Spiking Mixers for Robust and Energy-efficient Vision-and-Language Learning
Keywords: Spiking Neural Networks, Multimodal Learning, Mixture of Experts, Adversarial Robustness
Abstract: Multimodal learning is a fundamental challenge in artificial intelligence, with applications spanning computer vision, speech recognition, and natural language processing. This paper presents the pioneering incorporation of Spiking Neural Networks (SNNs) into the Vision-and-Language domain, introducing MLP-Mixer as a unified backbone and adapting mixture of experts approach to effectively fuse different modalities. The Mixer is directly trained using surrogate gradients and has small timesteps. We propose a SNN specific adversarial training technique, combined with the mixture of experts framework, leads to improvements in adversarial robustness. We hope these findings will shed light on future research in the field of Multimodal Spiking Neural Network and adversarial robustness of Multimodal Learning.
Primary Area: unsupervised, self-supervised, semi-supervised, and supervised representation learning
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2024/AuthorGuide.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors' identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 6677
Loading