IMDEA Networks researchers Andrea Fresa (PhD Student) and Jaya Prakash Champati (Research Assistant Professor) have conducted a study presenting the AMR² algorithm, which leverages edge computing infrastructure (processing, analyzing, and storing data closer to its source to enable faster, near real-time analysis and responses) to increase the inference accuracy of IoT sensors while mitigating latency limitations. They have demonstrated that the problem is successfully resolved. The paper, “An Offloading Algorithm for Maximizing Inference Accuracy on Edge Devices in an Edge Intelligence System,” was published this week at the MSWiM conference.
To understand what inference is, it's necessary to first explain that machine learning (ML) works in two main phases. The first refers to training, when the developer feeds their model a curated dataset so it can "learn" everything it needs to know about the type of data it will analyze. The next phase is inference: the model can make predictions based on real data to produce actionable results.
In their publication, the researchers concluded that the accuracy of inference increased by up to 40% when comparing the AMR² algorithm with basic programming techniques. They also discovered that an efficient programming algorithm is essential to adequately support machine learning algorithms at the edge of the network.
“The results of our study could be very useful for machine learning applications that need fast and accurate inference on end devices. Consider a service like Google Photos, for example, which categorizes elements in images. We can guarantee the execution delay using the AMR² algorithm, which can be very beneficial for a developer who can use it in the design to ensure that delays are not visible to the user,” explains Andrea Fresa.
The main obstacle they encountered in conducting this study was demonstrating the theoretical performance of the AMR² algorithm and validating it using an experimental testbed consisting of a Raspberry Pi and a server connected via a LAN. “To demonstrate the performance limits of AMR², we employed fundamental ideas from linear programming and tools from operations research,” Fresa emphasizes.
However, with this work, IMDEA Networks researchers have laid the groundwork for future research that will help make it possible to run machine learning applications at the network edge quickly and accurately.
