According to the first arXiv preprint, researchers propose an ‘Energy-Efficient Federated Learning’ approach that uses ‘Adaptive Encoder Freezing’ to reduce the computational load for MRI-to-CT conversion tasks in federated learning. The key idea is to selectively freeze parts of the neural network encoder to minimize the energy consumption of edge devices participating in the federated training process.
The second preprint discusses ‘Over-the-Air Federated Learning’ (AirFL), which integrates wireless signal processing and distributed machine learning to enable scalable AI at the network edge. AirFL leverages the superposition property of wireless signals to perform model aggregation directly over the air, avoiding the need for a central server.
Finally, the third preprint presents ‘Scalable Decision Focused Learning’, which aims to improve decision support systems by training estimators that are optimized for the downstream decision-making task, rather than just the prediction objective. This is achieved through the use of online trainable surrogates that can be updated during the optimization process.