EAI

What is EAI?

Embedded AI, also referred to as Embedded Artificial Intelligence (EAI), is a comprehensive framework system designed for performing AI tasks. It is integrated into network devices and offers a wide range of functionalities such as model management, data acquisition, and data preprocessing for AI algorithms implemented on these devices. Moreover, it supports the capability to transmit inference results to AI algorithm-based functions. By leveraging the device's data and computing resources, Embedded AI maximizes efficiency while providing advantages like reduced data transmission costs, ensured data security, and real-time inference and decision-making capabilities.

The Significance of EAI

The arrival of the fourth industrial revolution, driven by AI, is transforming human society and the world at an unprecedented pace. AI technologies have the potential to bring value to network devices in various aspects, including parameter optimization, application identification, security, and fault diagnosis.

AI comprises three fundamental elements: algorithms, computing power, and data. However, when AI algorithm-based functions rely heavily on a device to provide a significant amount of sample data and high-performance computing capabilities, it can severely impact the device's normal operations.

To address this challenge, the EAI system offers a comprehensive and adaptable framework for AI functions. Devices can subscribe to the services provided by the EAI system, allowing real-time data related to the AI function to undergo inference using the AI algorithm. This enables the implementation of AI functions based on the resulting inferences. By analyzing and inferring locally-generated data, the EAI system offers advantages such as reduced data transmission costs, enhanced data security, and the ability to make real-time inferences and decisions.

Operational Mechanism of EAI

The EAI system comprises three essential modules:

• Model module: Also known as the algorithm module, this module integrates multiple AI algorithms. It manages a collection of model files, with each file containing one or more models corresponding to different AI algorithms. Users have the ability to load and delete model files, thereby managing the AI algorithms utilized by the EAI system.

• Data module: This module is responsible for data acquisition, preprocessing, and management. It handles the vast amount of data required by all AI functions deployed on devices.

• Computing power module: The computing power module executes inference operations using algorithms from the model module and data from the data module. The resulting inferences are transmitted to the AI functions supported by the device. These functions analyze the inference results, generate specific configurations, and deliver them to the device.

The EAI system's implementation technique is depicted in the above image.

1. The data module of the EAI system is responsible for gathering extensive data associated with each AI function on the device. It preprocesses this data and utilizes the preprocessed data as input for the computing power module.

2. Within the EAI system, users have the capability to load or delete various model files. These model files contain trained models that are relevant to each AI function.

3. When AI functions on the device subscribe to the services of the EAI system, user configuration is not required. Subscription is automatically completed once an AI function is enabled. Once subscribed, the EAI system safeguards the subscribed model within the model file, ensuring that the latest version of the model cannot be deleted and is used as input for the computing power module.

4. The computing module carries out inference operations by utilizing algorithms from the model module and data from the data module. It then transmits the resulting inference to the enabled AI function.

5. The AI function responds to the inference result generated by the EAI system by delivering specific configurations based on the outcome.

Application of EAI

The AI ECN (Artificial Intelligence Explicit Congestion Notification) function dynamically adjusts the ECN thresholds of lossless queues based on the real-time traffic model in the live network. Its objective is to achieve optimal performance for lossless services by ensuring low latency, high throughput, and zero packet loss.

Unlike traditional static ECN functions that require manual parameter configuration, such as fixed ECN thresholds and marking probabilities, the AI ECN function adapts to the changing buffer space in the queue. By subscribing to the EAI function, the AI ECN function undergoes AI training using the live network's traffic model. This enables it to predict network traffic changes and infer the optimal ECN threshold in a timely manner. Additionally, the ECN threshold can be dynamically adjusted in response to real-time traffic changes, allowing precise management and control of the lossless queue buffer across the entire network.

Once enabled on a device, the AI ECN function automatically subscribes to the services provided by the EAI system. It intelligently analyzes the pushed traffic status information and determines the current traffic model based on the inference result from the EAI system. If the traffic model has been trained in the EAI system, the AI ECN function calculates the appropriate ECN threshold that aligns with the current network status. It then delivers this optimal ECN threshold to the device, adjusting the ECN threshold for lossless queues accordingly. This process is repeated for newly obtained traffic status, ensuring optimal performance of lossless queues in various traffic scenarios, thereby achieving low latency and high throughput for lossless services.