**Deep Learning** is a subfield of machine learning concerned with algorithms inspired by the structure and function of the brain called **artificial neural networks**.

If you are just starting out in the field of deep learning or you had some experience with neural networks some time ago, you may be confused. I know I was confused initially and so were many of my colleagues and friends who learned and used neural networks in the 1990s and early 2000s.

The leaders and experts in the field have ideas of what deep learning is and these specific and nuanced perspectives shed a lot of light on what deep learning is all about.

In this post, you will discover exactly what deep learning is by hearing from a range of experts and leaders in the field.

Let’s dive in.

## Deep Learning is Large Neural Networks

Andrew Ng from Coursera and Chief Scientist at Baidu Research formally founded Google Brain that eventually resulted in the productization of deep learning technologies across a large number of Google services.

He has spoken and written a lot about what deep learning is and is a good place to start.

In early talks on deep learning, Andrew described deep learning in the context of traditional artificial neural networks. In the 2013 talk titled “Deep Learning, Self-Taught Learning and Unsupervised Feature Learning” he described the idea of deep learning as:

Using brain simulations, hope to:

– Make learning algorithms much better and easier to use.

– Make revolutionary advances in machine learning and AI.

I believe this is our best shot at progress towards real AI

Later his comments became more nuanced.

The core of deep learning according to Andrew is that we now have fast enough computers and enough data to actually train large neural networks. When discussing why now is the time that deep learning is taking off at ExtractConf 2015 in a talk titled “What data scientists should know about deep learning“, he commented:

very large neural networks we can now have and … huge amounts of data that we have access to

He also commented on the important point that it is all about scale. That as we construct larger neural networks and train them with more and more data, their performance continues to increase. This is generally different to other machine learning techniques that reach a plateau in performance.

for most flavors of the old generations of learning algorithms … performance will plateau. … deep learning … is the first class of algorithms … that is scalable. … performance just keeps getting better as you feed them more data

He provides a nice cartoon of this in his slides:

Finally, he is clear to point out that the benefits from deep learning that we are seeing in practice come from supervised learning. From the 2015 ExtractConf talk, he commented:

almost all the value today of deep learning is through supervised learning or learning from labeled data

Earlier at a talk to Stanford University titled “Deep Learning” in 2014 he made a similar comment:

one reason that deep learning has taken off like crazy is because it is fantastic at supervised learning

Andrew often mentions that we should and will see more benefits coming from the unsupervised side of the tracks as the field matures to deal with the **abundance of unlabeled data available**.

Jeff Dean is a Wizard and Google Senior Fellow in the Systems and Infrastructure Group at Google and has been involved and perhaps partially responsible for the scaling and adoption of deep learning within Google. Jeff was involved in the Google Brain project and the development of large-scale deep learning software DistBelief and later TensorFlow.

In a 2016 talk titled “Deep Learning for Building Intelligent Computer Systems” he made a comment in the similar vein, that deep learning is really all about large neural networks.

When you hear the term deep learning, just think of a large deep neural net. Deep refers to the number of layers typically and so this kind of the popular term that’s been adopted in the press. I think of them as deep neural networks generally.

He has given this talk a few times, and in a modified set of slides for the same talk, he highlights the **scalability of neural networks** indicating that results get better with more data and larger models, that in turn require more computation to train.

## Deep Learning is Hierarchical Feature Learning

In addition to scalability, another often cited benefit of deep learning models is their ability to perform automatic feature extraction from raw data, also called feature learning.

Yoshua Bengio is another leader in deep learning although began with a strong interest in the automatic feature learning that large neural networks are capable of achieving.

He describes deep learning in terms of the algorithms ability to discover and learn good representations using feature learning. In his 2012 paper titled “Deep Learning of Representations for Unsupervised and Transfer Learning” he commented:

Deep learning algorithms seek to exploit the unknown structure in the input distribution in order to discover good representations, often at multiple levels, with higher-level learned features defined in terms of lower-level features

An elaborated perspective of deep learning along these lines is provided in his 2009 technical report titled “Learning deep architectures for AI” where he emphasizes the importance the hierarchy in feature learning.

Deep learning methods aim at learning feature hierarchies with features from higher levels of the hierarchy formed by the composition of lower level features. Automatically learning features at multiple levels of abstraction allow a system to learn complex functions mapping the input to the output directly from data, without depending completely on human-crafted features.

In the soon to be published book titled “Deep Learning” co-authored with Ian Goodfellow and Aaron Courville, they define deep learning in terms of the depth of the architecture of the models.

The hierarchy of concepts allows the computer to learn complicated concepts by building them out of simpler ones. If we draw a graph showing how these concepts are built on top of each other, the graph is deep, with many layers. For this reason, we call this approach to AI deep learning.

This is an important book and will likely become the definitive resource for the field for some time. The book goes on to describe multilayer perceptrons as an algorithm used in the field of deep learning, giving the idea that deep learning has subsumed artificial neural networks.

The quintessential example of a deep learning model is the feedforward deep network or multilayer perceptron (MLP).

Peter Norvig is the Director of Research at Google and famous for his textbook on AI titled “Artificial Intelligence: A Modern Approach“.

In a 2016 talk he gave titled “Deep Learning and Understandability versus Software Engineering and Verification” he defined deep learning in a very similar way to Yoshua, focusing on the power of abstraction permitted by using a deeper network structure.

a kind of learning where the representation you form have several levels of abstraction, rather than a direct input to output

## Why Call it “*Deep Learning*“?

Why Not Just “*Artificial Neural Networks*“?

