Explore
For Enterprise
Join for Free
Log In

艺术与人文
商务
计算机科学
数据科学
信息技术
健康
数学和逻辑
个人发展
物理科学与工程
社会科学
语言学习
學位
證書

瀏覽 Coursera 的全部課程

瀏覽
搜索
企業版
登錄
免費加入

Gradient Descent Updater Strategies

Loading...

Applied AI with DeepLearning

4.4（436 個評分） | 26K 名學生已註冊

課程 1（共 4 門，Advanced Data Science with IBM 專項課程）

>>> By enrolling in this course you agree to the End User License Agreement as set out in the FAQ. Once enrolled you can access the license in the Resources area <<< This course, Applied Artificial Intelligence with DeepLearning, is part of the IBM Advanced Data Science Certificate which IBM is currently creating and gives you easy access to the invaluable insights into Deep Learning models used by experts in Natural Language Processing, Computer Vision, Time Series Analysis, and many other disciplines. We’ll learn about the fundamentals of Linear Algebra and Neural Networks. Then we introduce the most popular DeepLearning Frameworks like Keras, TensorFlow, PyTorch, DeepLearning4J and Apache SystemML. Keras and TensorFlow are making up the greatest portion of this course. We learn about Anomaly Detection, Time Series Forecasting, Image Recognition and Natural Language Processing by building up models using Keras on real-life examples from IoT (Internet of Things), Financial Marked Data, Literature or Image Databases. Finally, we learn how to scale those artificial brains using Kubernetes, Apache Spark and GPUs. IMPORTANT: THIS COURSE ALONE IS NOT SUFFICIENT TO OBTAIN THE "IBM Watson IoT Certified Data Scientist certificate". You need to take three other courses where two of them are currently built. The Specialization will be ready late spring, early summer 2018 Using these approaches, no matter what your skill levels in topics you would like to master, you can change your thinking and change your life. If you’re already an expert, this peep under the mental hood will give your ideas for turbocharging successful creation and deployment of DeepLearning models. If you’re struggling, you’ll see a structured treasure trove of practical techniques that walk you through what you need to do to get on track. If you’ve ever wanted to become better at anything, this course will help serve as your guide. Prerequisites: Some coding skills are necessary. Preferably python, but any other programming language will do fine. Also some basic understanding of math (linear algebra) is a plus, but we will cover that part in the first week as well. If you choose to take this course and earn the Coursera course certificate, you will also earn an IBM digital badge. To find out more about IBM digital badges follow the link ibm.biz/badging.

查看授課大綱

您將學習的技能

Machine Learning, Deep Learning, Long Short-Term Memory (ISTM), Apache Spark

審閱

4.4（436 個評分）

5 stars
288 ratings
4 stars
93 ratings
3 stars
29 ratings
2 stars
9 ratings
1 star
17 ratings

MK

Apr 16, 2018

Useful information course have. There are some challenges.\n\nHowever, the instructor, Romeo is great!\n\nA real Jedi master!

FW

May 18, 2018

Good connection from the theory in Standford University: "Machine Learning" to modern day implementations of ML.

從本節課中

Introduction to deep learning

Linear algebra6:03

Deep feed forward neural networks12:57

Convolutional Neural Networks4:34

Recurrent neural networks1:46

Auto encoders and representation learning2:39

Methods for neural network training8:23

Gradient Descent Updater Strategies6:39

How to choose the correct activation function3:08

The bias-variance tradeoff in deep learning3:32

教學方

Romeo Kienzler
Chief Data Scientist, Course Lead
Niketan Pansare
Senior Software Engineer
Tom Hanlon
Training Director
Max Pumperla
Deep Learning Engineer
Ilja Rasin
Data Scientist

腳本

So, welcome to the lecture about Gradient Descent Updater Strategies.

This is a very important lecture because choosing

the correct Gradient Descent Updater Strategy might heavily influence your learning.

So remember, in Gradient Descent,

we are testing one random value because

that random value is generated by the random initialization of the weights.

And from there, you have to find the next step.

So next step, you find, of course,

by computing the first derivative of

the cost function and you go in direction of the steepest descent.

But there are some catches,

so let's start with Gradient Descent as it is.

So in Gradient Descent,

you update the weights,

theta, by subtracting a value from it.

So, the value which we are subtraction is this one.

So, let's have a look how this is computed.

Again, it starts with the cost function J.

The cost function J is,

of course, dependent on theta,

the weights, but also on the training data.

