Multimodal Spiking-Mixer with robustness-improved ODE-neuron

• Propose the first Vision-Language multi-modal Spiking Neural Network for Image-Caption application
• Demonstrate to view SNN as Neural-ODE and by analyzing the stability of the ODE to gain adverserial robustness
• Extend multimodal Vison-Language aversarial attack to SNN domain and demonstrate the effectiveness compared to naive unimodal implementation