Its deep since it works on pixel base transformation of image and pkl file having multiple layers is created in supervised / unsupervised classification. high power computer system is required to construct the decision model.
When used for image recognition, convolution neural networks (CNNs) consist of multiple layers of small neuron collections which process portions of the input image, called receptive fields. The outputs of these collections are then tiled so that their input regions overlap, to obtain a better representation of the original image; this is repeated for every such layer. Tiling allows CNNs to tolerate translation of the input image.
the following practice would provide you more assistance
http://www.robots.ox.ac.uk/~vgg/practicals/cnn/
https://www.kaggle.com/c/integer-sequence-learning
MINIST Digit problem, Dog Vs Cat are some suitable examples.
This is only with respect to image processing. Please provide a generalist view on deep learning.
The two hidden layers and the single output layer each have an array of associated bias values, named aBiases, bBiases, and oBiases respectively. The local outputs for the hidden layers are stored in class-scope arrays named aOutputs and bOutputs. These two arrays could have been defined locally to the ComputeOutputs method.
Please see example on
https://visualstudiomagazine.com/Articles/2014/06/01/Deep-Neural-Networks.aspx?Page=2
https://jamesmccaffrey.wordpress.com/2014/06/19/deep-neural-networks-using-c/
Dear Jayaram,
Deep means that the NN has many hidden layers. In 1990's the researcher had a problem to train NN with many hidden layers because the initialization weights of NN and also the hardware at that time is very limited. Now, the hardware is not the problem, but the initialization weights is still a problem and researched by some people.
From image classification, convolutional neural network usually used as the classifier and it trained using only SGD algorithm (Stochastic Gradient Descent). For CNN weights initialization is not the main problem, but for general NN it the main problem. Some approach that most used for the weights initialization method is RBM (Restricted Boltzmann Machine) or autoencoder.
The deepest NN comes from image classification method called deep residual convolutional neural network with 152 layers develop by Microsoft Asia Lab. Convolutional neural network (CNN) is quite different from original NN which CNN doesn't fully connected between layers (maybe that also why is very easy to train CNN). You can refer to the following paper for further explanation.
http://arxiv.org/abs/1512.03385
Best Regards,
Reza Fuad Rachmadi
It is kind of you for your answer. Is deep learning applicable only to neural networks.? Are there any other algorithms that belong to deep learning category?
Dear Mr. Jayaram,
Thank you for you reply.
As far as I know the term of deep learning is only used in neural network classifier or graph based classifier (because neural network just a graph with some operations in each node).
The modern neural network is very different from the neural network on 1990's. Right now, we do anything using neural network. Unlike the 1990's neural network, the output of the NN can be real values which very convenient for a lot of problems. We can also extrapolate the input using neural network (like super resolution problem in image processing).
Hopes it answer your question.
Thank you.
Best Regards,
Reza Fuad Rachmadi
Deep learning follows the classic hierarchical brain model for learning, which is a hierarchy of layers stacked one after another. Information is processed in one layer before is handled to a subsequent layer in a hierarchy. The more layers you use the deeper is your model; and for some extent, the better is the system accuracy.
Neural networks is mostly used for deep learning. However, there are other algorithms that are gaining popularity as well such as probabilistic graphical models and Bayesian frameworks.
Deep learning (also known as deep structured learning or hierarchical learning) is part of a broader family of machine learning methods based on learning data representations, as opposed to task-specific algorithms.
(wikipedia)
How do I start with deep learning?Let's GO!
One of our latest analysis showed that surface and deep learners are characterised as per their metacognition levels. The IR4.0 prerequisites all of us to move into deep learning as per level of surface learnings. The learnjng theories are needed to be revised too as per need in IR4.0. Though the Digital bloom‘s has been prooosed but needed to be again aligned as per requirement to be defined for surface and deep levels.
- Badges
- Science topic