18148418. SYSTEMS AND METHODS FOR NEURAL ARCHITECTURE SEARCH simplified abstract (Samsung Electronics Co., Ltd.)
Contents
SYSTEMS AND METHODS FOR NEURAL ARCHITECTURE SEARCH
Organization Name
Inventor(s)
Mostafa El-khamy of San Diego CA (US)
Yanlin Zhou of San Diego CA (US)
SYSTEMS AND METHODS FOR NEURAL ARCHITECTURE SEARCH - A simplified explanation of the abstract
This abstract first appeared for US patent application 18148418 titled 'SYSTEMS AND METHODS FOR NEURAL ARCHITECTURE SEARCH
Simplified Explanation
The abstract describes a system and method for neural architecture search, where a neural network is trained using a training data set and a smooth maximum unit regularization value to compute training loss. The connection weights of the neural network are adjusted to reduce the training loss.
- Training data set is processed with a neural network during the first epoch of training.
- Training loss is computed using a smooth maximum unit regularization value.
- Connection weights of the neural network are adjusted to reduce training loss.
Potential applications of this technology:
- Automated machine learning model development
- Optimization of neural network architectures
- Improving performance of deep learning models
Problems solved by this technology:
- Manual tuning of neural network architectures
- Time-consuming process of architecture search
- Improving efficiency and accuracy of neural network training
Benefits of this technology:
- Faster development of machine learning models
- Enhanced performance of neural networks
- Reduction in human effort required for architecture search.
Original Abstract Submitted
A system and a method are disclosed for neural architecture search. In some embodiments, the method includes: processing a training data set with a neural network during a first epoch of training of the neural network; computing a training loss using a smooth maximum unit regularization value; and adjusting a plurality of multiplicative connection weights and a plurality of parametric connection weights of the neural network in a direction that reduces the training loss.
