VR Data Evaluation for ML Model Retraining
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The difficulty in evaluating the output of machine learning models by individuals without expertise in statistics and artificial intelligence, particularly in domains like strawberry inspection, where domain experts may not be the best evaluators.
Innovation Solution
A system that uses virtual or augmented reality to allow users to interact with multidimensional data, allowing them to provide feedback on the output of machine learning models, which can then be used to retrain the models, utilizing modules like ML training, post-processing, and manifold creation to reduce dimensions for visualization and user interaction.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If traditional machine learning evaluation methods are used, then evaluation accuracy may be maintained, but the barrier to entry for domain experts without statistics and AI expertise becomes too high
Solution Approach 1:
The patent introduces an intermediary system that includes automatic evaluation algorithms, visualization tools, and guidance mechanisms. These intermediaries bridge the gap between domain experts and complex machine learning evaluation tasks, allowing experts to evaluate model output without needing deep statistics and AI knowledge. The system mediates by translating complex technical requirements into accessible interfaces while maintaining evaluation rigor.
Solution Approach 2:
The evaluation process is segmented into multiple independent components: data preprocessing modules, various evaluation metrics (accuracy, precision, recall, F1-score), visualization tools, and guidance systems. Each component can be used independently or combined based on specific needs, allowing domain experts to engage with only the portions relevant to their expertise while the system handles the technical complexity.
2Loss of information
If multidimensional data is presented for evaluation, then comprehensive assessment is achieved, but the complexity of visualization and user interaction increases
Solution Approach 1:
The patent transforms high-dimensional evaluation data into lower-dimensional visual representations by adding temporal and spatial dimensions. Evaluation metrics are displayed across time sequences, spatial distributions, and hierarchical levels, allowing comprehensive information to be perceived through multiple dimensional perspectives rather than overwhelming single-view complexity.
Solution Approach 2:
The visualization system dynamically adapts its complexity based on user interaction and evaluation needs. It provides dynamic filtering, drilling-down capabilities, and adaptive detail levels that adjust in real-time, allowing users to explore comprehensive data when needed while maintaining simplicity during routine evaluations.
Data Source
AI summary
Techniques for evaluating an output of a machine learning model and using the evaluation to retrain the machine learning model are described. For example, a data set that is output from a layer of the machine learning model is reduced to a 2-D or 3-D representation that is suitable for viewing. A user views the reduced data set in a viewing environment such as virtual reality or augmented reality. The user makes changes using that viewing environment. The changes are then used to retrain the machine learning model.