Method and Device for the Automatic Analysis of Models

Simple SummaryContent extracted from patent full text and abstract with AI.

The invention provides a method and device for automatically analyzing complex, non-linear machine learning models (especially those based on kernel methods) used to predict properties of objects (like chemical molecules or technical systems) that are not previously characterized. This solution not only produces accurate predictions but also identifies and prioritizes which training examples or features have the strongest influence on those predictions. The analysis generates interpretable insights, such as rankings or visualizations, which allow users to understand and potentially simplify the models.

Use CasesContent extracted from patent full text and abstract with AI.

• Drug discovery and toxicology research: understanding which molecular features contribute most to toxicity predictions for new chemical compounds.
• Chemical property prediction: estimation of properties such as solubility, melting point, or receptor binding for unknown molecules.
• Industrial process optimization: modeling and simplifying control systems for chemical plants by identifying critical process variables.
• Manufacturing and production: pinpointing the most influential process steps or component characteristics for quality prediction and process control.
• Diagnostic and predictive maintenance: applying the methodology to technical systems (e.g., machinery, buildings) to identify key factors influencing system performance, efficiency, or failure risks.
• Model transparency in AI/ML: integrating into software tools to make complex machine learning models more interpretable and auditable.

BenefitsContent extracted from patent full text and abstract with AI.

• Enables interpretability and explainability of complex machine learning models, helping experts understand which factors drive outcomes.
• Allows automated model simplification and feature selection, reducing complexity and computational costs while retaining predictive power.
• Improves trust and acceptability of predictive models in regulated domains (e.g., pharma, chemical, manufacturing) by providing transparent rationales for predictions.
• Supports targeted optimization by revealing which attributes or modifications most effectively change predicted properties (e.g., reducing toxicity).
• Facilitates model reduction and streamlined analysis, making it easier to deploy and maintain models in real-world applications.
• Enhances the decision-making process by translating 'black-box' model insights into understandable, actionable information for domain experts.

Technical Classifications (CPCs)

Inventors & Applicants

Patent Abstract

The invention relates to a method and a device for the automatic analysis of a non-linear model for predicting the properties of an object which is a priori not characterized. According to the method, a) the non-linear model is elaborated for training objects based on a mechanical learning method, especially a kernel-based learning method, in such a manner that it allows a statement regarding at least one property for at least one object, b) at least one measure is automatically determined by means of an analytical element using the representer theorem, said measure indicating which training object or which training objects that have become part of the non-linear model have the strongest influence on the predictions of the non-linear model, and c) a prioritized data set is automatically produced in which the measures of the influencing factors are put in the order of a predetermined condition.