深度神经网络三大关键点
- Hyper parameter -- could control parameter W and b
- Learning Rate
- #iteration
- #Hidden layer
- Hidden Units
- choice of activiative function
- 网络架构
- 不同的问题有不同架构去解决
- 主要两大类:
- CNN - 针对图像 ,对称性,平移,旋转
- RNN - 跟时间序列有关的数据 ,有记忆能力
- 训练方法
- 不同的训练方法,对性能的提升有重要的影响
- 论文:currleculum learning &
What is Deep Learning
Brief Theory:
Deep learning (also known as deep structured learning, hierarchical learning or deep machine learning) is a branch of machine learning based on a set of algorithms that attempt to model high-level abstractions in data by using multiple processing layers, with complex structures or otherwise, composed of multiple non-linear transformations.
深度学习本质
提炼特征(representation)
BP是深度神经网络的基础,没有它,误差没办法收敛。
从数学的角度:多层神经网络的本质就是一个多层复合的函数。
基本原理
DL 比传统的ML的优势在于: data越多,性能越好.
DL 的重新崛起的原因: scale drive DL progress
Data , computation , Algorithms(sigmod , ReLU)
Supervised Learning
Input | Output | Application | Tech | Data Type |
---|---|---|---|---|
Home Features | Price | Real Estate | Linear | Structured |
Ad,user info | Click on Ad ?(0 / 1) | Online Advertising | Standard NN | Structured |
Image | Object(1,....1000) | Photo tagging | CNN | Unstructured |
Audio | Text transcript | Speech Recognition | RNN | Unstructured |
English | Chinese | Machine translation | RNN | Unstructured |
Image,Radar info | Position of other car | Autonomous driving | Unstructured | |