Overlapping Chunk Attention for Explainable Predictions
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing machine learning methods for classifying information in source documents are labor-intensive, inefficient, prone to errors, and introduce unintended biases, lacking transparency in classification processes.
Innovation Solution
A multi-class ensemble classification model is utilized to generate classification outputs by splitting input data into overlapping chunks, using both class-agnostic and class-specific models to derive feature representations and probabilities, providing a classification output with improved accuracy and trustworthiness by including the data chunk responsible for the prediction.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If manual classification is used to evaluate source documents, then classification accuracy may be maintained, but labor intensity and time consumption increase significantly
Solution Approach 1:
The patent segments the input data object into multiple overlapping data chunks, allowing the classification system to process portions of the data independently and in parallel. This segmentation enables automated processing while maintaining accuracy by focusing on relevant local regions that contain classification evidence.
Solution Approach 2:
The patent introduces an attention mechanism as an intermediary between the input data chunks and the final classification output. This attention mechanism automatically identifies and weights the most relevant data chunks for each class, enabling the system to achieve manual-level accuracy through automated processing by focusing computational resources on critical regions.
2Speed
If traditional machine learning models are used for classification, then processing speed improves, but transparency and trustworthiness of classification decisions deteriorate
Solution Approach 1:
The patent implements a feedback mechanism where the attention mechanism provides backpropagation signals that highlight which data chunks contributed most to each classification decision. This feedback loop enables the system to maintain high processing speeds while providing transparent explanations of the classification rationale by identifying the specific input regions that influenced the output.
Solution Approach 2:
The attention mechanism serves as an intermediary layer that not only improves classification accuracy but also provides interpretability. By computing attention weights for different data chunks, the system can explain which portions of the input data were most influential in reaching the classification decision, thus maintaining transparency without sacrificing processing speed.
3Device complexity
If single classification model is used, then model complexity is reduced, but classification accuracy and robustness decrease
Solution Approach 1:
The patent merges multiple classification models into an ensemble architecture where each model processes the segmented data chunks. The attention mechanism then combines the predictions from these multiple models, weighting them based on their relevance to each class. This merging of multiple models improves classification accuracy and robustness while the attention mechanism manages the complexity by selectively focusing on the most relevant model predictions.
Data Source
AI summary
Embodiments of the present disclosure provide systems and methods for generating explainable predictions. One method may include generating a plurality of overlapping data chunks from an input data object, generating, using a machine learning class-agnostic model, a plurality of intermediate feature representations respectively corresponding to the plurality of overlapping data chunks, generating, using a machine learning class-specific model, a plurality of chunk-based classification probabilities from the plurality of intermediate representations that correspond to a particular prediction class, generating, using the plurality of chunk-based classification probabilities, a plurality of class scores for the plurality of overlapping data chunks, and, providing, by the one or more processors, a classification output that is based on the plurality of class scores and comprises a class prediction for the input data object and an overlapping data chunk from the plurality of overlapping data chunks that corresponds to the class prediction.