Geoffrey Hinton is a pioneer in the field of artificial neural networks and co-published the first paper on the backpropagation algorithm for training multilayer perceptron networks.

He may have started the introduction of the phrasing “*deep*” to describe the development of large artificial neural networks.

He co-authored a paper in 2006 titled “A Fast Learning Algorithm for Deep Belief Nets” in which they describe an approach to training “deep” (as in a many layered network) of restricted Boltzmann machines.

Using complementary priors, we derive a fast, greedy algorithm that can learn deep, directed belief networks one layer at a time, provided the top two layers form an undirected associative memory.

This paper and the related paper Geoff co-authored titled “Deep Boltzmann Machines” on an undirected deep network were well received by the community (now cited many hundreds of times) because they were successful examples of greedy layer-wise training of networks, allowing many more layers in feedforward networks.

In a co-authored article in Science titled “Reducing the Dimensionality of Data with Neural Networks” they stuck with the same description of “deep” to describe their approach to developing networks with many more layers than was previously typical.

We describe an effective way of initializing the weights that allows deep autoencoder networks to learn low-dimensional codes that work much better than principal components analysis as a tool to reduce the dimensionality of data.

In the same article, they make an interesting comment that meshes with Andrew Ng’s comment about the recent increase in compute power and access to large datasets that has unleashed the untapped capability of neural networks when used at larger scale.

It has been obvious since the 1980s that backpropagation through deep autoencoders would be very effective for nonlinear dimensionality reduction, provided that computers were fast enough, data sets were big enough, and the initial weights were close enough to a good solution. All three conditions are now satisfied.

In a talk to the Royal Society in 2016 titled “Deep Learning“, Geoff commented that Deep Belief Networks were the start of deep learning in 2006 and that the first successful application of this new wave of deep learning was to speech recognition in 2009 titled “Acoustic Modeling using Deep Belief Networks“, achieving state of the art results.

It was the results that made the speech recognition and the neural network communities take notice, the use “deep” as a differentiator on previous neural network techniques that probably resulted in the name change.

The descriptions of deep learning in the Royal Society talk are very backpropagation centric as you would expect. Interesting, he gives 4 reasons why backpropagation (read “deep learning”) did not take off last time around in the 1990s. The first two points match comments by Andrew Ng above about datasets being too small and computers being too slow.

## Deep Learning as Scalable Learning Across Domains

Deep learning excels on problem domains where the inputs (and even output) are analog. Meaning, they are not a few quantities in a tabular format but instead are **images of pixel data, documents of text data or files of audio data**.

Yann LeCun is the director of Facebook Research and is the father of the network architecture that excels at object recognition in image data called the Convolutional Neural Network (CNN). This technique is seeing great success because like multilayer perceptron feedforward neural networks, the technique scales with data and model size and can be trained with backpropagation.

This biases his definition of deep learning as the development of very large CNNs, which have had great success on object recognition in photographs.

In a 2016 talk at Lawrence Livermore National Laboratory titled “Accelerating Understanding: Deep Learning, Intelligent Applications, and GPUs” he described deep learning generally as learning hierarchical representations and defines it as a scalable approach to building object recognition systems:

deep learning [is] … a pipeline of modules all of which are trainable. … deep because [has] multiple stages in the process of recognizing an object and all of those stages are part of the training”

Jurgen Schmidhuber is the father of another popular algorithm that like MLPs and CNNs also scales with model size and dataset size and can be trained with backpropagation, but is instead tailored to learning sequence data, called the Long Short-Term Memory Network (LSTM), a type of recurrent neural network.

We do see some confusion in the phrasing of the field as “deep learning”. In his 2014 paper titled “Deep Learning in Neural Networks: An Overview” he does comment on the problematic naming of the field and the differentiation of deep from shallow learning. He also interestingly describes depth in terms of the complexity of the problem rather than the model used to solve the problem.

At which problem depth does Shallow Learning end, and Deep Learning begin? Discussions with DL experts have not yet yielded a conclusive response to this question. […], let me just define for the purposes of this overview: problems of depth > 10 require Very Deep Learning.

Demis Hassabis is the founder of DeepMind, later acquired by Google. DeepMind made the breakthrough of combining deep learning techniques with reinforcement learning to handle complex learning problems like game playing, famously demonstrated in playing Atari games and the game Go with Alpha Go.

In keeping with the naming, they called their new technique a Deep Q-Network, combining Deep Learning with Q-Learning. They also name the broader field of study “Deep Reinforcement Learning”.

In their 2015 nature paper titled “Human-level control through deep reinforcement learning” they comment on the important role of deep neural networks in their breakthrough and highlight the need for hierarchical abstraction.

To achieve this,we developed a novel agent, a deep Q-network (DQN), which is able to combine reinforcement learning with a class of artificial neural network known as deep neural networks. Notably, recent advances in deep neural networks, in which several layers of nodes are used to build up progressively more abstract representations of the data, have made it possible for artificial neural networks to learn concepts such as object categories directly from raw sensory data.

Finally, in what may be considered a defining paper in the field, Yann LeCun, Yoshua Bengio and Geoffrey Hinton published a paper in Nature titled simply “Deep Learning“. In it, they open with a clean definition of deep learning highlighting the multi-layered approach.

Deep learning allows computational models that are composed of multiple processing layers to learn representations of data with multiple levels of abstraction.

