17533779. EPHEMERAL LEARNING OF MACHINE LEARNING MODEL(S) simplified abstract (Google LLC)

From WikiPatents
Jump to navigation Jump to search

EPHEMERAL LEARNING OF MACHINE LEARNING MODEL(S)

Organization Name

Google LLC

Inventor(s)

Françoise Beaufays of Mountain View CA (US)

Khe Chai Sim of Dublin CA (US)

Trevor Strohman of Sunnyvale CA (US)

Oren Litvin of New York NY (US)

EPHEMERAL LEARNING OF MACHINE LEARNING MODEL(S) - A simplified explanation of the abstract

This abstract first appeared for US patent application 17533779 titled 'EPHEMERAL LEARNING OF MACHINE LEARNING MODEL(S)

Simplified Explanation

The patent application describes a method for training machine learning models using audio data without storing or logging the data. Here are the key points:

  • The system receives audio data from a user's device, capturing spoken utterances.
  • The data is processed by a fulfillment pipeline to perform certain actions based on the spoken utterances.
  • Simultaneously, a training pipeline processes the audio data using unsupervised learning techniques to generate gradients.
  • After processing, the audio data is discarded, ensuring that it is not stored or logged by the system.
  • This approach allows for efficient training of machine learning models without compromising the security of user data.

Potential Applications

This technology can have various applications, including:

  • Voice assistants: Training voice assistants to understand and respond to user commands without storing their audio data.
  • Speech recognition: Improving speech recognition systems by training them on large amounts of audio data without the need for data storage.
  • Language translation: Enhancing language translation models by training them on spoken utterances without compromising user privacy.

Problems Solved

The technology addresses the following problems:

  • Data storage and privacy: By discarding audio data after processing, the system ensures that user data is not stored or logged, increasing privacy and security.
  • Efficient training: The approach allows for training machine learning models without the need for storing large amounts of audio data, making the training process more efficient.

Benefits

The technology offers several benefits:

  • Enhanced privacy: User audio data is not stored or logged, ensuring the privacy and security of user information.
  • Efficient training: The system can train machine learning models without the need for storing or managing large amounts of audio data, improving training efficiency.
  • Versatile applications: The technology can be applied to various domains, such as voice assistants, speech recognition, and language translation, providing benefits across different industries.


Original Abstract Submitted

Implementations disclosed herein are directed to ephemeral learning of machine learning (“ML”) model(s) based on gradient(s) generated at a remote system (e.g., remote server(s)). Processor(s) of the remote system can receive stream(s) of audio data capturing spoken utterance(s) from a client device of a user. A fulfillment pipeline can process the stream(s) of audio data to cause certain fulfillment(s) of the spoken utterance(s) to be performed. Meanwhile, a training pipeline can process the stream(s) of audio data to generate gradient(s) using unsupervised learning techniques. Subsequent to the processing by the fulfillment pipeline and/or the training pipeline, the stream(s) of audio data are discarded by the remote system. Accordingly, the ML model(s) can be trained at the remote system without storing or logging of the stream(s) of audio data by non-transient memory thereof, thereby providing more efficient training mechanisms for training the ML model(s) and also increasing security of user data.

Retrieved from "http://wikipatents.org/index.php?title=17533779._EPHEMERAL_LEARNING_OF_MACHINE_LEARNING_MODEL(S)_simplified_abstract_(Google_LLC)&oldid=24318"