Recently, I've been missing the papers that made me fall in love with the field during my PhD. So I've decided to revisit a few of them in a "paper nostalgia" series of mine.
#3: "Augmented Neural ODEs" by Dupont et al from NeurIPS 2019. I love how much this paper explains with a few simple pictures. It finds a basic limit of neural ODEs, and the fix is just as simple as giving the dynamics extra dimensions.
1/20
Language models recognize dropout and Gaussian noise applied to their activations
3
뭔가 activation monitoring을 하고 있으면 GPU 연산에 변화가 있고 그걸 LLM이 감지하게 훈련할 수 있다 같은 논문을 트위터에서 봤었는데 이게 먼저인건가?
https://t.co/t2kiLPG28D