Later the multi-layered approach is described in terms of **representation learning and abstraction**.

Deep-learning methods are representation-learning methods with multiple levels of representation, obtained by composing simple but non-linear modules that each transform the representation at one level (starting with the raw input) into a representation at a higher, slightly more abstract level. […] The key aspect of deep learning is that these layers of features are not designed by human engineers: they are learned from data using a general-purpose learning procedure.

This is a nice and generic a description, and could easily describe most artificial neural network algorithms. It is also a good note to end on.

## Summary

In this post you discovered that deep learning is just very big neural networks on a lot more data, requiring bigger computers.

Although early approaches published by Hinton and collaborators focus on greedy layerwise training and unsupervised methods like autoencoders, modern state-of-the-art deep learning is focused on training deep (many layered) neural network models using the backpropagation algorithm. The most popular techniques are:

- Multilayer Perceptron Networks.
- Convolutional Neural Networks.
- Long Short-Term Memory Recurrent Neural Networks.

I hope this has cleared up what deep learning is and how leading definitions fit together under the one umbrella.

**If you have any questions about deep learning** or about this post, ask your questions in the comments below and I will do my best to answer them.

If the deep learning is such great algorithm, do you think that other older algorithms (like SVM) are no longer efficient to solve our problems?

I think that SVM and similar techniques still have their place. It seems that the niche for deep learning techniques is when you are working with raw analog data, like audio and image data.

Could you please give me some idea, how deep learning can be applied on social media data i.e. twitter.

Start by defining your problem:

http://machinelearningmastery.com/how-to-define-your-machine-learning-problem/

can you tell what could be research for MS level in deep learning?

This is a common question that I answer here:

https://machinelearningmastery.com/faq/single-faq/what-research-topic-should-i-work-on

first of all I would like to appreciate your effort. This is one of the best blog on deep learning I have read so far.

Well I would like to ask you if we need to extract some data like advertising boards from image, what you suggest is better SVM or CNN or do you have any better algorithm than these two in your mind?

CNN will give better result as compare to svm in image classification

CNN would be extremely better than SVM if and only if you have enough data. The reason that CNN would be better is that CNN work as an automatic feature extractor and you won’t need to bother yourself of feature selection and wondering if the extracted feature would weather work with the model or not. CNN extracts all possible features, from low-level features like edges to higher-level features like faces and objects.

As an Adult Education instructor (Andragogy), how can I apply deep learning in the conventional classroom environment?

Perhaps start by defining your problem clearly:

http://machinelearningmastery.com/how-to-define-your-machine-learning-problem/

Plz, anybody has Matlab code for CNN.

Matlab is not so useful in production environments, it’s more used at school. It’s not really the focus of this site.

Can CNNs perform tasks such as Medical Diagnosis or should they be

combined with another technique such as Reinforement Learning to

optimize performance?

Generally, CNNs are really good at working with image data.

Medical Diagnosis seems like a really broad domain. You may want to narrow your scope and clearly define and frame your problem before selecting specific algorithms.

ECG interpretation may be a good problem for CNNs in that they are images. Another project is the development of a Consultant in Cardiovascular Disease analogous to MYCIN, an Infectious Disease Consultatant developed by Shortliffe & Buchanan @ Stanford ~ 40 years ago which was Rule Based.

So Jason, what is the next discovery after “deep learning”?

No idea Napoleon. Deep learning has enough potential to keep us busy for a long while.

sir plz let me know on what basis cnn is extracting features from an image….

What do you mean exactly?

Okay

Good overview.

Take a look at this:

http://deeplearning4j.org/

It could be a good tool for DL?

Thanks Francesco.

hello, may deep learning apply to use in the stock market ?

What I mean : it doesn’t just only use to draw with old data diagram and use the old model but also write down how is the next day to give the number forecast ?

Hi Jason, deep learning may apply to the stock market.

I am not an expert in finance so I cannot give you expert advice. Try it and see.

You may be interested in this post on time series forecasting with deep learning:

http://machinelearningmastery.com/time-series-prediction-with-deep-learning-in-python-with-keras/

Thank for your reply, I have read some your posts and I am very impressed with your work. About myself , I just start to find out what is this filed and you have many experiences about them. I hope if you have some experiences about the finance especially in stock market…pls help me some reference to learn it by myself or find the “Tribute”as you mentioned 🙂

Thanks Jason, it’s great to have you here.

loved it , thanks for the overview , answered to a lot of my question

I am trying to find a topic for my Master-PHD proposal in Deep Learning in medical diagnosis and just wondering if there is any hot topic in this field at the moment ? and how can I learn more about this special field of Deep Learning

I’m glad to hear it was useful Maisie.

I would suggest talking to medical diagnosis people about big open problems where there is access to lots of data.

Hi maisie

I am looking for some information for my master thesis either about CNN and deep learning in medical diagnosis

So plz if u find anything usefull help me cause I’m new in the feild

Sounds like a great area.

Perhaps start by reviewing recent papers on the topic?

I am already but it seem a bit hard

I am looking for advices so I can continue

i am looking for M tech thesis in this topic…help me explore new areas….

Hi neha, the best person to talk to about research topic ideas is your advisor. Best of luck.

Hi Jason,

Thank you so much for your post. I am trying to solve an open problem with regards to embedded short text messages on the social media which are abbreviation, symbol and others. For instance, take bf can be interpret as boy friend or best friend. The input can be represent as character but how can someone encode this as input in neural network, so it can learn and output the target at the same time. Please help.

