ML Pipeline for Combinatorial Space Exploration Beyond Local Optima
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for exploring combinatorial spaces in engineering, biology, and chemistry are resource-intensive and prone to getting trapped in local optima, failing to efficiently discover optimal solutions due to reliance on prior knowledge and limited data sets.
Innovation Solution
An iterative machine learning pipeline that uses diverse ML models to identify top performers, generates augmented data from under-represented regions, and balances input data distribution to escape sub-optimal entrapment, ensuring comprehensive exploration of the combinatorial space.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional experimental methods are used to explore combinatorial spaces, then comprehensive coverage of all combinations is achieved, but resource consumption and time requirements become prohibitively high
Solution Approach 1:
The patent applies preliminary action by using machine learning models to predict and prioritize combinations before experimental screening. The system pre-processes the combinatorial space by identifying high-priority candidates based on learned patterns from initial data, allowing the most promising combinations to be tested first while less promising ones are filtered out, thus reducing overall screening time while maintaining exploration completeness.
Solution Approach 2:
The patent uses copying by creating virtual representations of experimental data through machine learning models. Instead of physically testing all combinations, the system creates computational copies and simulations of the combinatorial space, allowing virtual exploration and prediction of outcomes. This enables comprehensive analysis of combinatorial possibilities without the resource cost of physical experimentation for every combination.
2Reliability
If diverse machine learning models are trained and evaluated, then model performance and accuracy improve, but computational complexity and resource requirements increase
Solution Approach 1:
The patent applies segmentation by dividing the machine learning pipeline into distinct modular components: multiple diverse models are trained separately on different aspects of the data, then their predictions are aggregated or compared. This segmentation allows each model to specialize in specific patterns while the system as a whole benefits from diverse perspectives, improving accuracy without requiring a single overly complex model.
Solution Approach 2:
The patent merges multiple machine learning model outputs to generate final predictions. By combining the strengths of diverse models (e.g., ensemble methods, consensus approaches), the system achieves higher reliability and accuracy. The merging process integrates results from different model architectures and training approaches, allowing the system to leverage collective intelligence while managing individual model complexity.
3Stability of the object's composition
If model-driven augmented data is generated from under-represented regions, then data distribution balance improves, but data processing complexity increases
Solution Approach 1:
The patent implements feedback by continuously monitoring data distribution across the combinatorial space and using this information to guide augmented data generation. The system identifies under-represented regions through distribution analysis, then generates targeted augmented data for those specific areas. This feedback loop ensures data balance is maintained dynamically while focusing computational resources only on regions that need enhancement, rather than uniformly processing all data.
Solution Approach 2:
The patent applies local quality by generating augmented data with specific characteristics tailored to under-represented regions of the combinatorial space. Instead of applying uniform data augmentation across all regions, the system adapts the augmentation strategy to local data distribution needs, creating synthetic samples that specifically address gaps in particular areas while maintaining overall data quality and relevance.
Data Source
AI summary
A computer-implemented method and related system explore a combinatorial space. A combinatorial library and a desired output are identified. From the combinatorial library, an initial dataset is identified to be tested experimentally to create the combinatorial space. The following functions are iteratively performed: experimentally screening a set of diverse machine learning models (MLMs) using the initial data set or an augmented data set to produce experimental screening results; training the MLMs using the experimental screening results; selecting, from the MLMs, at least one MLM having a highest accuracy and performance; screening the combinatorial library; calculating a normalized similarity factor measured from top-ranked combinations; identifying, using the normalized similarity factor, an amount of the model-driven augmented data to be added to the top-ranked combinations; obtaining augmented data; and selecting the augmented data from the top-ranked combinations and the augmented combinatorial data. The iteration exits upon meeting an exit criterion.