Helsinki Institute for Information Technology HIIT

1
Helsinki Institute for Information TechnologyHIIT

• Martti Mäntylä
• Professor, Research Director
• Esko Ukkonen
• Professor, Research Director

2
HIIT in a Nutshell

• Joint research institute of University of
  Helsinki and Helsinki University of Technology
  founded in 1999
• Operates through research programmes led by
  professor-level Programme Directors
• Programmes operate through projects covering
  basic research, co-funded research, and EU
  activities
• Partners include Finnish ICT and media companies,
  research institutions, international partners
• Directors Prof. Martti Mäntylä (Advanced Research
  Unit), Prof. Esko Ukkonen (Basic Research Unit)
• 15 research groups in TKK or UH
• Personnel (py) 112 researchers, 8 staff (2006)

3
HIIT Vision

• Ubiquitous and universal computation,
  communication and digital media bridge people,
  communities, businesses, sciences, objects,
  environment and society, thereby producing and
  utilising vast amounts of data on complex
  phenomena.

4
HIIT Mission

• Recognised as an internationally leading research
  institution, HIIT conducts basic and
  multidisciplinary research of modern information
  and communication technology which includes
  theoretical and technological aspects, as well as
  applications, and has high scientific, industrial
  and societal impact.

5
Research Programmes

• Algorithmic Data Analysis Data analysis methods
  for other sciences and for industry through both
  basic research in computer science and applied
  work on problems arising from applications
• Future Internet Internet infrastructure for
  efficient, secure and trusted always-on
  connectivity and service
• Network Society Human-centric multidisciplinary
  anticipation and development of ubiquitous
  information and communication technology
• Probabilistic Adaptive Systems Sophisticated
  probabilistic models and their applications for
  solving problems appearing in complex real-world
  stochastic systems

6
Organisation
7
Personnel 2000-2006
8
Publications 2000-2006
9
Financing 2006
10
Administration

• Board
• Members from universities, industry, staff
• Overall planning, budget, personnel
• Scientific Advisory Board
• Strategic planning of research
• Evaluation of research programs and groups
• Steering Group
• Strategic planning of research
• Admission of groups and projects to HIIT
• Management Group, Programme Management Groups
• Coordination and planning of joint activities

11
Algorithmic Data Analysis

• Heikki Mannila
• Academy professor
• HIIT, Helsinki University of Technology and
  University of Helsinki

Helsinki University of Technology University of
Helsinki
12
Mission

• To develop useful algorithmic data analysis
  methods for other sciences and for industry.
• Basic research in computer science and applied
  work on problems arising from applications

13
Why?

• Computational methods are changing the way
  science and industry work
• Data analysis is becoming increasingly important
• Our ability to measure complex systems is far
  greater than our ability to model and understand
  them
• ? Need for the development of new methods

14
Profile Working cycle

• Close interaction with application experts
• Formulating computational concepts
• Analyzing the properties of the concepts
• Designing algorithms and analyzing their
  performance
• Implementing and experimenting with the
  algorithms
• Applying the results in practice

15
Groups in the ADA programme

• Pattern and link discovery (H. Toivonen)
• Adaptive computing (P. Floréen)
• Combinatorial pattern matching (E. Ukkonen) ()
• Data mining theory and applications (H. Mannila)
• Statistical machine learning and bioinformatics
  (S. Kaski) ()
• Parsimonious modelling (J. Hollmén)
• 
  () joined recently
• Discrete algorithms, combinatorial methods,
  probabilistic modeling
• Excellent network of collaborators in other
  sciences
• Strong international aspect (collaborations,
  postdocs from abroad)

Univ. Helsinki
TKK
16
Some highlights

• ContextPhone
• String algorithms
• Gene regulation
• Sums of products
• Randomization methods for discrete data
• Sequence segmentation
• Finding orders from data
• Ecological and biological data analysis

17
Sums of products

• Original motivation computing recurrence risks
  in genetics other computational problems
• M. Koivisto, An O(2n) algorithm for graph
  colouring and other partitioning problems via
  inclusionexclusion. FOCS 2006 A. Björklund, T.
  Husfeldt, P. Kaski, and M. Koivisto Fourier
  meets Möbius fast subset convolution. STOC 2007.