Regards

Abbey

Very cool problem Abbey.

I would suggest starting off by collecting a very high-quality dataset of messages and expected translation.

I would then suggest encoding the words as integers and use a word embedding to project the integer vectors into a higher dimensional space.

Let me know how you go.

Hi, thanks for the good overview.

In your opinion, on what field CNN could be used in developing countries?

Because there seems less raw data than developed countries, i couldn’t think of any use of CNN in developing countries.

Sorry Sam, I don’t know.

CNNs are state of the art on many problems that have spatial structure (or structure that can be made spatial).

Anything with images is a great start, domains like text and time series are also interesting.

Hi… I am an average developer in a developing country and my opinion is “yes”… if you find a way to get all these “disconnected” data together than you can help on gathering these data to make it easier for developing countries not to make the same mistakes as developed countries… thus bringing the cost down on “becoming” a developed country without the cost… the “research” exist… the implementation is the problem…

Hello Jason,

a very well and nicely explained article for the beginners.

I would like to ask one question, Please tell me any specific example in the area of computer vision, where shallow learning (Conventional Machine Learning) is much better than Deep Learning.

Great question, I’m not sure off hand. Computer Vision is not really my area of expertise.

This article is useful for learning deep learning .Nice article

Thanks priyanka.

Wonderful summary of Deep Learning – I am doing an undergraduate dissertation/thesis on applying Artificial Intelligence to solving Engineering problems.

Thanks Chris! Good luck with your thesis.

Hi… I am just an average normal developer, but I find this article very informative…

May I please ask one question:

If the “internet” and “line speed” was fast enough, would it mean these algorithms could learn itself or are the “programs” currently limited to human interaction during the learning stage…

So my actual question: the “data” according to me is available -> “internet” BUT do we (humanity currently) already have the computational ability to make “sense” of the data via these algorithms AND are the software developed in such a way to ignore human approval?

The data needed to learn for a given problem varies from problem to problem. As does the source of data and the transmission of data from the source to the learning algorithm.

Dr Jason, this is an immensely helpful compilation. I researched quite a bit today to understand what Deep Learning actually is. I must say all articles were helpful, but yours make me feel satisfied about my research today. Thanks again.

Based on my readings so far, I feel predictive analytics is at the core of both machine learning and deep learning is an approach for predictive analytics with accuracy that scales with more data and training. Would like to hear your thoughts on this.

Thanks. I’m very glad to hear it.

Predictive modeling is a sub-field of machine learning, and is by far the most useful area/the area of interest right now:

http://machinelearningmastery.com/gentle-introduction-to-predictive-modeling/

This article is very interesting and useful for a beginner in machine learning like me.

I am thinking about a project (just for my hobby) of designing a stabilization controller for a DIY Quadrotor. Do you have any advice on how and where I should start off? Can algorithms like SVM be used in this specific purpose? Is micro controller (like Arduino) able to handle this problem?

Thank you in advance

Sounds like a lot of fun.

Consider this process to work through your problem:

http://machinelearningmastery.com/start-here/#process

hi

Is the Deep Learning is suitable for prediction of any diseases like Diabetes using data mining algorithms?

If yes give some ideas to work in it

It may be good, but try a suite of algorithms to see what works best on your problem.

See this post:

http://machinelearningmastery.com/a-data-driven-approach-to-machine-learning/

hi jason are you fine i read your article that help me out and the comment section also

may i know that which is the latest algorithm in deep neural network

i need your help about neural network . i am working on neural network thanks

There are many deep learning algorithms.

The most popular are MLPs for tabular data, CNNs for image data and LSTMs for sequence data.

thanks jason in research base which algorithm you suggest for me to work

I would suggest you pick an area that most excites you.

JASON I WANT TO WORK IN MEDICAL AREA OR IMPLEMENT IN TO MEDICAL SITE

Great! I wish you the best of luck.

does deep learning is a solution of over-fitting problem in machine learning?

No, deep learning methods can overfit like any other.

Thanks for the great article. What is the best approach for classifying products based on product description?

Perhaps use an LSTM or CNN or both combined.

Lots of unnecessary points your explained which make difficult to understand what is actually deep learning is, also unnecessary explanaiton meke me bouring to read the document.

Sorry to hear that.

Jason,

What do you think is the future of deep learning?

How many years do you think will it take before a new algorithm becomes popular?

Hi Anthony,

There is not one algorithm, but a population with the headliners: MLP, LSTM and CNN.

I do not know where we are headed, sorry.

I am a student of computer science and am to present a seminar on deep learning, I av no idea of what is all about…..can I get articles dat can aid Me

This post would be a good start.

I do not have much general material, but if you want to know how to use deep learning to solve problems, start here:

http://machinelearningmastery.com/start-here/#deeplearning

Hi Jason, I have been referring to a few of your blogs for my Machine Learning stuff. One striking feature of your blogs is simplicity which draws me regularly to this place! This is very helpful.:) Talking about Deep Learning vs traditional ML, the general conception is that Deep Learning beats a human being at its ability to do feature abstraction. Is this true?

Also, could you tell me why Deep Learning fails to achieve more than many of the traditional ML algorithms for different datasets despite the assumed superiority of DL in feature abstraction over other algorithms?

Deep learning is great at feature extraction and in turn state of the art prediction on what I call “analog data”, e.g. images, text, audio, etc.

