17838722. NEURAL CAPACITANCE: NEURAL NETWORK SELECTION VIA EDGE DYNAMICS simplified abstract (INTERNATIONAL BUSINESS MACHINES CORPORATION)
NEURAL CAPACITANCE: NEURAL NETWORK SELECTION VIA EDGE DYNAMICS
Organization Name
INTERNATIONAL BUSINESS MACHINES CORPORATION
Inventor(s)
Pin-Yu Chen of White Plains NY (US)
Tejaswini Pedapati of White Plains NY (US)
Chuang Gan of Cambridge MA (US)
Chunheng Jiang of Troy NY (US)
Jianxi Gao of Niskayuna NY (US)
NEURAL CAPACITANCE: NEURAL NETWORK SELECTION VIA EDGE DYNAMICS - A simplified explanation of the abstract
This abstract first appeared for US patent application 17838722 titled 'NEURAL CAPACITANCE: NEURAL NETWORK SELECTION VIA EDGE DYNAMICS
Simplified Explanation
The abstract describes a method for modifying a pre-trained neural network model by incorporating a neural capacitance probe unit on top of one or more bottom layers. The probe unit is randomly initialized and the modified model is trained by fine-tuning the bottom layers on a target dataset. The method involves obtaining an adjacency matrix from the initialized probe unit and computing a neural capacitance metric using this matrix. An active model is selected based on this metric and a machine learning system is configured using the active model.
- The method modifies a pre-trained neural network model by adding a neural capacitance probe unit on top of one or more bottom layers.
- The probe unit is randomly initialized and the modified model is trained by fine-tuning the bottom layers on a target dataset.
- An adjacency matrix is obtained from the initialized probe unit and a neural capacitance metric is computed using this matrix.
- An active model is selected based on the neural capacitance metric.
- A machine learning system is configured using the selected active model.
Potential Applications
- This method can be applied in various fields where neural networks are used, such as computer vision, natural language processing, and speech recognition.
- It can be used to improve the performance and accuracy of pre-trained neural network models in specific tasks or domains.
Problems Solved
- The method solves the problem of modifying pre-trained neural network models to adapt them to specific tasks or datasets.
- It addresses the challenge of incorporating additional layers or units on top of existing neural network architectures without losing the knowledge learned by the pre-trained model.
Benefits
- The method allows for the modification and fine-tuning of pre-trained neural network models, saving time and computational resources compared to training from scratch.
- It enables the customization of neural network models for specific tasks or datasets, improving their performance and accuracy.
- The use of the neural capacitance metric helps in selecting the most suitable modified model, ensuring optimal performance for the target task.
Original Abstract Submitted
An output layer is removed from a pre-trained neural network model and a neural capacitance probe unit with multiple layers is incorporated on top of one or more bottom layers of the pre-trained neural network model. The neural capacitance probe unit is randomly initialized and a modified neural network model is trained by fine-tuning the one or more bottom layers on a target dataset for a maximum number of epochs, the modified neural network model comprising the neural capacitance probe unit incorporated with multiple layers on top of the one or more bottom layers of the pre-trained neural network model. An adjacency matrix is obtained from the initialized neural capacitance probe unit and a neural capacitance metric is computed using the adjacency matrix. An active model is selected using the neural capacitance metric and a machine learning system is configured using the active model.
