Analytics Cluster


Lead: Thorsten WOHLAND

Department of Biological Sciences, National University of Singapore

Developing task-specific evaluation schemes to achieve optimal detection outcomes in complex environments


To innovate data analysis strategies using modern methods of Mathematics, Statistics, and Artificial Intelligence coupled with the advantages of digital detection to improve information extraction and accurate and precise quantification of Detection Cluster’s outcomes. This will create new opportunities for developing readily deployable environmental and health monitoring devices with increased sensitivity and reliability, with faster results at reduced cost.


This cluster aims at answering three fundamental questions: (1) What is the quality of the data? (2) What is the information in a data set? (3) How can I extract information effectively? Answering these questions will produce novel data analysis paradigms that can be used to analyse sensing data in different formats and dependencies in real-time with high accuracy and precision. The three questions are addressed in three investigative strands:

(1) Image Mathematics and Statistics describe the image and its formation process in precise terms that provide a fundamental understanding of their characteristics in dependence on the technology developed in the Transduction Cluster. Taking advantage of these properties will lead to more accurate data analysis and parameter predictions, harnessing advantages in modern regression analysis and Bayesian statistics.

(2) Through Algorithmic Information Theory (AIT) we will explore how one can describe the information content of a data set and develop new strategies for efficient data analysis and novelty detection. By extracting a maximum of parameters, we will detect novel dimensions of information in existing and new data sets and explore parameter coherence and consistency to provide higher prediction probabilities.

(3) Artificial Intelligence, and especially Deep Learning, has proven in many cases to be less data hungry, quicker, and often more precise in its parameter predictions. These advantages of AI raise at least three distinct questions. First, how is the superior performance achieved within a convolutional neural network (CNN), a question we classify under interpretable AI. Second, can we rationally construct optimal CNNs adapted to particular sensing tasks and avoid erroneous predictions? Third, can we use CNNs or more generally AI for novelty detection in collaboration with the second strand on AIT?

Principal Investigators

Matthew R. FOREMAN
Duane LOH
Alexandre THIERY

© 2022 - 2024 Institute for Digital Molecular Analytics and Science (IDMxS)