It can be used on tabular data (e.g. spreadsheet of numbers) but this is not it’s sweet spot and often can be beaten by other methods, like gradient boosting.

There is no one algorithm to rule them all, just different algorithms for different problems and our job is to discover what works best on a given problem.

Thanks for your response Jason! I appreciate your clarification.

You’re welcome.

I am wondering that if I use a convolutional neural work in my train model, could I say it is deep learning?

Yes.

Sir , it’s a great review about deep learning

My question is what is the difference between deep neural network and CNN.

Is deep learning is applicable to quantitative data( tabular data)

I made a deep neural network model for bulk quantitative data and get a better result than traditional neural method.

What it means sir ?

Is my deep learning technique right?

A CNN is a type of neural network. It can be made deep. Therefore, it is a type of deep neural network.

Yes, neural nets require all input data to be tabular (vectorized).

I cannot know if your model is right. Evaluate it carefully and compare it to other models.

if we use hierarchal training algorithm such as we use unsupervised learning autoencoder with bottleneck (hidden layer, 10,2,10) for training then use the supervised learning with same autoencoder architecture ( hidden layer, 10,2,10) to tune the unsupervised model parameter (weights, bias). These training processes are performed separately.

now the unsupervised autoencoder works as dimension reduction and extract features. does the supervised model work in the same way and extract the feature or just this step conducted in the unsupervised learning

he supervised model will interpret the features and use them to make predictions.

can i say deeplerning == cnn, Do we have types in deep learning

CNN is a type of deep learning. So is RNN and MLP.

Sir, It is a good intro to deep learning. i’m planning to do phd in diagnosis of heart disease using deep learning. I have data’s of features. I don’t know how to classify those data. Can you please refer some material for numerical data classification using tensor flow.

I would recommend following this process when working through a new predictive modeling problem:

https://machinelearningmastery.com/start-here/#process

hi , i have started in my graduation project , and i am creating a medical system based on data comes from medical sensors , and i hope to use deep learning in detecting the disease or the health case based on that data , but i don’t know from where i should start in deep learning

can you help me in doing it ?

My best advice for getting started with deep learning in Python is here:

https://machinelearningmastery.com/start-here/#deeplearning

Hi Jason, thank you for the excellent overview. May I know how to apply deep learning in predicting adverse drug reactions, particularly in drug-drug interaction?

I recommend this process when starting on a new predictive modeling problem:

https://machinelearningmastery.com/start-here/#process

Hi Jason, Can you please tell me the unsupervised deep learning algorithms available? Please refer some link to learn about it.

Sorry, I don’t have material on unsupervised learning algorithms, I don’t find them useful in practice.

What does “bigger models” mean? Are there more equations in the model? Are there more variables in the model? Are there more for loops? How exactly is the model “bigger”? What’s an example of a “not big model” and why is that worse? And what exactly is meant by the term “model” in this field? I’m still not clear on that. Like what exactly are the characteristics that make something a “model” and why is another general programming algorithm like, say, a loop that divides numbers successively to determine if an integer is prime, not a “model”? To me that sounds like a “model” for determining if a number is prime, so what is meant in this field by “model”? What are the inherent properties that make something a “model”? Is a model a type of algorithm? Is it a class in object-oriented design? What’s a “model” and what;s a “bigger model”?

Too many questions for one comment, sorry.

Ok, let’s start by what exactly is meant by the term “model” in this field? Because I’m still absolutely not clear on what that means. If I write up some code to solve a problem, how do I know whether or not I’ve “built a model”?

A model is often referred to as the weights + structure after the training algorithm has been run on data.

It is the “program” that can make predictions on new data.

And then what’s the “bigger model” you refer to in this article? Are there more weights and more structure in the training algorithm? How is that achieved? Do you plug in more equations and more variables and decision parameters/whatever? How do you know what additional equations and parameters to plug in, and how do you know those are the right ones as opposed to others? Or does “bigger models” simply refer to running on more overall data, and the training algorithm is the same with the same amount of equations/variables/computations per training example?

“Larger Models” refers to deeper (more layers) or wider (more neurons per layer), ultimately more representational capacity.

Best summary for students to learn basics of Deep Learning. Thank you Jason

Thanks.

Thank you very much. It is very good summary about deep learning.

Could you give some algorithms used in deep learning , please.

The three to focus on are: Multilayer Perceptron, Convolutional Neural Network and Long Short-Term Memory Network.

how deep learning can be applied in music ? is it possible to analyze existing music to compose new music by computers ?? If yes what type of algorithm should be used ?

Yes, you could classify music or generate music.

Dear Dr Jason Brownlee, I really found this very useful and helpful for beginners to this domain.

Thanks.

I am familiar with machine learning and neural networks. My expertise is optimization and I am just interested in this field. What do you suggest as a good starting point? I prefer to learn it through experience and see how it works on different cases.

I have some advice for getting started here that might help:

https://machinelearningmastery.com/start-here/#deeplearning

Thanx a lot 🙂

Do you think a deep-learning system could be “taught to read” in a manner similar to a young child? That is:

1. Visual input of the words on each page

2. Coordinated (i.e. synchronized) sound input of those words being read

3. Repeat/reinforce as needed by “reading” the same book multiple times

4. Expand the data set by “reading” a variety of books

…with the ultimate goal & test being to present the system with a book it hasn’t “seen” before, and have it “read” that book via a synthesized voice.

