Managing and Updating AI on Many Smart Devices from the Cloud

This patent describes a system for remotely monitoring and updating artificial intelligence models running on a fleet of connected smart devices, like factory robots or smart cameras, all managed from a central cloud-based control panel.

What it covers

This patent describes a method for managing many smart devices, called a 'fleet of edge devices,' that run artificial intelligence (AI) or machine learning (ML) programs. It works by receiving 'monitoring information' about the AI programs and 'status information' from each device (Claim 1). This information is then shown on a 'remote fleet management graphical user interface (GUI)' in the cloud, allowing a user to see what's happening with a selected group of devices. The user can then provide 'user configuration inputs' through this GUI to update how an AI program works on one or more devices. For example, a user could tell a group of smart cameras to collectively 'retrain' their object recognition AI model using new data they've collected, with the cloud system sending 'orchestration information' to coordinate this distributed retraining (Claim 1). The system can also manage local retraining or fine-tuning of models on individual edge devices (Claim 4, 7).

What it doesn't cover

• —Does not cover managing a single edge device; it specifically requires a 'fleet of edge devices' (Claim 1).
• —Does not cover managing devices that do not implement 'machine learning (ML) or artificial intelligence (AI) workloads' (Claim 1).
• —Does not cover systems where the AI models are not 'pre-trained' before being deployed to the edge compute units (Claim 1).
• —Does not cover local management of edge devices without a 'remote fleet management graphical user interface (GUI)' in a cloud environment (Claim 1).
• —Does not cover systems where the updated configuration is not related to retraining or finetuning a pre-trained ML/AI model (Claim 1, 4, 7, 10).
• —Does not cover systems that only monitor edge devices without the ability to transmit control information for updated configurations (Claim 1).

The clever bit

The innovation lies in providing a centralized, cloud-based system to remotely monitor and, critically, *update* (retrain or finetune) pre-trained AI models across an entire fleet of distributed edge devices, even coordinating distributed retraining efforts among them.

Why it matters

As more AI and ML applications move from central data centers to 'edge devices' like smart sensors and robots, managing these distributed systems becomes complex. This patent addresses the challenge of keeping many AI models updated and performing well across a large network of devices. It allows for centralized control and updates, which is crucial for maintaining performance and security in large-scale deployments like smart factories or autonomous vehicle fleets.

Real-world examples

• 1.Managing AI models on industrial robots in a smart factory
• 2.Updating object detection AI on a fleet of autonomous delivery vehicles
• 3.Controlling AI-powered security cameras across multiple retail stores
• 4.Orchestrating AI updates for smart city traffic sensors
• 5.Managing predictive maintenance AI on wind turbines in a wind farm