The recent developments in the field of machine learning and signal processing are increasingly focusing on the optimization of models for edge computing and real-time applications. A significant trend is the creation of lightweight, efficient models that maintain high accuracy and robustness, especially under challenging conditions such as low signal-to-noise ratios or edge cases in data. Innovations include the development of hybrid neural networks that combine the strengths of convolutional neural networks (CNNs) and transformers, recursive convolution strategies for efficient multi-frequency representations, and multimodal approaches to enhance robustness in specific domains like air-traffic control. These advancements aim to reduce computational overhead and parameter counts without sacrificing performance, making them suitable for deployment on resource-constrained devices.

Noteworthy papers include:

• An ultralight hybrid neural network optimized for edge applications, achieving high accuracy with minimal computational resources.
• RecConv, introducing a recursive decomposition strategy for efficient multi-frequency representations, significantly reducing parameter growth and computational complexity.
• A multimodal call-sign-command recovery model (CCR) that enhances edge-case robustness in air-traffic control speech processing.
• FAST, a Fast Audio Spectrogram Transformer, combining CNNs and transformers for efficient and robust audio classification, achieving state-of-the-art performance with significantly fewer parameters.