Master Thesis - Exploring Semi-Quantitative Methods for Interpreting Correlation Matrices

In many fields of science and engineering, correlation matrices are used to explore relationships between variables in complex systems. These matrices often serve as a foundation for building predictive models, identifying clusters, or guiding decision-making. However, in some cases — such as early-stage data collection, aggregated datasets, or constrained measurement environments — the available correlation data may be of limited resolution. This means that while the matrix may reveal general patterns or associations, it lacks the precision or granularity required for robust mathematical modeling.

Despite these limitations, correlation matrices can still offer valuable insights. They may highlight broad trends, suggest potential groupings, or indicate areas worth further investigation. Traditional visualization techniques like heatmaps can help surface these patterns, but they often fall short of enabling deeper, semi-quantitative interpretation.

This diploma work aims to explore the space between visual exploration and formal modeling, investigating how structured but coarse correlation data can be used to extract meaningful insights. By combining advanced visualization methods with semi-formal analytical tools — such as fuzzy logic, graph-based representations, or heuristic modeling — the project seeks to develop a framework for working effectively with imperfect but informative data.

Project Aim
To investigate and develop methods that bridge the gap between visual exploration and mathematical modeling when working with correlation data of limited resolution, quality, and depth.

Key Components of the Project
Characterize Existing Data
-    Analyze the structure, resolution and quality of the available correlation data.

Explore Visualization Techniques
-    Explore network graphs, clustering, and dimensionality reduction methods.
-    Use graph-based representations to capture structure without assuming precise numerical relationships.

Qualitative Modeling
-    Investigate fuzzy logic, qualitative reasoning, or probabilistic graphical models to represent uncertain or imprecise relationships.

Application and Evaluation
-    Apply the developed methods to a real or synthetic dataset.
-    Compare the insights gained through visual, semi-quantitative, and formal modeling approaches.

Expected Outcome
An alternative approach to model correlations, suitable for visualization, extraction, hypothesis generation and quantitative analysis, including a concrete framework which can be used as basis for future modelling.

Timeframe and Schedule
Full time work one semester.

The project require knowledge and studies in Engineering  Physics.

Selection is made continuously, send in your application today!

Contact Information – Tetra Pak, D&T Industrial Base Engineering
Fredrik Andersson
Systems Engineer A
Mail: fredrikp.andersson@tetrapak.com
Phone: 046-362614

Erik Hamberg
Mail: erik.hamberg@tetrapak.com
Manager, Converting Process Integration
Phone: 046-363410

Du vill göra skillnad. Det vill vi också. Varje dag arbetar vi för att förbättra tillgången till säkra livsmedel för människor över hela världen – och samtidigt skydda planeten. Oavsett om du precis har inlett karriären eller är redo för nästa steg – kom till oss. Tillsammans kan vi bli en del av lösningen.