How to overcome transformers’ flaw of long-document inputs
Can we use hyperspherical prototype networks to compete against softmax?
The 3x3 approach that Slimmer AI uses to drive AI innovation