A(A1,A2,...,Ak) runs through ordered
partitions of the n-element set
Sums of products occur in parameter estimation
of probabilistic models
18
Randomization methods for discrete data

• Randomization methods generate randomized
  versions of the data, run the algorithm on them,
  and compare with the results on the actual data
• How to generate datasets that have the same first
  order statistics (row and column counts) as the
  dataset?
• Doable (Gionis et al., KDD 2006) (runner-up, best
  paper)
• Segmentation randomization (BMC Bioinformatics
  2007)

19
Sequence segmentation

• How to segment a sequence into homogeneous
  segments?
• Genome structure telecommunications
  paleontology
• New concepts, algorithms, approximation results
• Applications isochore structure, context
  sensitivity in mobile communications,
  paleontology
• Publications ICDM 2001, ReComb 2003, ECCB 2002,
  KAIS 2006, SDM 2006, KDD 2006, Gene 2007,
  Paleobiology 2006, Oncogene 2006, BMC Bioinf. 2007

20
Finding orders from data

• Seriation problem for 0-1 data in paleontology
  Find an ordering of the rows such that the 1s are
  as consecutive as possible
• No errors ? polynomial errors ? NP-hard
• Spectral techniques MCMC finding partial orders
  (Paleobiology 2006 PLoS Comput Biol 2006 KDD
  2006 KDD 2005 PKDD 2007)

21
Vision

• Keep up the high quality
• Strong interaction with departments in TKK, UH,
  and other parts of HIIT
• Strong interaction with industry and other
  sciences
• About current size
• Keep a sufficiently sharp focus
• International recruiting and mobility
• Quality is the key world-leading research

22
Future Internet Program at HIIT

• Kimmo Raatikainen
• program director
• kimmo.raatikainen_at_hiit.fi

23
FI Team

• Director Prof. Kimmo Raatikainen
• wireless Internet group, mobile computing group
• Research co-ordinator Oriana Riva
• Seniors
• Adj. Prof. Patrik Floréen (adaptive computing
  group)
• Adj. Prof. Andrei Gurtov (networking research
  group)
• Dr. Arto Karila (networking research group)
• Univ. Lect. Markku Kojo (wireless Internet group)
• Prof. Jukka Manner (wireless Internet group,
  distributed networking and security group)
• Dr. Pekka Nikander (networking research group)
• Dr. Ken Rimey (distributed applications group)
• Adj. Prof. Sasu Tarkoma (mobile computing group,
  distributed networking and security group)
• Prof. Antti Ylä-Jääski (distributed networking
  and security group)

24
Research Groups in FI Program
Distributed Applications
Adaptive Computing
Mobile Computing
Distributed Networking and Security
Wireless Internet
Networking Research
25
Result Highlights

• Host Identity Protocol
• Contributions to IETFs RFCs
• Reference implementation for Linux
• Mobility middleware
• Contributions to W3Cs Efficient XML Interchange
• Intelligent synchronization of XML documents
• Contributions to WWRF Service Architecture
• Fuego open source implementation widely reused

26
Directions in Future Internet

• Three focus areas of research
• Security-Trust-Privacy
• dynamic requirements, chains of trust
• Mobile Always-on Connectivity
• reasonable solutions at each layer, cross-layer
  interactions, locator-identity split
• Scalable Open Service Architectures
• interoperability rapid, reliable and
  fault-tolerant service provision supporting
  context-awareness
• Solutions are sought in distributed algorithms
  and structures, middleware, and (Internet)
  protocols.

27
Directions in Future Internet

• Dual approach
• improving current Internet
• protocol enhancements HIP, TCP, DCCP, SIP, etc
• overlay networks
• secure push
• starting from clean table
• publish-subscribe paradigm
• applying microeconomics and game theory

28
Network Society

• Marko Turpeinen
• Program Director, Professor
• Giulio Jacucci
• Senior Research Scientist

29
Mission

• Human-centric multidisciplinary anticipation and
  development of ubiquitous information and
  communication technology, which is based on deep
  understanding of needs and practices of our
  everyday life and our social relations in a
  network society.

