View source for 18159349. DEEP LEARNING OPTIMIZER FOR FINE-TUNING WHILE DYNAMICALLY MITIGATING CATASTROPHIC FORGETTING simplified abstract (GM Cruise Holdings LLC)

=DEEP LEARNING OPTIMIZER FOR FINE-TUNING WHILE DYNAMICALLY MITIGATING CATASTROPHIC FORGETTING=


==Organization Name==

[[:Category:GM Cruise Holdings LLC|GM Cruise Holdings LLC]]


[[Category:GM Cruise Holdings LLC]]

==Inventor(s)==

[[:Category:Zhao Chen of Mountain View CA (US)|Zhao Chen of Mountain View CA (US)]][[Category:Zhao Chen of Mountain View CA (US)]]

[[:Category:Shuai Zheng of Santa Clara CA (US)|Shuai Zheng of Santa Clara CA (US)]][[Category:Shuai Zheng of Santa Clara CA (US)]]

==DEEP LEARNING OPTIMIZER FOR FINE-TUNING WHILE DYNAMICALLY MITIGATING CATASTROPHIC FORGETTING - A simplified explanation of the abstract==


This abstract first appeared for US patent application 18159349 titled 'DEEP LEARNING OPTIMIZER FOR FINE-TUNING WHILE DYNAMICALLY MITIGATING CATASTROPHIC FORGETTING

'''Simplified Explanation:'''
The patent application describes a deep learning optimizer that fine-tunes a model while preventing catastrophic forgetting.

'''Key Features and Innovation:'''
* Saving the original state of the model to a buffer.
* Applying new gradients to intermediate model states during training.
* Calculating a displacement vector to adjust the model's weights.
* Performing an orthogonal projection of the final gradient to mitigate inconsistencies.
* Applying the orthogonal gradient to the original state of the model.

'''Potential Applications:'''
This technology can be applied in various fields such as computer vision, natural language processing, and speech recognition for improving model performance while preventing catastrophic forgetting.

'''Problems Solved:'''
This technology addresses the issue of catastrophic forgetting in deep learning models, where new information can cause the model to forget previously learned information.

'''Benefits:'''
The technology allows for fine-tuning of models without losing previously learned information, leading to improved performance and efficiency in deep learning tasks.

'''Commercial Applications:'''
Potential commercial applications include enhancing the accuracy and reliability of AI systems in industries such as healthcare, finance, and autonomous vehicles.

'''Prior Art:'''
Researchers can explore prior art related to deep learning optimization techniques, model fine-tuning, and methods to prevent catastrophic forgetting in neural networks.

'''Frequently Updated Research:'''
Stay informed about the latest advancements in deep learning optimization, model fine-tuning, and techniques to mitigate catastrophic forgetting to ensure the technology remains cutting-edge.

'''Questions about Deep Learning Optimizer:'''
1. How does the deep learning optimizer prevent catastrophic forgetting in neural networks?
2. What are the potential implications of using this technology in real-world applications?


==Original Abstract Submitted==

Disclosed are embodiments for a deep learning optimizer for fine-tuning while dynamically mitigating catastrophic forgetting. In some aspects, a method includes saving an original state of a model to a buffer, the original state comprising original weights of the model; apply new gradients to intermediate model states of the model during at least one intermediate training step of training of the model; calculate, at a final training step of the training, a displacement vector representing a difference between current weights of the model and the original weights of the model; responsive to the displacement vector and a final gradient of the final training step being inconsistent in terms of gradient direction, perform an orthogonal projection of the final gradient on the displacement vector to generate an orthogonal gradient; and applying the orthogonal gradient to the original state of the model at the final training step of the training.



[[Category:G06N3/08]]
[[Category:CPC_G06N3/08]]