Multimedia Classifier Training with Negative Data Filtering

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current techniques for data classification in computer vision and multimedia are prone to errors, particularly in rejecting objects or multimedia data items that do not belong to a specific class, leading to inaccurate classifications and increased computational resource usage.

Innovation Solution

The use of negative multimedia data items to train classifiers, which improves accuracy by differentiating between data items associated with a topic and those that are not, through iterative testing and updating of model vectors based on similarity values.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Use of energy by moving object

If threshold techniques are used to reject objects with low post-classification values, then computational resources are reduced, but classification accuracy deteriorates causing legitimate objects to be mistakenly rejected

Engineering Contradiction:
Improvecomputational resourcesVSAvoidclassification accuracy
Core Design Contradiction:
Use of energy by moving objectVSReliability

Solution Approach 1:

The system performs preliminary classification to identify candidate objects, then applies negative data filtering to eliminate false positives before final classification. This preliminary action using negative multimedia data allows the system to reject unlikely candidates early, reducing computational resources while maintaining accuracy for legitimate objects.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

Negative multimedia data acts as an intermediary between the initial classification and final decision. By introducing this intermediate filtering layer that uses learned negative examples, the system can confidently reject low-scoring candidates without mistakenly rejecting legitimate objects, thus reducing computational waste while preserving accuracy.

Inventive Principle:
Principle #24Intermediary (Mediator)

2Reliability

If n+1 classification techniques with diverse negative data are used, then rejection capability is improved, but model accuracy deteriorates due to the diversity of negative labels

Engineering Contradiction:
Improverejection capabilityVSAvoidmodel accuracy
Core Design Contradiction:
ReliabilityVSMeasurement precision

Solution Approach 1:

The system segments the diverse negative data into multiple negative data sets, each associated with different topics or categories. By training separate classifiers or using topic-specific negative data, the system can handle diverse negative examples more effectively, improving rejection capability while maintaining model accuracy through specialized rather than generalized approaches.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

Instead of using a single uniform model for all negative data, the system applies local quality by using topic-specific negative data sets and potentially different classification strategies for different domains. This allows the model to achieve high accuracy within each topic area while maintaining strong overall rejection capability across diverse categories.

Inventive Principle:
Principle #3Local quality

3Reliability

If more negative multimedia data items are used for training, then classification accuracy is improved, but training computational resources increase

Engineering Contradiction:
Improveclassification accuracyVSAvoidtraining computational resources
Core Design Contradiction:
ReliabilityVSUse of energy by moving object

Solution Approach 1:

The system uses partial action by selecting and using only the most relevant negative multimedia data items for training, rather than processing all available negative data. By applying relevance filtering and focusing on representative samples, the system achieves good classification accuracy with reduced training computational resources.

Inventive Principle:
Principle #16Partial or excessive action

Solution Approach 2:

The system changes parameters by adjusting the size and composition of negative data sets based on training progress and performance metrics. By dynamically modifying the number and type of negative examples used during training, the system can achieve desired accuracy levels while optimizing computational resource usage through adaptive parameter tuning.

Inventive Principle:
Principle #35Parameter changes

Data Source

PatentEP3248144B1Optimizing multi-class multimedia data classification using negative data
Publication Date: 2024.03.13 MICROSOFT TECHNOLOGY LICENSING LLC
  • EP3248144B1 patent drawingFigure 1
  • EP3248144B1 patent drawingFigure 2
  • EP3248144B1 patent drawingFigure 3

AI summary

Techniques for optimizing multi-class image classification by leveraging negative multimedia data items to train and update classifiers are described. The techniques describe accessing positive multimedia data items of a plurality of multimedia data items, extracting features from the positive multimedia data items, and training classifiers based at least in part on the features. The classifiers may include a plurality of model vectors each corresponding to one of the individual labels. The system may iteratively test the classifiers using positive multimedia data and negative multimedia data and may update one or more model vectors associated with the classifiers differently, depending on whether multimedia data items are positive or negative. Techniques for applying the classifiers to determine whether a new multimedia data item is associated with a topic based at least in part on comparing similarity values with corresponding statistics derived from classifier training are also described.