Helsinki Institute for Information Technology HIIT

1
Helsinki Institute for Information Technology HIIT

• Esko Ukkonen
• Director of the Basic Research Unit

Department of Computer Science HIIT Basic
Research Unit
2
HIIT Mission and research profile

• Joint research institute of University of
  Helsinki and Helsinki University of Technology
  (TKK)
• Goals
• strategic research in information technology and
  related topics, aiming at high scientific,
  industrial, and societal impact
• strong and attractive research environment
  joining the forces of the two universities
• Main research themes mobile computing, user
  experience, intelligent systems, semantic
  Internet, societal media, digital economy,
  adaptive computing, theory and applications of
  data mining, computational neuroscience

3
Organization

• Two research units with common Board, Scientific
  Advisory Board, and Industrial Advisory Board
• Senior researchers of HIIT typically have
  positions also in one of the departments of
  computer science of the host universities
• No permanent positions

4
Organization (cont)

• Advanced Research Unit (ARU) 1999 ?
• 2-3 year industry co-funded strategic research
  projects, typically funded by Tekes
• Located primarily in Ruoholahti
• Basic Research Unit (BRU) 2002 ?
• Long-term research of computer science in areas
  relevant to other sciences and to industry
• Located in the premises of the departments of
  computer science of the University of Helsinki
  and Helsinki University of Technology
• Basic funding from University of Helsinki

5
The Basic Research Unit (BRU) of HIIT is
evaluated in this assessment

• the evaluation of Computer Science Department
  covers all university-based groups that have
  activity in HIIT
• HIIT-BRU is also evaluated separately

6
Basic Research Unit (BRU)

• established 2002
• basic funding from UH
• main location at the premises of CS Dept of UH
• activities also on Otaniemi campus of TKK, CS
  Dept
• infrastructure of CS Dept
• Director Heikki Mannila 2002 8/2004 /
  Esko Ukkonen 9/2004

7
Mode of operation

• high-quality basic research of computer science
  on areas that have application potential in other
  sciences or in industry
• collaboration between universities
• close co-operation with CS departments
• strong participation in teaching
• international networking, international recruiting

8
Personnel
9
Publications
10
Funding
11
Research programme

• Theory and applications of data mining
• Academy Professor Heikki Mannila
• Professor Hannu Toivonen
• Neuroinformatics
• Dr. Aapo Hyvärinen
• Adaptive computing
• Dr. Patrik Floreen
• Professor Hannu Toivonen

12
Strategy What application areas?

• More than enough of good application areas and
  interested application research groups
• What is a good application area?
• Important on its own
• Requires computational advances
• Algorithmic component
• World-class excellence in collaborating groups
• Analogous criteria for industrial co-operations

13
Future vision

• strong international center in computational data
  analysis
• innovative multi-disciplinary research programmes
• senior researcher positions
• internationally leading scientific status
• recruiting
• networking

14
(No Transcript)
15
(No Transcript)
16
Data Mining Theory and Applications

• Heikki Mannila

Department of Computer Science HIIT Basic
Research Unit
17
Goals and approach

• Develop novel data analysis techniques for the
  use of other sciences and industry
• Why?
• How? Combine basic research in computer science
  with applications
• Look at data analysis problems arising in
  practice
• Abstract new computational concepts from them
• Develop new computational methods
• Analyze complexity
• Implement
• Take the results into practice
• Theoretical work in algorithms can have fast
  impact in the application areas
• The applications feed interesting novel questions
  to theoretical research in computer science

18
Structure of the data mining group

• Seniors (2) Heikki Mannila, Hannu Toivonen
• Postdocs (8) Aristides Gionis, Panyiotis
  Tsaparas, Ella Bingham, Marko Salmenkivi, Mikko
  Koivisto, Saara Hyvönen, Taneli Mielikäinen,
  Petteri Sevon
• Recruiting from abroad (Stanford, Toronto
  Purdue) postdoc education
• 8 Ph.D. students
• Funding HIIT BRU, Academy of Finland, Tekes,
  companies, other departments
• Group belongs to the FDK Center of Excellence
• Everybody teaches, everybody does research
• Very good international visibility

19
Themes in methods

• Algorithms and concrete complexity
• Data mining pattern discovery and probabilistic
  modelling
• Combinatorial pattern matching
• Methods for sequential data

20
Theory and applications

• Theory and practice interact a lot!
• Most theoretical directions are motivated by
  practical issues
• Application areas selected by theoretical
  interest

21
Application areas

• Genome structure
• Gene mapping
• Ubiquitous computing (adaptive computing)
• Palaeontology, ecology, paleoecology
• Climate studies
• Linguistics

