Download
1 / 14
140 likes | 289 Views
Complex Systems Computation Research Group (CoSCo). Professor Petri Myllymäki Helsinki Institute for Information Technology HIIT Petri.Myllymaki@hiit.fi. Outline. The CoSCo group Research objectives Examples of past successes Highlights of ongoing research Future directions.
E N D
Complex Systems Computation Research Group (CoSCo) Professor Petri Myllymäki Helsinki Institute for Information Technology HIIT Petri.Myllymaki@hiit.fi
Outline • The CoSCo group • Research objectives • Examples of past successes • Highlights of ongoing research • Future directions
CoSCo Group (cosco.hiit.fi) • Director: Professor Petri Myllymäki • Senior researchers: • Dr. Wray Buntine • HIIT Fellow Jorma Rissanen • Nokia Research Fellow Henry Tirri • some 20 researchers, mostly quite experienced • a good combination of theoretical competence with top-level programming skills • International recruits: • Dr. Wray Buntine (2002 ) • Dr. Huizhen Yu (MIT, 2005 ) • Active collaboration with: • UCB (Michael Jordan) • CWI (Paul Vitanyi, Peter Grünwald) • Tsinghua University (Lizhu Zhou)
Our mission • Investigate computational issues related to complex systems, focusing on prediction and model selection tasks. • Develop theoretically and methodologically elegant solutions to practically important problems related to complex systems. • A complex system is a collection of interacting agents, elements or processes whose collective behavior exhibits interesting large scale patterns. Such systems can be found in various disciplines, including economics, computer science, mathematical biology and physics. • Examples of complex systems: • The Internet • WLAN and GSM networks • Social networks • Sensor networks • Models for complex systems: • Bayesian and causal networks • Markov Random Fields • Markov decision processes • ... and other probabilistic graphical models • “Intellectual home conferences”: • Uncertainty in AI (4 Cosco members in the program committee) • AI and Statistics (3 Cosco members in the program committee)
B-course Data Analysis Server(http://b-course.hiit.fi) • commercial-quality ASP tool for research and teaching purposes • 10 000 users world-wide in 2 years • Sampsa Hautaniemi, Henrik Edgren, Petri Vesanen, Maija Wolf, Anna-Kaarina Jarvinen, Olli Yli-Harja, Jaakko Astola, Olli Kallioniemi and Outi Monni, A Novel Strategy for Microarray Quality Control Using Bayesian Networks. To appear in Journal of Bioinformatics. • Jarvis, Smith, Wada, Rivas, McElroy, Smulders, Carninci, Hayashizaki, Dietrich, Wu, McConnell, Yu, Wan, Hartemink, Lin, A framework for integrating the songbird brain. J Comp Physiol A (2002) 188: 961-980. • K Deforche, K Van Laethem, A Abecasis, J Snoeck, AP Carvalho, I Derdelinckx, P Gomes, J Cabanas, MA Soares, RM Brindeiro, A Tanuri, R Camacho and AM Vandamme, Bayesian Network Reveals Linkage for Mutations at Position 89 of Hiv-1 Protease to other Protease Codons and Therapy with Protease Inhibitors. Proceedings of the 2nd European HIV Drug Resistance Workshop, March 11th-13th, 2004, Rome.
Ekahau Positioning Engine • Software for locating devices in Wi-Fi networks • European Union: The European Information Society Technology Prize 2002. • Technology Marketing Corporation (TMC): Best product of the year 2002. • Planet PDA, the Global Summit on Enterprise & Custom Volume Handheld Computing: Best of show. • Software Industry Summit: Best commercialized innovation in Finland in 2002. • SearchNetworking.com: Bronze medal, best product of the year 2003 (enterprise wireless applications and systems). • Wi-Fi Planet 2004: Best of Show.
Koptimi BayMiner • Intelligent algorithms for constrained bin-packing • In fielded use at StoraEnso since 2000 • Commercial ASP tool for data analysis and visualization • Helsingin Sanomat’s “Election machine”
Collaborative search YDIN Personalized search Next Generation Information Retrieval Localized search Topic-specific search
Scalable Probabilistic Methods for the Next Generation Search Engine (PROSE) Search-Ina-Box NGIR projects ALVIS (FP6 STREP) Pascal (FP6 NoE)
Prima: Proactive Information Retrieval by Adaptive Models of Users' Attention and Interests • Basic research project funded by the Academy of Finland (2003-2005) • Joint work with the Neural Networks Research Centre (NNRC) and Center for Knowledge and Innovation Research (CKIR) • Information retrieval enhanced with relevance data and context-sensitive information
Kolmogorov Clustering • Joint work with CWI (Paul Vitanyi & Peter Grünwald) under the Pascal EU Network of Excellence • Basic idea: two objects are similar if they compress well together Kolmogorov similarity metric • In practice: instead of the non-computable Kolmogorov complexity, use domain-specific compression methods • Research issues: supervised clustering, polymorphic (non-homogeneous) data, computational issues,...
Future directions • Search: query relevance modeling, more sophisticated language models, model regularization, hierarchical models, grid computing, intelligent crawl,... • Modeling in general: on-line learning, learning with different loss functions, MDL vs. Bayes, combining prior knowledge with statistical data (in medical diagnosis and bioinformatics), causal inference,... • Visualization of multidimensional data • Polymorphic data, temporal data, intelligent pre-processing, multi-level modeling,... • Multi-agent modeling, distributed inference in sensor networks, positioning in ad hoc wireless networks,...
