I asked #ChatGPT o1 mini to generate a pic of a christmas tree in the middle of a crowded market place of #Darjeeling on a snowy evening. Instead it responded with a short para asking me to imagine with it! I feed in to #Gemini step by step and got this! #MerryChristmas

.. of equiv-CNNs against deformations which would include adversarial examples. Some experiments with SE(2)-CKNs and SO(3)-CKNs have been provided to support the claims. To be presented virtually at @GRaM_workshop Paper link: https://t.co/arLpQROEJe Questions? Send an email!

We generalize the notion of convolutional kernel networks (ckns) to make it equivariant to any Lie group transformations and then study it's stability bounds. As one can embed equiv-CNNs in equiv-CKNs, such Lipschitz styled bound can be useful in understanding the robustness..

How robust are equivariant convolutional representations w.r.t. deformations (incld. adversarial examples)? One answer lies in understanding equivariant CNNs through the lens of equivariant CKNs, following a standard analysis of embedding CNNs in RKHSs. @GRaM_workshop