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 | 31 |
Tags
- Copilot
- 표현론
- matrix multiplication
- 딥러닝
- Private ML
- 머신러닝
- bert
- NLP
- Language Modeling
- Natural Language Processing
- ICML
- Homomorphic Encryption
- attention
- math
- Knowledge Tracing
- Deep learning
- 자연어처리
- Machine Learning
- Residual Connection
- GPT
- Model Compression
- Computer Vision
- Pre-training
- Data Augmentation
- AI
- Transformer
- KT
- Knowledge Distillation
- Github Copilot
- 동형암호
Archives
- Today
- Total
목록Language Modeling (1)
Anti Math Math Club
ReZero is All You Need: Fast Convergence at Large Depth
딥러닝의 발전에 있어서 중요한 발견 중 한가지는 ResNet의 발명이라고 할 수 있습니다. 매우 deep한 뉴럴넷을 학습시키기위해서 input의 정보를 그대로 output에 더해줌으로써 모델은 input과 output의 "차이"에 해당하는 부분만 학습할 수 있도록 해주는 것 입니다. 실제로 Residual Connection을 사용했을 대 vanishing gradient problem도 어느정도 해결이 된다고 알려져 있습니다. 하지만 시간이 지날수록 점점 더 크고 깊은 모델을 사용하게 되면서, Residual Connection만으로는 부족하다고 느껴지고 이를 개선한 여러가지 모델들과 방법론이 등장하게 됩니다. 예를 들어서, Batch Normalization이나 Layer Normalization과 ..
Machine Learning & Deep Learning/Algorithms
2021. 3. 14. 14:30