Probabilistic Adaptive Systems

Our goal is to automate intelligent behavior by building robust probabilistic models for a complex world. The work has a strong basic research component that intersects artificial intelligence, machine learning, computer science, information theory and mathematical statistics. The results of this methodological work are applied to both scientific and industrial applications.

Research challenges

Theoretical frameworks for probabilistic modeling. To develop computationally efficient, general-purpose methods for probabilistic modeling, focusing on issues related to model selection, parameter estimation and inference.
Models for intelligent information access. In many modern information networks (like the Internet and various sensor networks), the data can not be found in a well structured format, and accessing the information may be a problem even if the information is in principle available. The goal is to apply probabilistic models to perform information retrieval tasks in these kinds of environments.
Models for image analysis. To develop probabilistic methods for processing two- or three-dimensional measurement data, with applications in data visualization and de-noising, and in the analysis of brain imaging data.
Models for information processing in the visual system of the brain. To develop probabilistic computational models that show how vision is possible in the brain, and to generalize these principles to different domains of computational neuroscience and computational intelligence.
Models for probabilistic data fusion. To develop probabilistic methods for combining inputs originating from heterogeneous data sources.
User modeling. To develop probabilistic modeling methods for personalization, profiling and segmentation.

Current Research Projects

Neuroinformatics
Modest: Applications of the MDL Principle to Prediction and Model Selection and Testing
STAM: Algorithmic Methods in Stemmatology
PinView: Personal Information Navigator Adapting Through Viewing
SULRSL: Supervised Unsupervised Learning and Relevant Subtask Learning
UI-ART: Urban contextual information interfaces with multimodal augmented reality

Past Research Projects

PMMA: Probabilistic Methods for Microarray Data
KUKOT: MDL-Based Methods for Image Denoising
CLASS: Cognitive-Level Annotation using Latent Statistical Structures
CIVI: Cognitively Inspired Visual Interfaces for Representing Multidimensional Information
SIB: Search-ina-Box
Alvis: Superpeer Semantic Search Engine
PASCAL Pump Priming