일 | 월 | 화 | 수 | 목 | 금 | 토 |
---|---|---|---|---|---|---|
1 | 2 | 3 | 4 | 5 | ||
6 | 7 | 8 | 9 | 10 | 11 | 12 |
13 | 14 | 15 | 16 | 17 | 18 | 19 |
20 | 21 | 22 | 23 | 24 | 25 | 26 |
27 | 28 | 29 | 30 |
Tags
- 폰트생성
- GAN
- BERT
- RNN
- 모션매칭
- dl
- ddpm
- motion matching
- userwidget
- Stat110
- animation retargeting
- Diffusion
- 언리얼엔진
- cv
- 오블완
- WinAPI
- Font Generation
- NLP
- WBP
- Generative Model
- Unreal Engine
- UE5
- 딥러닝
- multimodal
- ue5.4
- CNN
- 디퓨전모델
- deep learning
- Few-shot generation
- 생성모델
Archives
- Today
- Total
목록복습 (1)
Deeper Learning

전체적인 내용 리뷰가 아닌 그룹 스터디 중 토론 나눴던 주제 중심으로 정리. Lecture 5 ~ 7 Sigmoid의 문제점 Saturated neuron kills gridients Not zero centered exp() is expensive 만약 input으로 모두 양수가 주어지면 w의 gradient는 모두 양수거나 모두 음수가 된다. w1, w2가 x, y축을 이루는 사분면을 생각해보면 만약 4사분면에 최적의 W가 존재할 경우 w1, w2의 grad는 각각 양수, 음수로 주어져야 빠른 수렴이 가능하다. 하지만 위 상황에서 이는 불가능하여 oscillation 형태로 수렴이 느려진다. Tanh zero centered saturated kill Relu no saturated kill comp..
AI/Deep Learning
2021. 9. 1. 22:14