22
Application partners

• Genome structure and gene mapping
• Leena Peltonen (KTL), Juha Kere (Karolinska), Anu
  Jalanko (KTL), Irma Thesleff (Institute of
  Biotechnology), Orion Pharma, GeneOS, Jurilab
• Linguistics
• R.-L. Pitkänen (Research Center for the Languages
  of Finland) Terttu Nevalainen (Department of
  English)
• Paleontology
• Mikael Fortelius (Dept. of Geology), Jukka
  Jernvall (Institute of Biotechnology)
• Climate studies
• Markku Kulmala (Dept. of Physics)
• Telecommunications (Nokia Research Center)

23
Interaction with society

• Direct impact of research in computer science
• Industrial collaborations
• Industry ?? university interchange of personnel
• Impact through the applications

24
Recent highlights

• Segmentation problems
• How to understand the structure of the genome
• Definition of the (k,h)-segmentation problem
• Algorithms with constant factor approximation
  guarantees (ReComb 2003)
• Very good practical performance
• New views on genome structure
• Ongoing practical work
• Recent very general theorem

25
Recent highlights

• Seriation problems in paleontology ? novel
  algorithmic techniques ? reinterpretation of
  certain fossil sites (Paleobiology, in press)
• Algorithmic techniques
• spectral ordering, combinatorial search over
  permutions, finding partial orders that fit the
  data

26
Recent highlights

• D. Gunopulos, R. Khardon, H. Mannila, S. Saluja,
  H. Toivonen, and R.S. Sharma. Discovering all
  most specific sentences. ACM Transactions on
  Database Systems 28 (2) 140-174, June 2003.
• F. Geerts, H. Mannila, E. Terzi Relational
  link-based ranking . The 30th International
  Conference on Very Large Data Bases (VLDB'04) ,
  2004.
• A. Gionis, H. Mannila, P.Tsaparas, Clustering
  aggregation, 21st International Conference on
  Data Engineering (ICDE) 2005.
• A. Gionis, A. Hinnenburg, S. Papadimitriou, P.
  Tsaparas, Dimension-induced clustering, 11th
  International Conference on Knowledge Discovery
  and Data Mining (KDD 2005 )

27
Recent highlights

• ACM SIGKDD Innovation Award 2003
• Principles of Data Mining (Hand, Mannila, Smyth,
  MIT Press 2001)

28
Collaborations and publications

• Publications with CS people from
• University of California at Irvine, University of
  California at Riverside, Stanford University,
  Microsoft Research, Universität Freiburg, Tufts
  University, University of Illinois at
  Urbana-Champaign, University of Minnesota,
  Technical University of Athens, ATT Research,
  University of British Columbia, University of
  Memphis, Technion, Nokia Research Center,
  University of Antwerpen, TU München, University
  of Wales, Oxford University, New Jersey IT, RPI
  (NY), Carnegie-Mellon University
• Conference publications in 2004
• SIGMOD, PODS, VLDB, KDD, PSB, ICDE, ICDM, PKDD,
  EDBT,
• Journals
• ACM TODS, IEEE TKDE, Information Retrieval, IEEE
  Pervasive Computing, Bioinformatics, Annals of
  Human Genetics, Ecology, Quaternary Science
  Reviews, Br J Haematol, Leukemia, CACM,
  Ecological Applications,

29
Connections networking

• Editorial board memberships
• ACM TODS
• IEEE TKDE
• Data Mining Knowledge Discovery
• Program Committee activities
• ACM SIGMOD
• ACM PODS
• ACM SIGKDD
• ICDM
• SIAM DM
• ISMB,
• EU projects
• April
• IQ
• networks

30
Vision

• Theory and applications
• Fundamental issues in computation
• Strong focus on 35 application areas
• More industrial applications
• Good application partners
• Publications software actual use
• Strong international recruiting, excellent
  networking
• Internationally leading unit in its area

31
Personnel

• Seniors
• Mannila, Heikki
• Toivonen, Hannu
• Postdocs
• Gionis, Aristides
• Tsaparas, Panayiotis
• Salmenkivi, Marko
• Koivisto, Mikko
• Bingham, Ella
• Hyvönen, Saara
• Mielikäinen, Taneli
• Sevon, Petteri

• Ph.D. students
• Eronen, Lauri
• Heino, Jaana
• Hintsanen, Petteri
• Haiminen, Niina
• Laasonen, Kari
• Terzi, Evimaria
• Leino, Antti
• Kollin, Jussi

32
Past personnel and postdocs

• Previous postdocs
• Geerts, Floris 09/200204/2004
• Goethals, Bart 01/200309/2004
• Hinneburg, Alexander 04/200404/2005
• Ollikainen, Vesa 01/200212/2002
• Onkamo, Päivi 11/200212/2004
• Vasko, Kari 04/200212/2003
• Visitors etc.
• Muilu, Juha 09/200212/2004 (20)
• Afrati, Foto 10/200411/2004
• Jaeger, Manfred 03/200205/2002, 11/200212/2002
  06/200006/2003
• de Raedt, Luc 03/200203/2002, 06/200006/2003
• Zaki, Mohammed, J. 05/200306/2003
• Papadimitriou Spiros 08/200410/2004
• Sood, Kismat 01/200312/2003

