Make structures in high-dimensional data visible and interpretable.
Large datasets and many measurements make it difficult to see what matters. Relevant groups, outliers and differences remain hidden in tables and simple charts.
IAP makes similarities and differences accessible through projection, clustering and databionic self-organisation. A topographic representation of gene-expression profiles with 7,500 measurements each shows how complex data can be interpreted together with domain experts.
You receive visualisations of groups, outliers and differences in your data, interpreted together with your experts. The analysis identifies which features merit further investigation for classification or forecasting.
Read the technical detail ↓In the example shown, each point represents a gene-expression profile with 7,500 dimensions. Swarm intelligence searches for structure without using diagnosis labels. Similar profiles occupy valleys; ridges expose differences between groups. The landscape can also be represented as a three-dimensional model.
Known diagnoses are colour-coded for subsequent interpretation. Their agreement with the discovered groups shows which diagnostic distinctions are represented in the measurements. This examines the structure and meaning of the data before a classifier is applied.
Two atypical cases in the example warrant closer examination of diagnosis and data quality. Such evidence helps experts investigate measurement problems, rare cases or unsuitable assignments.
Stier, Q., Hoffmann, J. & Thrun, M. C. (2026). Classifying with the Fine Structure of Distributions: Leveraging Distributional Information for Robust and Plausible Naïve Bayes. Machine Learning and Knowledge Extraction (MAKE), 8(1), 13. MDPI.
Uses fine distributional structure to provide a robust and interpretable classification baseline. The implementation supports parallel computation across multiple processor cores with shared memory.
Thrun, M. C. & Märte, J. (2026). Memshare: Memory Sharing for Multicore Computation in R with an Application to Feature Selection by Mutual Information using PDE. The R Journal, 17(4), 306–322.
Shared memory for parallel computation in R, demonstrated through feature selection in high-dimensional gene-expression data.
Thrun, M. C. & Ultsch, A. (2021). Swarm intelligence for self-organized clustering. Artificial Intelligence, 290, 103237. Elsevier.
DatabionicSwarm uses swarm intelligence and self-organisation to discover groups and structures in high-dimensional data. Clustering follows similarities and density differences in the data without requiring predefined classes.
Hospital prototype · collaborative research
The quality of blood and bone-marrow samples affects interpretation. Dilution and complex cell populations need to be considered in the analysis.
Project details →
Collaborative project · Proof of Concept (PoC)
Company fundamentals were to support systematic stock selection that people could inspect.
Project details →
Data analysis · further method development
How does use of a company website relate to later project enquiries? The analysis for LANCOM Systems examined browsing behaviour alongside associated project information.
Project details →Option 1
IAP investigates the problem, develops and evaluates the approach, and hands over the agreed prototype with documentation. Your IT team takes responsibility for further product development and operation.
Option 2
IAP develops the agreed application, including its interface and integrations, connects it to the intended workflow, and completes acceptance and handover. Your organisation or operating partner takes responsibility for subsequent operation.
In a free, non-binding 20-minute introductory discussion, we clarify your objective, starting point and a useful next step.
