Managing AI and Machine Learning on Many Devices from the Cloud

This patent describes a system for remotely monitoring and updating artificial intelligence and machine learning programs running on a large collection of smart devices, like robots or sensors, from a central cloud system.

What it covers

This patent details a system for managing a 'fleet' of edge devices, which are small computers or sensors that perform tasks locally. It works by receiving monitoring information about AI or machine learning programs running on these edge compute units, as described in claim 1. It also collects status information from connected edge assets, such as cameras or environmental sensors. All this data is shown on a remote graphical user interface (GUI) in a cloud environment, allowing a user to see how a selected group of devices is performing. The user can then provide input through the GUI to update the configuration of an AI or machine learning workload on specific edge devices. For example, a user could tell a group of factory robots to update their object recognition model (a pre-trained ML model) to better identify new product defects, with the cloud system transmitting the necessary control information, as outlined in claim 1.

What it doesn't cover

• —Does not cover managing individual edge devices without a central cloud-based GUI for fleet-wide oversight and configuration.
• —Does not cover systems where the AI/ML model updates are not initiated by a user configuration input received via a remote GUI.
• —Does not cover edge device management where the updated configuration is not specifically for a pre-trained ML or AI model or network connectivity.
• —Does not cover systems that only monitor edge devices without the capability to transmit control information for updated configurations.
• —Does not cover local, on-device management of AI/ML workloads without interaction with a cloud computing environment.

The clever bit

The novelty lies in providing a unified, cloud-based graphical interface to not only monitor a diverse fleet of edge devices and their AI workloads, but also to centrally push complex updates like local retraining, finetuning, or distributed learning orchestration back to specific devices or groups of devices.

Why it matters

Managing vast numbers of smart devices, especially those running complex AI or machine learning tasks, is a significant challenge. This patent offers a structured way to centrally monitor and update these distributed systems. It is particularly relevant for industries deploying large-scale IoT (Internet of Things) solutions, where manual updates for each device are impractical. This approach helps ensure that AI models on edge devices remain accurate and efficient without needing physical access to each unit.

Real-world examples

• 1.Autonomous vehicle fleets updating their perception models
• 2.Smart factory robots receiving new defect detection algorithms
• 3.Remote sensor networks for environmental monitoring adjusting their data analysis models
• 4.Retail store cameras updating their customer behavior analytics AI
• 5.Industrial IoT deployments managing predictive maintenance models on machinery