(Never mind the robotics to turn the pages, we can leave that part to a human assist.)

A general system of this form could be constructed, not sure about the “like a child” part though.

i am nazek hassouneh

i am a master student and my thesis in deep learning

i have data set with class label

i will use deep learning for classification

data set is an excel format and will convert to CSV form

data set with 49 attribute and 131,000 records

how can i use deep learning i need tutorial for that

i need specific nueral networks suitable for this data set

Perhaps this tutorial will help you get started:

https://machinelearningmastery.com/tutorial-first-neural-network-python-keras/

thank you very much

Very Usefull Article In Deep Learning

Thanks Alexander.

Thanks Jason. Great article as always.

Glad to hear that.

Hi Jason

loving your articles on ML and NLP! Apologies if this is a daft question but do the extra layers in deep learning models make them more or less transparent? I am thinking in the context of looking for patterns of helpful comments in forum exchanges – would I be able to recognise the features discovered (e.g. as a series of IFTTT rules) or would they just come out as a series of factor weights?

Very new to this so any pointers most welcome

Keep up the good work

best wishes

Mat

Layers add layers of abstraction which makes the model more complex/opaque.

Thanks Jason. I see that people have been making progress on increasing transparency in deep learning https://arxiv.org/abs/1710.09511

I am working my way through your NLP crash course to understand more

Thanks for the link.

Hi ..

It’s a great blog I’ve ever seen , thanks a lot for all your efforts, wish you the best 😊

I am a master student and my thesis is about breast cancer diagnosis , I want to know is deep learning algorithm appropriate for my thesis

and if the answer is yes which one is the best

I’m so confused plz help me 😢

I would recommend starting with CNNs for image data.

Very nice one.I want to use deep learning in tourism sector. I can manage to get the tourists data.

Can you tell me how can i use deep learning in tourism sector.

Thanks

Perhaps start by framing your predictive modeling problem:

http://machinelearningmastery.com/how-to-define-your-machine-learning-problem/

Thanks for sharing these types of soul idea especially for like underresourced country. I am trying to do my thesis by “Object detection by Deep learning for autonomous Vehicle” which method prefer for accomplished my thesis.

Good luck with your project!

a very useful for beginner like me. and right now i’m in doing research with title “analysis on deep learning method for heart disease. which suitable technique i should use ?

Thanks.

Perhaps try a suite of methods and see what works best for your specific dataset.

thanks very much jason for your valuable information , i have question about features extraction ;

i know features selection , but what means by features extraction which is a property of deep Learning.

It refers to methods that create features from analog data like images, audio and text:

https://en.wikipedia.org/wiki/Feature_extraction

There is a lot of overlap with feature engineering.

Want to learn more about deep neural network and its variants. Could you please tell me how?

Yes, start here:

https://machinelearningmastery.com/start-here/#deeplearning

Hello Jason,

I’m a PhD student working on a decentralized IDS (Intrusion Detection System) platform utilizing Big Data, and I’m using machine learning algorithms to detect some signature based attacks.

I’m researching if deep learning would be a good AI method to detect non-signature based attacks (anomalies). I’ve used neural networks to train a specific node to detect a specific signature based attack, but I’m new to deep learning and not sure how apply it into my model.

Would Multilayer Perceptron, Convolutional Neural Network or Long Short-Term Memory Network algorithms applicable at detecting anomalies with gigantic amounts of raw data?

Thanks in advance and great article, very useful.

I would recommend testing a suite of methods to see what works best for your specific dataset.

If i am new to this where can i start , eventhough i read the full article its difficult for me to get some technical terms ? So where can i start if i am starting from scratch?

THANK YOU

You can start here:

https://machinelearningmastery.com/start-here/#deeplearning

What about using Deep Learning in regression problems? Can it be useful for problems like ocean wave forecasting in univariate mode?

It may be. Perhaps try it and see how you go.

thank you Doctor Jason Brownlee

For the great effort you make

I have several questions and I hope to be as broad as we used to

CNN is one of the neural networks that can be very deep but my question here is the code that distinguishes between being a normal neural network and being a deep neural network knowing that it can be used in both cases

I have implemented a deep learning application that predicts the status of my client as he will continue his service or not

The Code

import theano

import theano

import tensorflow

# Importing the libraries

import numpy as np

import matplotlib.pyplot as plt

import pandas as pd

# Importing the dataset

dataset = pd.read_csv(‘C:/Python34/Churn_Modelling.csv’)

X = dataset.iloc[:, 3:13].values

y = dataset.iloc[:, 13].values

from sklearn.preprocessing import LabelEncoder, OneHotEncoder

labelencoder_X_1 = LabelEncoder()

X[:, 1] = labelencoder_X_1.fit_transform(X[:, 1])

labelencoder_X_2 = LabelEncoder()

X[:, 2] = labelencoder_X_2.fit_transform(X[:, 2])

onehotencoder = OneHotEncoder(categorical_features = [1])

X = onehotencoder.fit_transform(X).toarray()

X = X[:, 1:]

# Splitting the dataset into the Training set and Test set

from sklearn.model_selection import train_test_split

X_train, X_test, y_train, y_test = train_test_split(X, y, test_size = 0.4)

# Feature Scaling

from sklearn.preprocessing import StandardScaler

sc = StandardScaler()

X_train = sc.fit_transform(X_train)

X_test = sc.transform(X_test)