So the training data,

I have exemplified here using Matrix X,

and the Matrix Y.

So, this is basically the complete data set.

And once this cost function is computed under complete data set,

we take the first derivative,

we get the gradient,

then we multiply by the learning rate.

So, the update would be smaller,

and then we subtract from the actual value of theta_t,

and that gives us theta_t plus one.

So, that's how gradient descent works.

So, there's a variation.

This is called Stochastic Gradient Descent,

and the only difference is that you are not computing the gradient under

complete matrix X and the complete matrix Y.

You just take one training example. This is X_i and Y_i.

So once you have computed the gradient,

you update already the parameters theta.

So, a variation of this is the so-called Mini Batch Gradient Descent,

so that's somehow in between.

You don't use the complete data set and just don't use a single instance of your data,

but just use a batch.

So the batch size, you have to define.

Usually, you take values between 32 and 1024 or so,

and then you compute the gradient for that particular batch.

And once this gradient has been computed,

you update the parameter matrix theta.

So now, another way of doing this is called momentum.

So momentum, the idea is that we take also an update of

a past timestep into consideration for computing the update of the current timestep.

So now, we have here a variable called nu.

So, nu is computed from nu_t minus one.

And usually, we take a gamma of 0-9.

And from this, it basically subtract the actual gradient.

Once you've computed nu_t,

we subtract nu_t from theta_t and we get the updated theta_t plus one.

So, there's a variation of momentum.

This is called Nesterov Accelerated Gradient,

and the only addition here is that in the cost function,

we subtract this term here already from theta_t.

So the usual way, this is like a ball rolling down the hill and it's a smart ball.

So whenever the slope starts to increase,

the ball stops accelerating and breaks a bit.

So Adagrad, which stands for a Adaptive Gradient,

tries to change the learning rate over time and not only for a complete batch.

It tries to change the learner rate for each individual example.

So you see here, this formula is dependent on i.

So, i stands for extra training example.

You'll see here that the learning rate eta

is modified and it's modified by this term here.

So, G_t, ii is the matrix which contains

information on the past gradients per training example.

And taking this into account,

it just reduces the learning rate.

And note that there is little term, e,

so that we avoid division by zero.

This whole thing cannot only be computed per

example but also using matrix-matrix multiplication,

and therefore, we can omit the i,

because G_t is a diagonal matrix.

So, don't worry if you don't understand this,

so maybe you have to revisit the linear algebra basics,

but anyway, it's just the way to compute the whole math in just one go.

So, eta data is only a variation of Adagrad,

and the difference is that it doesn't use a matrix G,

but it continuously computes the mean of the historic gradients.

So there are many, many variations of gradient descent updates,

and I recommend you to have a look at Sebastian's blog later,

where everything is described really nicely.

But the key take home point here is that

the Gradient Descent Updater strategy is very important note you have to tune,

and as usual, tuning a neural network is considered as black magic or trial and error.

So, you just have to try a couple of those.

So, let's actually have a look at Sebastian's blog.

So, Sebastian is this funny guy here.

And you see here,

those are the trajectories of different.

learning course, using different gradient descent updater strategies.

So, it's pretty interesting because it's a very important hyper parameter you can tune.

So, we have covered most of those.

But what I wanted to show you is two daily figures here.

So, those two are really interesting.

So you see here, different trajectories in

optimization space using different gradient descent updaters.

So, the red one here is stochastic gradient descent and you see here in both problems,

it performs really poor.

And it even gets stuck.

So maybe in the first image,

it won't get stuck but it won't converge for ages and in the second,

this is a settle point. It even gets stuck.

And you can see here that for example,

AdaDelta is performing best among all of those, Okay?

I think that's enough for now.

I hope that you now understood that this is

an important hyper parameter that you can tune, and that's basically it.

探索我們的目錄

免費加入並獲得個性化推薦、更新和優惠。

Coursera provides universal access to the world’s best education, partnering with top universities and organizations to offer courses online.

© 2019 Coursera Inc. 保留所有權利。

通過 App Store 下載

通過 Google Play 獲取

Coursera
關於
管理團隊
工作機會
目錄
證書
MasterTrack™ 證書
學位
企業版
政府版

社區
Learners
Partners
Developers
Beta Testers
Translators

連接
博客
Facebook
領英
Twitter
Instagram
YouTube
技術博客

更多
條款
隱私
幫助
內容訪問
媒體
聯繫我們
目錄
附屬公司