Rewrite the Stars
This work addresses a foundational gap in network design for machine learning researchers, though it appears incremental as it builds on existing intuitive explanations of the star operation.
The paper tackles the problem of understanding and utilizing the star operation (element-wise multiplication) in neural networks, revealing its ability to map inputs into high-dimensional, non-linear feature spaces without widening the network, and demonstrates that StarNet achieves impressive performance and low latency under compact structures.
Recent studies have drawn attention to the untapped potential of the "star operation" (element-wise multiplication) in network design. While intuitive explanations abound, the foundational rationale behind its application remains largely unexplored. Our study attempts to reveal the star operation's ability to map inputs into high-dimensional, non-linear feature spaces -- akin to kernel tricks -- without widening the network. We further introduce StarNet, a simple yet powerful prototype, demonstrating impressive performance and low latency under compact network structure and efficient budget. Like stars in the sky, the star operation appears unremarkable but holds a vast universe of potential. Our work encourages further exploration across tasks, with codes available at https://github.com/ma-xu/Rewrite-the-Stars.