18455182. Framework for Learning to Transfer Learn simplified abstract (GOOGLE LLC)
Contents
Framework for Learning to Transfer Learn
Organization Name
Inventor(s)
Sercan Omer Arik of San Francisco CA (US)
Tomas Jon Pfister of Foster City CA (US)
Linchao Zhu of Mountain View CA (US)
Framework for Learning to Transfer Learn - A simplified explanation of the abstract
This abstract first appeared for US patent application 18455182 titled 'Framework for Learning to Transfer Learn
Simplified Explanation
The abstract describes a method for deep learning model training using a Learning to Transfer Learn (L2TL) architecture, involving encoder weights, source classifier layer weights, target classifier layer weights, coefficients, and a policy weight.
- The method applies gradient descent-based optimization to minimize the loss function during the first phase of learning iterations.
- Coefficients are determined by sampling actions of a policy model during the first phase.
- The policy weight that maximizes an evaluation metric is determined during the second phase of learning iterations.
Potential Applications
This technology can be applied in various fields where transfer learning is needed, such as image recognition, natural language processing, and speech recognition.
Problems Solved
1. Efficient transfer learning: The method allows for efficient transfer of knowledge from a source data set to a target data set. 2. Optimization: By optimizing the encoder and classifier weights, the method improves the performance of the deep learning model.
Benefits
1. Improved performance: By minimizing the loss function and maximizing the evaluation metric, the method enhances the performance of the deep learning model. 2. Transfer learning efficiency: The method streamlines the transfer learning process, making it more effective and efficient.
Original Abstract Submitted
A method includes receiving a source data set and a target data set and identifying a loss function for a deep learning model based on the source data set and the target data set. The loss function includes encoder weights, source classifier layer weights, target classifier layer weights, coefficients, and a policy weight. During a first phase of each of a plurality of learning iterations for a learning to transfer learn (L2TL) architecture, the method also includes: applying gradient decent-based optimization to learn the encoder weights, the source classifier layer weights, and the target classifier weights that minimize the loss function; and determining the coefficients by sampling actions of a policy model. During a second phase of each of the plurality of learning iterations, determining the policy weight that maximizes an evaluation metric.