30
Groups and Selected Themes
Digital Economy Prof. Jukka Kemppinen, Dr. Perttu
Virtanen, Dr. Olli Pitkänen Virtual
consumerism Innovation and society IPR creative
commons
Ubiquitous Interaction Dr. Giulio Jacucci, Prof.
Martti Mäntylä Awareness cues Technology
ecologies Performative interaction
Digital Content Communities Prof. Marko
Turpeinen, Dr. Risto Sarvas Creative
communities Open platforms Remixing formats
31
Social Snapshot Photography
2003
2004
2006
2005
2007
Dr. Risto Sarvas receives the best computer
science thesis in Finland 2006 award
A more commercially oriented version of MobShare.
A full-blown commercial service with MTV3.
User-centric mobile phone photo sharing system.
Metadata-centric mobile phone photo system.
Immortalidad _at_HIIT with KCL, Futurice,
Yliopistopaino
MC2 _at_HIIT with Futurice
MC2 _at_HIIT with Futurice
MMM-1 _at_Berkeley with Futurice
32
Multitouch Interactive Screen
33
Awareness Cues

• In-depth intervention studies of teenagers and
  information workers, analyzing their
  interpretation and use of cues
• Oulasvirta, A., Petit, R., Raento, M., Tiitta,
  S. (2007). Interpreting and acting on mobile
  awareness cues. /Human-Computer Interaction/, 22
  (12), 97-135.

34
Key Research Domains (2009)

• 1. Mobile and ubiquitous interaction
• 2. Open media creation, management and
  distribution
• 3. Tools and methodology for service innovation
• 4. Development of a sustainable network society

35
Probabilistic Adaptive Systems
Aapo Hyvärinen Senior Research Scientist Petri
Myllymäki Programme Director, Professor
36
Probabilistic Adaptive Systems Groups

• Complex Systems Computation Group (CoSCo)
• Prof. Petri Myllymäki, program director
• Dr. Jorma Rissanen, HIIT Fellow
• Neuroinformatics Group
• Dr. Aapo Hyvärinen
• Statistical Machine Learning and Bioinformatics
  Group (from 2007)
• Prof. Samuel Kaski

37
Mission

• Background
• When automating intelligent behaviour modelling
  plays a central role
• Due to uncertainty and incompleteness of
  available information such models are commonly
  based on probabilities
• Goal
• theoretical understanding
• development of models
• which are
• probabilistic or information-theoretic
• computationally efficient
• with applications in engineering and science

38
Background Independent Component Analysis

• Linear decomposition of multivariate data
• Finds hidden directions, in contrast to classic
  PCA
• Based on non-Gaussianity of the data
• FastICA the most popular algorithm (Hyvärinen.
  IEEE Trans. NN, 1999)
• Standard reference book on the theory
  Independent Component Analysis by Hyvärinen,
  Karhunen, Oja, 2001. (Translated into Japanese
  and Chinese)

39
Non-Gaussian Bayesian Networks

• Non-Gaussianity enables learning network
  structure and weights in basic linear DAG case
  (Shimizu, Hoyer, Hyvärinen, Kerminen. J. Mach.
  Learn. Res., 2006)
• Enables inference on the direction of causality
• Extensions to, e.g.
• hidden confounding variables
• nonlinearities
• binary data
• Application to neuroscientific sensor data
  (functional brain imaging)

40
Theory of probabilistic modelling

• New principle for estimation of non-normalized
  statistical models (Hyvärinen. J. of Machine
  Learning Research, 2005)
• Minimum Description Length (MDL) Normalized
  Maximum Likelihood (NML)
• Conditional NML Universal Model (Rissanen and
  Roos, ITA, 2007)
• Linear-time algorithm for multinomial stochastic
  complexity (Kontkanen and Myllymäki, Inform.
  Proc. Letters, 2007)
• The first computationally feasible algorithm
  computing NML for tree-structured Bayesian
  networks (Kontkanen, Wettig and Myllymäki,
  EURASIP J. on Bioinform. and Sys. Biol., 2007).
• Applications clustering, histogram estimation,
  image denoising

41
Future Vision

• Paradigms in cognitive science / intelligent
  systems
• 1960s classic AI / symbolic cognitive science
• 1980s parallel distributed processing / neural
  networks
• 2000s probabilistic inference, possibly based
  on real sensory data
• Internet various sensors (e.g. in
  neuroscience) provide huge amounts of data
• Unconventional properties, e.g. nongaussianity
• Need for new analysis methods
• Spin-off data analysis methods to be applied
  anywhere