Machine Learning Perforations for Isolating Exception Decisions

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing machine learning models struggle to effectively consider and learn from exception decisions, leading to potential weakening of model performance due to attempts to incorporate anomalous feedback without proper differentiation.

Innovation Solution

Generate and utilize perforations within machine learning models to identify and store additional features contributing to exception decisions, allowing for separate validation of feature boundaries and deviation scores, thereby improving model accuracy without altering the base model.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If machine learning models attempt to incorporate anomalous feedback from exception decisions, then the models can consider more diverse decision criteria, but the model performance may be weakened due to lack of proper differentiation between normal and exception cases

Engineering Contradiction:

Improveability to consider exception decisionsVSAvoidmodel performance

Core Design Contradiction:

Adaptability or versatilityVSReliability

Solution Approach 1:

The patent segments the decision space by creating perforations - specific sub-regions within the base model's decision boundary where exception decisions are allowed. This segmentation enables the model to handle exception cases without compromising overall performance by isolating anomalous feedback to specific perforated regions rather than incorporating it globally into the base model.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The patent extracts exception decisions from the base model by identifying and isolating anomalous feedback cases. These extracted exception cases are then handled separately through perforation models, preventing them from degrading the base model's performance while still allowing the system to learn from diverse decision criteria.

Inventive Principle:

Principle #2Taking out (Extraction)

2Adaptability or versatility

If machine learning models incorporate exception decisions without differentiation, then more decision criteria are considered, but the complexity of managing both normal and anomalous feedback increases

Engineering Contradiction:

Improveability to learn from exception decisionsVSAvoidmodel management complexity

Core Design Contradiction:

Adaptability or versatilityVSDevice complexity

Solution Approach 1:

The patent divides the feedback processing into two distinct segments: base model for normal decisions and perforation models for exception decisions. This segmentation simplifies management by providing clear boundaries and separate processing paths, reducing the complexity of managing both normal and anomalous feedback simultaneously.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The patent introduces perforation models as intermediary components between the base model and exception decisions. These intermediary perforation models act as mediators that handle anomalous feedback specifically, simplifying the overall system architecture by providing a dedicated layer for exception management rather than mixing all feedback types in a single complex model.

Inventive Principle:

Principle #24Intermediary (Mediator)

3Adaptability or versatility

If perforation models are generated for each exception decision, then the ability to handle exception decisions improves, but the computational overhead and processing time increases

Engineering Contradiction:

Improveexception decision handling capabilityVSAvoidprocessing time

Core Design Contradiction:

Adaptability or versatilityVSLoss of time

Solution Approach 1:

The patent performs preliminary action by pre-generating perforation models for anticipated exception scenarios and storing them in advance. When exception decisions occur, the system can quickly retrieve and apply pre-generated perforation models rather than creating them in real-time, significantly reducing processing time while maintaining the ability to handle diverse exception cases.

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The patent uses copying by generating perforation models that replicate the structure and learning capabilities of the base model but are specialized for exception handling. These copied models can be efficiently stored and reused across multiple exception scenarios, reducing computational overhead compared to creating entirely new models for each exception type.

Inventive Principle:

Principle #26Copying

Data Source

PatentUS12354020B2Generating and utilizing perforations to improve decision making

Publication Date: 2025.07.08 INTERNATIONAL BUSINESS MACHINE CORPORATION

AI summary

An embodiment for managing machine learning models to generate and utilize perforations within machine learning models to improve their ability to consider and learn from exception decisions. The embodiment may detect an exception decision in a base model. The embodiment may automatically determine data considered in making the exception decision and identify and store in a database known features from the gathered data. The embodiment may automatically identify and store in the database remaining additional features considered, and generate and store perforations corresponding to the remaining additional features considered. The embodiment may, in response to detecting subsequent decisions involving shared additional features contained in the generated perforations, automatically validate feature boundaries within the generated perforations from a set of data sources. The embodiment may automatically calculate scores for the subsequent decisions using both the base model and corresponding perforation and output decision recommendations for the subsequent decisions.