Semantic Meta-Split Learning: A TinyML Scheme for Few-Shot Wireless Image Classification

TL;DR

This work presents a TinyML-based semantic communication framework for few-shot wireless image classification that integrates split-learning and meta-learning, and exploits split-learning to limit the computations performed by the end-users while ensuring privacy-preserving.

Abstract

Semantic and goal-oriented (SGO) communication is an emerging technology that only transmits significant information for a given task. Semantic communication encounters many challenges, such as computational complexity at end users, availability of data, and privacy-preserving. This work presents a TinyML-based semantic communication framework for few-shot wireless image classification that integrates split-learning and meta-learning. We exploit split-learning to limit the computations performed by the end-users while ensuring privacy-preserving. In addition, meta-learning overcomes data availability concerns and speeds up training by utilizing similarly trained tasks. The proposed algorithm is tested using a data set of images of hand-written letters. In addition, we present an uncertainty analysis of the predictions using conformal prediction (CP) techniques. Simulation results show that the proposed Semantic-MSL outperforms conventional schemes by achieving 20 % gain on classification accuracy using fewer data points, yet less training energy consumption.

Figures (7)

• Figure 1: Multiple devices aim to classify the correct letters in images sampled from different languages. Each device transmits the semantics of the image through a wireless channel to the aggregator, which predicts the image class and transmits it back to the device.
• Figure 2: Illustration of the architecture of split learning. The model is divided at the cut layer into a device-side model and an aggregator-side model, where the training is taking part at both entities.
• Figure 3: Meta-learning architecture. It consists of i) a meta-train that has a support set, which samples meta-tasks for training, and ii) a query set that tests the meta-learning model on new, unseen tasks.
• Figure 4: The proposed Semantic-MSL architecture. The model consists of multiple convolutional layers, each followed by a normalization layer, a ReLu layer, and a pooling layer. At the end, multiple fully connected layers output the predicted class. At the cut layer, a reshape layer corrects the size of the smashed data to be transmitted through the channel.
• Figure 5: The convergence of the accuracy as a function of the SGD steps (iterations) of the proposed Semantic-MSL compared to the Semantic-SL scheme.