# Importing the Keras libraries and packages

import keras

from keras.models import Sequential

from keras.layers import Dense

#Initializing Neural Network

classifier = Sequential()

# Adding the input layer and the first hidden layer

classifier.add(Dense(activation=”relu”, kernel_initializer=”uniform”, units=7, input_dim=11))

# Adding the second hidden layer

classifier.add(Dense(activation=”relu”, kernel_initializer=”uniform”, units=7))

# Adding the output layer

classifier.add(Dense(activation=”sigmoid”, kernel_initializer=”uniform”, units=1))

# Compiling Neural Network

classifier.compile(optimizer = ‘adam’, loss = ‘binary_crossentropy’, metrics = [‘accuracy’])

# Fitting our model

classifier.fit(X_train, y_train, batch_size = 50, nb_epoch = 500)

# Predicting the Test set results

y_pred = classifier.predict(X_test)

y_pred = (y_pred > 0.5)

# Creating the Confusion Matrix

from sklearn.metrics import confusion_matrix

cm = confusion_matrix(y_test, y_pred)

The Data set

This application was found on the internet and its code is (Can I specify what functions (code) that make it a deep learning)

And you have implemented an application for you to learn about God called Your First Machine Learning Project in Python

What functions are used make me know whether this code is deep learning or machine learning

In the past it was possible to put a lot of layers and use a large data size in MLP

MLB is now one of the deep learning algorithms

What code you have completed makes it a profound learning

Sorry for the delay I would like to explain to me the application has been applied using the deep learning algorithm which is the same algorithm application is not using deep learning

You can define deep.

Learn more here:

https://machinelearningmastery.com/what-is-deep-learning/

I’m sorry to urge you Dr. Jason

I would also like a small code showing the use of deep learning about traditional learning

What is “traditional learning” in the context of deep learning?

I mean traditional learning is the algorithms in which we do not use depth but similar in use

Like RNN was used by the production of deep learning idea

But I mean what the code will differentiate between RNN and DNN, knowing that RNN and many of the previous algorithms are deep learning algorithms

Thank you for your patience with me

Generally, any neural network may be referred to as deep learning now. I don’t see a practical distinction other than for marketing.

Dear Mr. Jason, there is one thing I don’t understand. In the article it is said that Deep Learning’s performance get better with more data while other, older algorithms tends to reach a plateau. Can you explain more and give an example about the plateau?

Initially I think the plateau is there because more data can cause overfitting, but after some browsing I found out that more data will decrease the chance of overfitting. It is the number of feature, not the number of data that causes overfitting. The only thing I can think about how more data can create plateau is on heuristic algorithm, which can create more local minima where algorithms can get stuck on.

Sorry, I do not have a direct reference.

Thanks John. I found the article very useful.

I am now confident I know what deep learning is.

Glad to hear it.

A very good blog John. I am a newbie to the field of Deep Learning and this blog has helped me well.

Thanks

I’m glad it helped, also, my name is “Jason”.

Can we detect Malware Infections/DOS/Brute Force Attacks on any Network using Deep Learning?

Probably.

Hi, I want to know what are the deep learning methods using PAC Bayesian. And then compare them with other kind of methods.

would you please clarify for me this problem?

Sorry, I don’t have material on this topic.

https://www.technoonews.com/2018/08/what-is-neural-network.html

technology lover

Thanks for sharing.

My brain just imploded.

Sorry to hear that Sean.

Amazing read and this article has given me a head start to write my Master thesis in this field. Thanks.

I’m glad to hear that.

Hai Sir,

I am Preethi, working with Visual Studio C++. My research problem is related to classification and prediction. Now I am trying to use the deep concept (sparse autoencoder ) in my problem, but I didnt get any C++ package for autoencoder. I found so many python packages, but I have to follow VS C++. OpenCV offers modules for CNN ,not for autoencoders.

Could you pl suggest me any package to use with VS C++?

Waiting for your kind response.

Sorry, I’m not familiar with cpp libraries for deep learning.

Very useful article on Deep Learning

Thanks.

Kindly explain Convolutional Nets Model of Deep Learning in detail.

Sure:

http://machinelearningmastery.com/crash-course-convolutional-neural-networks/

Does Particle Swarm Optimization and ant bee colony algorithms which comes under artificial intelligence is related to deep learning method or unsupervised learning techniques in machine learning

Not really, they are optimization algorithms.

Neural networks are function approximation algorithms.

This article has given me great knowledge on deep learning. currently doing research on data mining. Could you please suggest me how to apply deep learning for cancer classification. Right now I am applying cuckoo search optimization algorithm.

Perhaps try this process:

https://machinelearningmastery.com/start-here/#process

Hey I’m wondering what are your thoughts about AI being implemented into healthcare

Sounds great!

HI,

I am beginner and I want to make a CVML- Computer Vision and Machine Learning lab. What tools and requirement have I need.

Sorry, I don’t know about setting up a lab.

Thanks Jason. This article is so well written and informative.

Thanks.

Thanks. This article is so informative.

I’m glad it helped.

Hi Jason and thank you for such an enlightening article.

What I understood is that the hidden layers act as feature learners from the data. They achieve

this by successively applying nonlinear transformations (with the activation functions) on the input data to map them into a new space: the feature space.

In case of a classification task, the classes become easier (linearly) to separatein this feature space.

What about in the case of regression? What is the function of the hidden layers in a deep neural network model for regression.

Thank you in advance!

