MLP-Mixer
A vision architecture based on multilayer perceptrons (MLPs) that dispenses entirely with convolution and self-attention, mixing information by alternately applying MLPs along the spatial dimension and the channel dimension.
Timeline (last 90 days)
A developer has open-sourced a Graph HyperNetwork (GHN) project that can predict all weights in a single forward pass for previously unseen ResNet, MLP-Mixer, MobileNet and even Transformer architectures.
The token mixing and channel mixing of EG-RL draw on the design philosophy of MLP-Mixer-style architectures, decoupling the two dimensions of the feature tensor for separate processing.
All Facts (2)
A developer has open-sourced a Graph HyperNetwork (GHN) project that can predict all weights in a single forward pass for previously unseen ResNet, MLP-Mixer, MobileNet and even Transformer architectures.
50%UnverifiedThe token mixing and channel mixing of EG-RL draw on the design philosophy of MLP-Mixer-style architectures, decoupling the two dimensions of the feature tensor for separate processing.
50%