Update README.md

README.md

@@ -8,12 +8,16 @@ library_name: timm
 
 # Model Card for Model ID
 
-Based on **quasi-linear hyperbolic systems of PDEs** [1](https://github.com/liuyao12/ConvNets-PDE-perspective), the QLNet enters an uncharted water of ConvNet model space marked by the use of (element-wise) multiplication instead of ReLU as the primary nonlinearity. It achieves comparable performance as ResNet50 on ImageNet-1k (acc=78.4), demonstrating that its capacity/expressivity, and deserves more study (hyper-paremeter tuning, optimizer, etc.) by the community.
+Based on **quasi-linear hyperbolic systems of PDEs** [[Liu et al, 2023](https://github.com/liuyao12/ConvNets-PDE-perspective)], the QLNet enters an uncharted water of ConvNet model space marked by the use of (element-wise) multiplication instead of ReLU as the primary nonlinearity. It achieves comparable performance as ResNet50 on ImageNet-1k (acc=78.4), demonstrating that its capacity/expressivity, and deserves more study (hyper-paremeter tuning, optimizer, etc.) by the community.
 
-*This modelcard aims to be a base template for new models. It has been generated using [this raw template](https://github.com/huggingface/huggingface_hub/blob/main/src/huggingface_hub/templates/modelcard_template.md?plain=1).*
 
 ![](https://huggingface.co/liuyao/QLNet/resolve/main/QLNet.jpeg)
 
+One notable feature of the architecture is the absence of ReLU or other element-wise activation functions, which enables the model to admit a *continuous* symmetry in its parameters. Check out the [notebook](https://colab.research.google.com/#fileId=https://huggingface.co/liuyao/QLNet/blob/main/QLNet_symmetry.ipynb) for a demonstration that transforms the weights while leaving the output unchanged.
+
+*This modelcard aims to be a base template for new models. It has been generated using [this raw template](https://github.com/huggingface/huggingface_hub/blob/main/src/huggingface_hub/templates/modelcard_template.md?plain=1).*
+
+
 ## Model Details
 
 ### Model Description