Model Generation Reducing Weight Variability for Accuracy

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing models, such as SVM and DNN, face limitations in accuracy enhancement, as adjusting hyperparameters alone may not suffice to improve model performance, indicating a need to optimize model parameters directly.

Innovation Solution

An information processing apparatus that obtains training data and generates models by reducing variability in weight parameters, using techniques like normalization and batch normalization, to enhance model accuracy.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If hyperparameters are adjusted to improve model accuracy, then model performance may improve, but the variability in weight parameters remains high and accuracy enhancement is limited

Engineering Contradiction:

Improvemodel accuracyVSAvoidweight variability

Core Design Contradiction:

Measurement precisionVSStability of the object's composition

Solution Approach 1:

The patent changes the parameter being optimized from hyperparameters to weight parameters themselves. By directly optimizing the weight parameters through the generating unit that creates models with decreased weight variability, the system achieves better accuracy without the limitations of hyperparameter tuning. This fundamental parameter change resolves the contradiction by targeting the actual source of model behavior rather than external control parameters.

Inventive Principle:

Principle #35Parameter changes

Solution Approach 2:

The generating unit performs preliminary action by pre-processing the training data and pre-defining the model generation approach before actual model training. This preliminary preparation ensures that when models are generated, they inherently have reduced weight variability, allowing accuracy improvement without needing to adjust hyperparameters during training.

Inventive Principle:

Principle #10Preliminary action

2Measurement precision

If model parameters are optimized to reduce weight variability, then model accuracy improves, but the complexity of the generation process increases

Engineering Contradiction:

Improvemodel accuracyVSAvoidgeneration process complexity

Core Design Contradiction:

Measurement precisionVSDevice complexity

Solution Approach 1:

The generating unit performs self-service by automatically obtaining training data, processing it, and generating models with reduced weight variability without requiring external intervention for hyperparameter tuning. The system serves itself by integrating data processing and model generation into a unified automated process, reducing the apparent complexity despite the advanced techniques employed.

Inventive Principle:

Principle #25Self-service

Solution Approach 2:

The patent merges multiple functions into the generating unit: data obtaining, data processing, and model generation with weight variability reduction. By combining these previously separate processes into one integrated unit, the system manages complexity through functional consolidation while achieving improved accuracy through coordinated optimization.

Inventive Principle:

Principle #5Merging (Combining)

Data Source

PatentUS20220198329A1Information processing apparatus, information processing method, and information processing program

Publication Date: 2022.06.23 ACTAPIO INC

AI summary

An information processing apparatus according to the application concerned includes an obtaining unit that obtains a dataset of training data to be used for the training of a model; and a generating unit that uses the dataset and generates a model in such a way that there is a decrease in the variability in the weight.