Progressive Machine Learning Model with Qualifier-Based Data Selection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Machine learning models face challenges in maintaining accuracy over time, especially in fields where data evolves rapidly, due to the difficulty in obtaining and updating qualified training data, particularly in applications requiring expertise like medical diagnostics.
Innovation Solution
Implementing a progressive machine learning system that allows for continuous updating of models using both local and remote training data, enabling end-users to contribute new examples and improve model performance without requiring extensive machine learning expertise.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If classical machine learning models are trained on initial training data, then the models can make predictions, but the model accuracy deteriorates over time as data evolves rapidly
Solution Approach 1:
The patent implements a progressive machine learning model that dynamically updates itself over time by incorporating new training data. The model transitions from a static structure to a dynamic one that continuously adapts to evolving data patterns, resolving the contradiction between maintaining accuracy and the time required for updates.
Solution Approach 2:
The system establishes a feedback loop where model predictions are continuously evaluated against new data, and successful predictions are fed back as additional training examples. This feedback mechanism enables the model to self-improve and maintain accuracy without requiring extensive manual retraining cycles.
2Reliability
If additional training data is collected to improve model performance, then model accuracy improves, but the complexity of data collection and management increases
Solution Approach 1:
The progressive machine learning model performs self-service by automatically selecting and incorporating its own successful predictions as training data. The system autonomously identifies valuable training examples from its operational predictions, eliminating the need for complex external data collection and management processes.
Solution Approach 2:
The feedback loop captures predictions made during normal model operation and routes them back as training data. This internal feedback mechanism provides a continuous stream of relevant training examples without requiring external data collection infrastructure or complex data management systems.
3Adaptability or versatility
If the model is frequently updated with new training data, then model relevance improves, but the computational resources and time required for training increase
Solution Approach 1:
Instead of performing complete model retraining with all available data, the system applies partial updates using only the subset of predictions identified as valuable training examples. This partial action approach maintains model relevance while significantly reducing the computational burden and training time associated with full retraining cycles.
Solution Approach 2:
The model employs dynamic update strategies where the extent and frequency of training adjustments are adapted based on the rate of data evolution and model performance requirements. This dynamic approach allows the system to balance adaptability with computational efficiency, updating only when and where necessary.
Data Source
AI summary
This disclosure describes techniques in which an artificially intelligent system includes a progressive machine learning model. In some examples, a system includes both a static and progressive model, each trained to make predictions or other assessments through machine learning techniques. The progressive model may be progressively updated, modified, and/or improved through additional training data or training examples, which may be derived from a local source and/or from one or more remote sources that may be chosen by a qualifier. The qualifier may be a human decision-maker that chooses, selects, and/or evaluates potential data or new training examples for use in updating the progressive model. The qualifier may, in other examples, be an artificially intelligent system trained to perform the functions of the qualifier.