33
(No Transcript)
34
Adaptive Computing Systems

• Patrik Floréen

Department of Computer Science HIIT Basic
Research Unit
35
Mission and Approach

• Mission Research on multi-optimisation problems
  for resource-constrained distributed environments
• Context-aware mobile systems ad hoc and sensor
  networks
• Methodological research and software development
• Methods Combinatorial algorithms, probabilistic
  modelling techniques and component-based software
  development

36
Structure

• Started 2003
• Funding from the Academy of Finland, Nokia and EU
  IST
• Collaboration in-house and with several
  organisations
• TKK, MIT, Universität Kassel, Nokia, Ericsson,
  Suunto, Vaisala, NEC, Telematica Instituut,
  Fraunhofer FOKUS, Finnish Meteorological
  Institute,

37
Present ACS Researchers

• Prof. Hannu Toivonen
• Kari Laasonen
• Renaud Petit
• Mika Raento
• Dr. Patrik Floréen
• Petteri Nurmi
• Michael Przybilski
• Jukka Suomela
• In addition Dr. Greger Lindén Coordinator of
  PROACT

38
Themes and Results

• Context reasoning process and algorithms
• On-device location recognition algorithm
  (Pervasive 2004)
• Structuring of context reasoning process for
  activity recognition
• Software architectures and software development
• Distributed context management framework based on
  software components
• Implementation work on Symbian
• Algorithms for ad hoc and sensor networks
• Multicast lifetime maximisation under energy
  constraints (IEEE JSAC 2005)
• Balanced data gathering (Theor. Comp. Sci. 2005)

39
Highlight ContextPhone

• ContextPhone software in use at MIT (100
  persons), U. Berkeley, U. Oslo and University of
  Art and Design Helsinki
• Presented in IEEE Pervasive Computing 4 (2005) 2,
  p. 51-59

40
Future Vision

• Continue on path taken, guided by our group
  strategy
• Data-centric view quality and quantity of data
• Example how to construct and operate a sensor
  network to ensure minimum error in the data
  forwarded and in inferences drawn from that data
• Computational requirements and efficiency
• Example how to minimise the data and resources
  required in a context-aware system to recognise
  the activities of a user and predicting the
  users next actions
• Continued industrial collaboration

41
(No Transcript)
42
(No Transcript)
43
Neuroinformatics

• Aapo Hyvärinen

Department of Computer Science HIIT Basic
Research Unit
44
Mission and goals

• Neuroinformatics means intersection of
  information-processing sciences and neuroscience
  
• Three interconnected goals
• Build computational models of information
  processing in the brain
• Understanding the brain
• Development of "intelligent" computer algorithms
• Develop data analysis methods for neuroscience
• Advance general theory of multivariate
  computational data analysis

45
Structure

• People
• Leader Aapo Hyvärinen
• Postdocs Patrik Hoyer, Jarmo Hurri
• PhD students Urs Köster, Ilmari Kurki, Jussi
  Lindgren, Jukka Perkiö (50)
• Visitors Shohei Shimizu, Michael Gutmann, Asun
  Vicente
• Moved from TKK ( Helsinki U of Tech) in 2003
• Partners
• Neuroscience Dept of Psych (UH), Brain imaging
  centers (Naples, Maastricht, Würzburg,
  TKK, Oulu)
• Theory TKK, Osaka U, U of Alicante
• Networking Canadian Inst. of Adv. Res. (G.
  Hinton)

46
Core competence

• Unsupervised statistical multivariate models with
  nongaussian factors
• Starting point independent component analysis
  (ICA)
• Statistical models of early visual processing in
  the brain
• Adapted to statistical structure of natural images

47
Highlights

• At TKK Book Independent component analysis
  (Hyvärinen, Karhunen Oja, Wiley, 2001 in
  Japanese 2005)
• Models of statistical structure of natural
  images prediction of cells properties in new
  areas, i.e. predictive computational
  neuroscience (BMC Neurosci., 2005)

48
Highlights

• At TKK Book Independent component analysis
  (Hyvärinen, Karhunen Oja, Wiley, 2001 in
  Japanese 2005)
• Models of statistical structure of natural
  images prediction of cells properties in new
  areas, i.e. predictive computational
  neuroscience (BMC Neurosci., 2005)
• Stability/reliability analysis of ICA
  (NeuroImage, 2004)
• New principle for estimation of non-normalized
  statistical models (J. of Machine Learning
  Research, 2005)
• Linear nongaussian structural equation model for
  causal discovery (Proc. Uncertainty in AI, 2005)

49
Future vision

• Paradigms unifying computer science and
  neuroscience
• 1960s classic AI / symbolic cognitive science
• 1980s parallel distributed processing / neural
  networks
• 2000s probabilistic inference based on real
  sensory data
• Brain imaging methods produce huge amounts of
  data
• need for analysis methods
• nongaussianity prominent, as in ICA
• Spin-off data analysis methods to be applied
  anywhere