- Reinforcement Learning with TensorFlow
- Sayon Dutta
- 82字
- 2021-08-27 18:51:57
Overcoming the limitations of deep learning
These two possible problems can be overcome by:
- Minimizing the use of the sigmoid and tanh activation functions
- Using a momentum-based stochastic gradient descent
- Proper initialization of weights and biases, such as xavier initialization
- Regularization (add regularization loss along with data loss and minimize that)
For more detail, along with mathematical representations of the vanishing and exploding gradient, you can read this article: Intelligent Signals : Unstable Deep Learning. Why and How to solve them ?
推薦閱讀
- Ansible Quick Start Guide
- 物聯(lián)網(wǎng)與云計(jì)算
- Visual C# 2008開(kāi)發(fā)技術(shù)詳解
- Maya 2012從入門到精通
- 可編程控制器技術(shù)應(yīng)用(西門子S7系列)
- 機(jī)器人編程實(shí)戰(zhàn)
- 基于多目標(biāo)決策的數(shù)據(jù)挖掘方法評(píng)估與應(yīng)用
- 計(jì)算機(jī)系統(tǒng)結(jié)構(gòu)
- MATLAB/Simulink權(quán)威指南:開(kāi)發(fā)環(huán)境、程序設(shè)計(jì)、系統(tǒng)仿真與案例實(shí)戰(zhàn)
- 水下無(wú)線傳感器網(wǎng)絡(luò)的通信與決策技術(shù)
- SAP Business Intelligence Quick Start Guide
- 自動(dòng)化生產(chǎn)線安裝與調(diào)試(三菱FX系列)(第二版)
- 精通LabVIEW程序設(shè)計(jì)
- Photoshop CS4數(shù)碼攝影處理50例
- Learning Cassandra for Administrators