18541972. SYSTEMS AND METHODS FOR ON-DEVICE VALIDATION OF A NEURAL NETWORK MODEL simplified abstract (SAMSUNG ELECTRONICS CO., LTD.)

SYSTEMS AND METHODS FOR ON-DEVICE VALIDATION OF A NEURAL NETWORK MODEL

Organization Name

SAMSUNG ELECTRONICS CO., LTD.

Inventor(s)

Gokulkrishna M of Bengaluru (IN)

Siva Kailash Sachithanandam of Bengaluru (IN)

Prasanna R of Bengaluru (IN)

Rajath Elias Soans of Bengaluru (IN)

Alladi Ashok Kumar Senapati of Bengaluru (IN)

Praveen Doreswamy Naidu of Bengaluru (IN)

Pradeep Nelahonne Shivamurthappa of Bengaluru (IN)

SYSTEMS AND METHODS FOR ON-DEVICE VALIDATION OF A NEURAL NETWORK MODEL - A simplified explanation of the abstract

This abstract first appeared for US patent application 18541972 titled 'SYSTEMS AND METHODS FOR ON-DEVICE VALIDATION OF A NEURAL NETWORK MODEL

Simplified Explanation

The abstract describes a method for validating a trained artificial intelligence model on a device by deploying a validation model that incorporates anticipated configurational changes and actual deviations that occurred during training. The output of both models is combined to validate the trained AI model.

• The method involves deploying a validation model that considers anticipated configurational changes and actual deviations during training.
• Input data is provided to both the validation model and the trained AI model to receive outputs for comparison.
• The outputs of both models are combined to validate the trained AI model.

Potential Applications

This technology could be applied in various industries where the validation of AI models is crucial, such as healthcare, finance, autonomous vehicles, and manufacturing.

Problems Solved

This method addresses the challenge of ensuring the accuracy and reliability of trained AI models on devices by validating them against potential configurational changes and deviations that may have occurred during training.

Benefits

The method provides a more robust validation process for AI models, increasing their trustworthiness and performance on devices. It helps in identifying and correcting any discrepancies that may arise due to configurational changes or deviations.

Potential Commercial Applications

Potential commercial applications of this technology include AI model validation services for companies developing AI solutions, AI model testing tools for device manufacturers, and AI quality assurance services for various industries.

Possible Prior Art

One possible prior art could be the use of simulation environments to validate AI models against different scenarios and configurations. Another could be the use of statistical methods to assess the performance of AI models in real-world settings.

What are the potential limitations of this method in real-world applications?

One potential limitation of this method could be the computational resources required to deploy and run the validation model alongside the trained AI model on devices with limited processing power.

How does this method compare to existing AI model validation techniques?

This method stands out by incorporating anticipated configurational changes and actual deviations during training to provide a more comprehensive validation process compared to traditional validation techniques that may not account for such factors.

Original Abstract Submitted

A method for validating a trained artificial intelligence (AI) model on a device is provided. The method includes deploying a validation model generated by applying a plurality of anticipated configurational changes associated with the trained AI model requiring validation. Further, the method includes providing input data to each of the validation model and the trained AI model for receiving an output from each of the validation model and the trained AI model, wherein the output of the validation model is further based on one or more actual configurational deviations that occurred during training of the trained AI model since deployment of the trained AI model on the device. Furthermore, the method includes combining the output of each of the validation model and the trained AI model to validate the trained AI model.