17932809. FEDERATED LEARNING SURROGATION WITH TRUSTED SERVER simplified abstract (QUALCOMM Incorporated)

FEDERATED LEARNING SURROGATION WITH TRUSTED SERVER

Organization Name

QUALCOMM Incorporated

Inventor(s)

Jian Shen of San Diego CA (US)

Jamie Menjay Lin of San Diego CA (US)

FEDERATED LEARNING SURROGATION WITH TRUSTED SERVER - A simplified explanation of the abstract

This abstract first appeared for US patent application 17932809 titled 'FEDERATED LEARNING SURROGATION WITH TRUSTED SERVER

Simplified Explanation

The present disclosure relates to techniques for surrogated federated learning, where a trusted server receives intermediate activations from a node device to refine weights associated with a neural network.

• Explanation:

 * Intermediate activations are received at a trusted server from a node device.
 * The node device generated the intermediate activations using a first set of layers of a neural network.
 * Weights associated with a second set of layers of the neural network are refined using the intermediate activations.
 * Weight updates corresponding to the refined weights are transmitted to a federated learning system.

• • Potential Applications:**

This technology can be applied in various fields such as healthcare, finance, and telecommunications for collaborative machine learning tasks.

• • Problems Solved:**

1. Enhances privacy and security by allowing the refinement of weights without sharing raw data. 2. Facilitates efficient model training across multiple devices without compromising data confidentiality.

• • Benefits:**

1. Improved model accuracy through collaborative learning. 2. Reduced communication overhead by transmitting weight updates instead of raw data.

• • Potential Commercial Applications:**

"Surrogated Federated Learning in Healthcare: Enhancing Model Training while Protecting Patient Data"

• • Possible Prior Art:**

Prior art in federated learning and privacy-preserving machine learning techniques may exist, but specific examples are not provided in this disclosure.

• • Unanswered Questions:**

• • 1. How does the trusted server ensure the security and privacy of the intermediate activations received from the node device?**

The process of securing and encrypting the intermediate activations for transmission is not detailed in the abstract.

• • 2. What types of neural networks are most suitable for surrogated federated learning, and are there any limitations in terms of network architecture?**

The abstract does not specify the compatibility or restrictions of neural network architectures for this technique.

Original Abstract Submitted

Certain aspects of the present disclosure provide techniques and apparatus for surrogated federated learning. A set of intermediate activations is received at a trusted server from a node device, where the node device generated the set of intermediate activations using a first set of layers of a neural network. One or more weights associated with a second set of layers of the neural network are refined using the set of intermediate activations, and one or more weight updates corresponding to the refined one or more weights are transmitted to a federated learning system.