Deep Learning Model Segmentation for Interpretable Feature Extraction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Deep learning models in computer vision lack interpretability, making it difficult for humans to understand their decision-making processes, which is a critical issue in medical AI where trust and explanation of model judgments are essential.
Innovation Solution
A method that involves a deep learning model with a pre-trained loss function combining a cross-entropy loss function for classification and a hyperbolic log loss function for regression, with a regularization factor to adjust the weight between these functions, allowing for the extraction of feature vectors interpretable based on domain knowledge and improving model performance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a deep learning model is used for classification tasks, then classification performance can be improved, but interpretability of the model's decision-making process deteriorates
Solution Approach 1:
The deep learning model is segmented into two distinct neural networks: a first neural network responsible for extracting interpretable feature vectors, and a second neural network responsible for generating probability values for classification. This segmentation allows each component to specialize in its respective function, maintaining interpretability through the first network while achieving accurate classification through the second network.
Solution Approach 2:
An interpretable feature vector serves as an intermediary between the input image and the final classification output. The first neural network extracts this intermediate representation that maintains domain knowledge interpretability, which then serves as input to the second neural network for probability estimation, thus bridging the gap between interpretability and accuracy.
2Measurement precision
If a complex loss function combining multiple objectives is used, then model performance and interpretability can be improved, but training complexity increases
Solution Approach 1:
The loss function merges two distinct objectives into a unified training framework: a classification loss (cross-entropy) for accuracy and a regularization loss for interpretability. By combining these losses with appropriate weighting, the model simultaneously optimizes for both performance and interpretability during training.
Solution Approach 2:
A regularization factor parameter is introduced to control the relative weight between the classification loss and the interpretability regularization loss. By adjusting this parameter, the training process can balance between achieving high classification accuracy and maintaining interpretable feature representations, allowing flexible optimization based on specific application needs.
Data Source
AI summary
According to an exemplary embodiment of the present disclosure, a method for classification by using a deep learning model, the method being performed by a computing device, is disclosed. The method may include: extracting a feature vector interpretable based on domain knowledge by inputting an image including at least one object of interest into a first neural network of a deep learning model; and estimating a probability value corresponding to a classification task by inputting the feature vector into a second neural network of the deep learning model. In this case, the deep learning model may be pre-trained based on a loss function having an output value of the first neural network and an output value of the second neural network as input variables.