18220567. METHOD AND SYSTEM FOR PERSONALISING MACHINE LEARNING MODELS simplified abstract (SAMSUNG ELECTRONICS CO., LTD.)

METHOD AND SYSTEM FOR PERSONALISING MACHINE LEARNING MODELS

Organization Name

SAMSUNG ELECTRONICS CO., LTD.

Inventor(s)

Alberto Gil C. P. Ramos of Staines (GB)

Abhinav Mehrotra of Staines (GB)

Sourav Bhattacharya of Staines (GB)

METHOD AND SYSTEM FOR PERSONALISING MACHINE LEARNING MODELS - A simplified explanation of the abstract

This abstract first appeared for US patent application 18220567 titled 'METHOD AND SYSTEM FOR PERSONALISING MACHINE LEARNING MODELS

Simplified Explanation

The present techniques describe a method and system for personalizing machine learning models on devices with limited resources using conditional neural networks. This involves incorporating a conditioning vector into the weights learned during training, resulting in more efficient use of computational resources during inference.

• The techniques focus on personalizing machine learning models on resource-constrained devices.
• Conditional neural networks are utilized to achieve this personalization.
• The conditioning vector is incorporated into the weights learned during training.
• This approach reduces the computational resources required during inference time.

Potential Applications

The potential applications of this technology include:

• Personalized recommendation systems on mobile devices.
• Speech recognition and natural language processing on IoT devices.
• Image recognition and object detection on edge devices.
• Real-time monitoring and analysis on wearable devices.

Problems Solved

The problems solved by this technology are:

• Overcoming resource limitations on devices with limited computational power.
• Enabling personalization of machine learning models on resource-constrained devices.
• Reducing the computational resources required during inference.

Benefits

The benefits of this technology are:

• Improved performance and accuracy of machine learning models on resource-constrained devices.
• Efficient use of computational resources during inference.
• Personalization of models without compromising device limitations.
• Enhanced user experience and responsiveness on low-power devices.

Original Abstract Submitted

Broadly speaking, embodiments of the present techniques provide a method and system for personalising machine learning models on resource-constrained devices by using conditional neural networks. In particular, the present techniques allow for resource-efficient use of a conditioning vector by incorporating the conditioning vector into weights learned during training. This reduces the computational resources required at inference time.