DNN Outlier Detection Using T-Way Feature Combinations
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Deep Neural Networks (DNNs) struggle with outlier detection due to the need for large training data sets and unpredictability in handling unseen image types, leading to erroneous results, particularly in medical imaging.
Innovation Solution
A method using t-way feature combinations to analyze DNN feature vectors, flagging abnormal inputs by quantizing feature vectors to discrete values and comparing against thresholds, reducing the impact on normal inputs.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If DNNs are used for outlier detection, then processing efficiency and data representation capability are improved, but reliability deteriorates due to heavy dependence on training data and unpredictability on unseen inputs
Solution Approach 1:
The system performs preliminary action by computing statistics of feature vector values during the training phase and storing them for later use. This pre-computed statistical information is then applied during outlier detection to reliably identify outliers without requiring the DNN to have seen similar inputs during training, thus improving reliability while maintaining processing efficiency.
2Measurement precision
If large training data sets are used to improve DNN performance, then detection accuracy is improved, but the need for extensive retraining and data collection increases
Solution Approach 1:
The invention extracts the essential statistical characteristics of normal data by computing statistics of feature vector values during training. These extracted statistical parameters are then used for outlier detection without requiring the full training dataset, thereby maintaining detection accuracy while eliminating the need for extensive retraining and data collection.
3Ease of operation
If traditional outlier detection methods are used, then interpretability is improved, but processing capability for large amounts of data deteriorates
Solution Approach 1:
The system segments the complex outlier detection problem into two parts: first, the DNN processes large amounts of data efficiently to extract feature vectors; second, simple statistical comparisons are performed on these feature vectors to detect outliers. This segmentation allows the system to maintain both high data processing capability and interpretability, as the statistical comparison step is easily understandable.
Data Source
AI summary
Outlier detection using a Deep Neural Network (DNN) includes running a trained DNN model on an received input item. A first feature vector is extracted from the input item and quantized to discrete values. A first number of special t-way feature combinations are computed in the input item and compared against a computed threshold. Based on the comparison, the input item is flagged as an outlier and an alert is generated notifying of the flagged input item.