Pre-trained Model Selector for Automated Machine Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Users in domains such as medical or autonomous driving lack the resources and time to deploy state-of-the-art machine learning solutions, as training sophisticated neural networks for image recognition is resource-intensive and requires significant expertise, limiting the widespread adoption of machine learning services.
Innovation Solution
The solution involves using pre-trained models that are dynamically adapted to specific user tasks by fine-tuning portions of existing models, reducing the need for extensive retraining and resource consumption, allowing users with little AI knowledge to generate hosted machine learning services for image recognition, audio processing, and other tasks.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If sophisticated neural networks are trained from scratch for image recognition, then model accuracy is improved, but resource consumption and training time increase significantly
Solution Approach 1:
The patent applies preliminary action by using pre-trained models that have already been trained on large datasets before being deployed for specific tasks. This allows the model to leverage previously acquired knowledge and features, eliminating the need to train from scratch while maintaining high accuracy for new applications.
Solution Approach 2:
The patent implements parameter changes through fine-tuning, where specific parameters of the pre-trained model are adjusted and optimized for the target task. This involves modifying certain layers or parameters of the existing model to adapt to new data distributions while preserving the beneficial features learned during pre-training.
2Measurement precision
If sophisticated neural networks are trained from scratch, then model accuracy is improved, but training time increases significantly
Solution Approach 1:
The patent applies preliminary action by using pre-trained models that have already been trained on large datasets before being deployed for specific tasks. This allows the model to leverage previously acquired knowledge and features, eliminating the need to train from scratch while maintaining high accuracy for new applications.
3Use of energy by moving object
If pre-trained models are used and fine-tuned, then resource consumption is reduced, but model adaptability to specific tasks may be limited
Solution Approach 1:
The patent implements parameter changes through fine-tuning, where specific parameters of the pre-trained model are adjusted and optimized for the target task. This involves modifying certain layers or parameters of the existing model to adapt to new data distributions while preserving the beneficial features learned during pre-training.
Solution Approach 2:
The patent applies dynamics by enabling flexible configuration of fine-tuning parameters, allowing users to adjust the degree and scope of adaptation based on their specific needs. This dynamic approach lets users balance between maintaining resource efficiency and achieving sufficient task-specific performance.
4Measurement precision
If extensive retraining is performed, then model performance on specific tasks is improved, but complexity of the process increases
Solution Approach 1:
The patent implements parameter changes through fine-tuning, where specific parameters of the pre-trained model are adjusted and optimized for the target task. This involves modifying certain layers or parameters of the existing model to adapt to new data distributions while preserving the beneficial features learned during pre-training.
Data Source
AI summary
Features related to systems and methods for automated generation of a machine learning model based in part on a pretrained model are described. The pretrained model is used as a starting point to augment and retrain according to client specifications. The identification of an appropriate pretrained model is based on the client specifications such as model inputs, model outputs, and similarities between the data used to train the models.