to share our experiences and lessons learned from the design of attacks mitigation mechanisms in different types of networks such as enterprise networks, wireless ad hoc networks, and software defined networking (SDN). Specifically, in enterprise networks, we show how the legacy security mechanisms like attack graph can be bridged with organization level security metrics to aid security administrators in taking cost-effective countermeasures. Also, we will discuss the potential of SDN to reshape the landscape of today’s defense mechanisms, by specifically exemplifying several implementations of traditional security mechanisms in SDN controllers.

November 19

14:45 -16:15, Joho 2, Research Bldg. No.7 (総合7), Main Campus
Endres Boros, Rutgers University
Title: NIM, Co-NIM, and hypergraph-NIM
Abstract: NIM is a well-known simple game, in which players alternate in taking away some positive number of stones form one of the heaps until no stones are left. The player who takes the last stone is the winner. Bouton (1901) analyzed and completely solved this game. The Sprague–Grundy theorem states that every impartial game under the normal play convention is equivalent (in some sense) to a NIM game. Sprague (1935) and Grundy (1939) proved this independently. What happens if the ``stones’’ are elements in an abstract set and the ``heaps’’ are subsets, which possibly intersect? What happens if a player’s move involves taking a positive number of stones from exactly k heaps for some fixed integer k? In this talk we address these questions, recall the SG-theory and extend it to a few generalizations of NIM, and prove some connections between the different variants. We close with open ends and some conjectures. (Joint work with V. Gurvich, N.B. Ho, K. Makino and P. Music.)

November 26

16:30 -18:00 (irregular time slot), Joho 2, Research Bldg. No.7 (総合7), Main Campus
Francois Le Gall, University of Tokyo
Title: Overview of the Recent Progress on Matrix Multiplication
Abstract: Until a few years ago, the fastest algorithm for matrix multiplication was the “Coppersmith-Winograd algorithm” designed in 1987. In 2010, Stothers gave an improvement to the algorithm, the first in 23 years. Further improvements have then been obtained in 2012 and 2014 by Vassilevska Williams and Le Gall, respectively. In this talk I will describe the long history of work on the complexity of matrix multiplication, and discuss these very recent improvements.

December 17

14:45 -16:15, Joho 2, Research Bldg. No.7 (総合7), Main Campus
Rajeev Raman, University of Leicester

Title: Encodings = (Data Structures) - (Data)

Abstract: Driven by the increasing need to analyze and search for complex patterns in very large data sets, the area of compressed and succinct data structures has grown rapidly in the last 10-15 years. Such data structures have very low memory requirements, allowing them to fit into the main memory of a computer, which in turn avoids expensive computation on hard disks.

This talk will focus on a topic that has become popular recently: encoding ``the data structure'' itself. Some data structuring problems involve supporting queries on data, but the queries that need to be supported do not allow the original data to be deduced from the queries. This presents opportunities to obtain space savings even when the data is incompressible, by pre-processing the data, extracting only the information needed to answer the queries, and then deleting the data. The minimum information needed to answer the queries is called the effective entropy of the problem: precisely determining the effective entropy can involve interesting combinatorics.

January 28

14:45 -16:15, Joho 2, Research Bldg. No.7 (総合7), Main Campus
Ichiro Fujinaga, McGill University

Title: The Research Program of the Distributed Digital Music Archives and Libraries Laboratory at McGill University

Abstract: The main goal of this research program is to develop and evaluate practices, frameworks, and tools for the design and construction of worldwide distributed digital music archives and libraries. Over the last few millennia, humans have amassed an enormous amount of information and cultural material that is scattered around the world. It is becoming abundantly clear that the optimal path for the acquisition of this material is to distribute the task of digitizing the wealth of historical and cultural heritage material that exists in analogue formats, which may include books, manuscripts, music scores, maps, photographs, videos, analogue tapes, and phonograph records. In order to achieve this goal, libraries, museums, and archives throughout the world, large or small, need well-researched policies, proper guidance, and efficient tools to digitize their collections and to make them available economically. The research conducted within the program addresses unique and imminent challenges posed by the digitization and dissemination of music media. In this talk various projects conducted at our laboratory will be presented; including large-scale optical music recognition, workflow management for automatic metadata extraction of LP recordings, creation of ground truth for structural and chord analysis of music, evaluation of digitization methods for analogue recordings, and digital prosopography of Renaissance musicians.