Probability Distribution Learning for Multi-Peak Model Accuracy

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing machine learning techniques face difficulties in accurately modeling complex probability distributions, particularly multi-peaked distributions, whether with or without pre-existing learning data.

Innovation Solution

A multi-staged learning approach is employed, starting with a simpler probability distribution and progressively increasing its complexity, using a parameter λ to expand the objective distribution, and employing a monotonous increasing sequence to refine the model, leveraging techniques like Wasserstein metric for accuracy.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Device complexity

If a complex probability distribution is modeled directly by machine learning, then the model complexity increases, but the learning accuracy deteriorates

Engineering Contradiction:
Improvemodel complexityVSAvoidlearning accuracy
Core Design Contradiction:
Device complexityVSMeasurement precision

Solution Approach 1:

The learning process is segmented into multiple stages: first learning a simple probability distribution, then progressively learning more complex distributions by incorporating additional peaks. This divides the complex learning task into manageable sub-tasks, where each stage builds upon the previous one, preventing the model from being overwhelmed by the full complexity at once.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The method performs preliminary learning of a simple probability distribution before attempting to learn the complex target distribution. The simple distribution serves as a foundation or initialization, preparing the model in advance for the more challenging task of capturing complex multi-peaked structures, thereby improving final learning accuracy.

Inventive Principle:
Principle #10Preliminary action

2Ease of operation

If learning data is not prepared in advance, then the ease of operation improves, but the learning accuracy deteriorates

Engineering Contradiction:
Improveease of learningVSAvoidlearning accuracy
Core Design Contradiction:
Ease of operationVSMeasurement precision

Solution Approach 1:

The system performs self-service by automatically generating learning data from the simple probability distribution that it has already learned. Instead of requiring external preparation of learning data, the model uses its own learned simple distribution to create training data for the next stage, enabling autonomous progressive learning without manual data preparation.

Inventive Principle:
Principle #25Self-service

Solution Approach 2:

The method performs preliminary learning of a simple probability distribution that can then serve as a source for generating learning data. This preliminary action creates a self-sufficient system where the model generates its own training data, eliminating the need for external data preparation while maintaining high learning accuracy through the progressive complexity approach.

Inventive Principle:
Principle #10Preliminary action

Data Source

PatentEP4654096A1Learning program, learning method, and information processing apparatus
Publication Date: 2025.11.26 FUJITSU LTD
  • EP4654096A1 patent drawingFigure 1A~1B
  • EP4654096A1 patent drawingFigure 2A~2B
  • EP4654096A1 patent drawingFigure 3

AI summary

Provided is a learning program causing a computer to execute a process. The process includes a first process of generating a probability distribution model by learning a probability distribution having fewer peaks than an objective probability distribution, and a second process of generating a new probability distribution model by learning a probability distribution closer to the objective probability distribution than a learned probability distribution using a parameter of a generated probability distribution model.