I would say: In case of regression, there is the nonlinear transformation of the input data to the feature space and there a linear regression in that new feature space can be applied to aproximate the numerical target variable.

Not quite, the model can learn a non-linear separation of classes.

Same idea, but it learns a non-linear fit for association between the inputs to the output value.

ok, thank you for the response.

Is what I said the case for SVR?

Not quite, an SVM can have a non-linear kernel.

yes sure. It is the non linear kernel that enables the non linear transformation of the input data to the feature space.

Isn’t the same principle as the activation functions in the hidden layers of a deep neural network?

Yes, similar.

Hello Jeson

this is regarding deep learning mini cource. i have done some research and read some material about deep learning and for deep learning mini cource – lession 1 – task , below are the details as requested

10 impressive applications of deep learning methods in the field of natural language processing

1. Text summarization

https://arxiv.org/pdf/1707.02268.pdf

2. Robust real-time detection, tracking, and pose estimation of faces in video streams

https://ieeexplore.ieee.org/abstract/document/1334689

3. Deep Residual Learning for Image Recognition

http://openaccess.thecvf.com/content_cvpr_2016/html/He_Deep_Residual_Learning_CVPR_2016_paper.html

4. Multi-digit Number Recognition from Street View Imagery

https://arxiv.org/abs/1312.6082

5. Automatic whale counting in satellite images with deep learning

https://www.biorxiv.org/content/10.1101/443671v1.abstract

6. A short-term building cooling load prediction method using deep learning algorithms

https://www.sciencedirect.com/science/article/pii/S0306261917302921

7. Deep learning for pixel-level image fusion: Recent advances and future prospects

https://www.sciencedirect.com/science/article/pii/S1566253517305936

8. Reading Text in the Wild

https://link.springer.com/article/10.1007/s11263-015-0823-z

9. Neural Machine Translation by Jointly Learning to Align and Translate

https://arxiv.org/abs/1409.0473

10. Interleaved Text/Image Deep Mining on a Large-Scale Radiology Database for Automated Image Interpretation

http://www.jmlr.org/papers/volume17/15-176/15-176.pdf

going through 2nd lession . Thank you for availing this information.

Highly appreciated !!

Thanks

Very nice!

Hi

Can you help me for my mastery

Study thé properties of deep learning approximation

Yes, you can get started here:

https://machinelearningmastery.com/start-here/#deeplearning

Hi sir ,I have one question that how to train the cifar 10 dataset with opencv and python such that after training it will generate xml file.

My aim is to ” Detect objects using OpenCV an python ”.

As I am new in this field, so please consider me.

Sorry, I don’t have any tutorials on this topic, I may cover it in the future.

hello dear

how can I start with (Deep learning in medical image analysis) for my thesis

thanks

Perhaps start here:

https://machinelearningmastery.com/start-here/#dlfcv

generally, I should start with computer vision after that and the more specific subject is in medical image analysis, is it right …..thanks

I recommend that you start working on your project directly, rather than study subjects to get ready to work on your project.

Perhaps the most appropriate methods will be deep learning models like pre-trained convolutional neural networks.

Hello Everyone my thesis topic is “CLEFT FACIAL AESTHETIC OUTCOME EVALUATION BASED ON DEEP TRANSFER LEARNING” how can I use deep learning and transfer learning model and to combine them and develop cleft facial aesthetic outcome measure system please help me I am really worried about it

Perhaps start with a strong definition of the problem you’re solving.

This framework will help:

http://machinelearningmastery.com/how-to-define-your-machine-learning-problem/

Thanks for the information, now I understand

You’re welcome.

Thanks Jason.

The post really brought me to light about Deep learning. I don’t know if you can be of help for my M.Sc thesis. I intend to use deep learning to obtain sistolic and diastolic data readings from a wearable device then run it through CNN to produce a more accurate value as its output.

The CNN will run on a parallel architecture to accommodate the processing power.

And being a consultant for an ICT firm, i will also want to know if you are open to take up some consultancy contract with the firm? You can reach me on my email: if you are interested.

Thanks for asking, I no longer take consulting projects:

https://machinelearningmastery.com/faq/single-faq/can-you-do-some-consulting

You will find these tutorials very helpful:

https://machinelearningmastery.com/start-here/#deep_learning_time_series

Been reading up a lot on machine learning/deep learning, AI thinking how I can use them to UP my marketing ANTE. Will probably need a tech guy to really do it, but just wanted to get a good grasp about the topic and then I came across yours. Great article!

Thanks James.

Hi James ..your articles are always helping. can you please tell me ..how we can train a DQN model with the help of given dataset

I hope to cover that topic in the future.

I am waiting for your tutorial …….as they are the BEST

Thanks.

Thanks for the article. I read a few more articles and decided to work in Tensorflow for deep learning.

Sounds great Harvey!

Hi jason

Which part of deep learning needs to cogitated to improve deep learning?

Is it approch of weigh choosing or the structure of neurals (number of layers and number of neuron in each layers or relation between each other)…. ?

Which part it is????

Great question. Learning rate for sure.

I outline more methods and a framework here:

https://machinelearningmastery.com/start-here/#better

deep learning in itself is an intense topic the way you have elaborate it is great job.please keep sharing such topic.

Thanks.

how can I start with (Deep learning in bitcoin price prediction) for my thesis

This sounds like a time series problem.

Perhaps start here:

https://machinelearningmastery.com/start-here/#deep_learning_time_series