18170888. DISTILLATION OF DEEP ENSEMBLES simplified abstract (GE Precision Healthcare LLC)
Contents
DISTILLATION OF DEEP ENSEMBLES
Organization Name
Inventor(s)
Hariharan Ravishankar of Bengaluru (IN)
Prasad Sudhakara Murthy of Bengaluru (IN)
DISTILLATION OF DEEP ENSEMBLES - A simplified explanation of the abstract
This abstract first appeared for US patent application 18170888 titled 'DISTILLATION OF DEEP ENSEMBLES
Simplified Explanation: The patent application describes systems and techniques that improve the distillation process of deep learning ensembles. These systems can access a deep learning ensemble and iteratively distill it into a smaller ensemble for inferencing tasks.
- Systems and techniques for distilling deep learning ensembles
- Accessing a deep learning ensemble for inferencing tasks
- Iteratively distilling the ensemble into a smaller one
- Training new neural networks based on previous distillation iterations
- Loss function based on neural networks of the smaller ensemble
Potential Applications: 1. Enhancing the performance of deep learning models 2. Improving efficiency in inferencing tasks 3. Reducing the computational resources required for deep learning ensembles
Problems Solved: 1. Overcoming the challenges of distilling large deep learning ensembles 2. Enhancing the accuracy and speed of inferencing tasks 3. Optimizing the use of computational resources in deep learning applications
Benefits: 1. Improved performance of deep learning models 2. Increased efficiency in inferencing tasks 3. Resource optimization for deep learning applications
Commercial Applications: Optimizing Deep Learning Ensembles for Enhanced Performance in Inferencing Tasks
Questions about Deep Learning Ensemble Distillation: 1. How does distilling deep learning ensembles improve inferencing tasks? 2. What are the key benefits of using smaller deep learning ensembles for inferencing tasks?
Original Abstract Submitted
Systems/techniques that facilitate improved distillation of deep ensembles are provided. In various embodiments, a system can access a deep learning ensemble configured to perform an inferencing task. In various aspects, the system can iteratively distill the deep learning ensemble into a smaller deep learning ensemble configured to perform the inferencing task, wherein a current distillation iteration can involve training a new neural network of the smaller deep learning ensemble via a loss function that is based on one or more neural networks of the smaller deep learning ensemble which were trained during one or more previous distillation iterations.
