Artificial intelligence apparatus using a plurality of output layers and method for same

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Deep learning models face challenges in achieving high accuracy with small amounts of learning data and are prone to overfitting due to the limitations of transfer learning, especially when the last classification layer is learned with a small amount of new data, leading to focused results.

Innovation Solution

An AI apparatus with multiple output layers learned through transfer learning, where weights are adjusted based on brightness and time information, and new layers are trained if results are unknown or inconsistent, allowing for dynamic adaptation and improved classification accuracy.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Loss of time

If transfer learning is applied to learn only the last classification layer with small amounts of new data, then the learning time is reduced, but the classification accuracy deteriorates due to overfitting

Engineering Contradiction:
Improvelearning timeVSAvoidclassification accuracy
Core Design Contradiction:
Loss of timeVSMeasurement precision

Solution Approach 1:

The patent segments the classification layers into multiple groups, where each group is trained with different amounts of new data. Specifically, it divides the classification layers such that some layers use a first amount of training data while other layers use a second amount of training data, allowing different portions of the model to adapt to new data at different rates and preventing overfitting in any single layer.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent applies local quality by making each classification layer have different learning characteristics. Each classification layer is trained with different amounts of new data according to its specific needs, rather than applying a uniform training approach to all layers. This allows each layer to develop optimal local properties for its specific function.

Inventive Principle:
Principle #3Local quality

2Productivity

If the last classification layer is focused on relevant new data, then the learning speed is improved, but the result value becomes overfitted

Engineering Contradiction:
Improvelearning speedVSAvoidresult value reliability
Core Design Contradiction:
ProductivityVSReliability

Solution Approach 1:

The patent segments the classification layers into multiple groups that can be trained at different speeds with different amounts of data. This segmentation allows some layers to learn quickly from new data while other layers maintain their pre-trained knowledge, preventing any single layer from becoming overfitted while maintaining overall learning efficiency.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent implements dynamic training where the amount of new data used for training each classification layer can be adjusted. The system dynamically determines how much new data each layer should receive based on the specific task requirements, allowing flexible adaptation that balances learning speed with generalization capability.

Inventive Principle:
Principle #15Dynamics

Data Source

PatentEP3893162A1Artificial intelligence apparatus using a plurality of output layers and method for same
Publication Date: 2021.10.13 LG ELECTRONICS INC
  • EP3893162A1 patent drawingFigure 1
  • EP3893162A1 patent drawingFigure 2
  • EP3893162A1 patent drawingFigure 3

AI summary

The present disclosure discloses an artificial intelligence apparatus including an input interface configured to obtain input data, a sensing interface configured to obtain environment information, and one or more processors configured to classify an object by inputting the input data obtained from the input interface to an artificial intelligence model, in which the artificial intelligence model uses a first learning model and a second learning model which is connected with the first learning model and includes a plurality of output layers to respectively assign weights to the respective result values output by the plurality of output layers and combine the respective result values to which the weights are assigned to derive the final result.