As we know that AI has been one of the thriving fields in this world. Deep learning is a component of Artificial Intelligence and uses Machine Learning. On a broad view Deep learning involves many techniques using Neural Networks to execute a particular task. Taking input and moving that through a number of hidden layers containing nodes and finally returning the input is the basic thing about neural networks. Deep learning is kinda complex where we need a good deal of mathematics too. In this particular post we will try to learn very basic skeleton of Deep Learning without anything in depth learning to solve real time problems. This post contains :
- Neural networks
- Techniques used in Deep Learning
- How to implement Deep Learning in practice
Not going to dive deep in this topic. Let’s have slight knowledge to get started! Neural networks are artificial neurons inspired by biological neurons. Nodes are nothing but clusters or point where many lines intersect or connect. So this should give a proper understanding what actually a neural networks are. To be more clear check out this image
The circles there in the image are nothing but nodes. That forms complete neural network. Interested to know more about NN checkout here: Neural networks
Techniques used in DL:
- Fully connected neural networks
- Convolutional neural networks
- Recurrent neural networks
- Generative adversarial neural networks
- Deep reinforcement learning
Fully connected neural networks:
Check out the image above. This seems like a fully connected forward neural network. Fully connected means each node connects with all the nodes in the next adjacent layer. Forward means this flow of lines is unidirectional and does not have a arrow pointing in opposite direction.
Convolutional neural networks
In this type we will be using grid type typography to process. More you checkout here: CNN
Recurrent neural networks
In this technique nodes get connected recursively accordingly with the input. Let’s take a basic example when type in ‘h’ possible prediction made by the machine during its learning let be ‘e’ which make it ‘he’ in the same way you continue to type ‘hel’ possible prediction would be p to make it ‘help’.
Generative adversarial neural networks
Generator and discriminator are the two major components we deal with. It consists of two neural networks. One is Generator and other is Discriminator. Please look into this two know more in detail.
Two major components are there in reinforcement learning: Agent and the environment. Agent can take state from the environment. Environment can take action front the agent. Then environment can return reward back to the agent. This could be the basic workflow of Reinforcement learning. But this have variety of application and most advanced technique.
How to implement Deep Learning in practice:
These are the most possible ways to use.
We have many services which help you to implement these directly in to your project using a very little code. The major Services are
- Google cloud platform
- Amazon webservices
- Microsoft congitive services
These make your life simpler in using DL applications like identifying objects when you drop a random image and more… But there isn’t you all done because they hold their own cons as well.
You select the model, train by dumping data and test. That’s it you are done! Platforms:
- Microsoft cognitive services
- Azure Machine Learning
The most difficult way and the best way to learn and implement AI using these libriaries:
- TensorFLow by Google
- CNTK by Microsoft
I think we have gone a little deeper and the few things might be extremely confusing. Please note that this blog only gives a basic idea about DL and does not intend anything in depth knowledge. I personally suggest you to learn from other sources having this blog as checklist.
We will share and learn more on AI diving deep. Get Hyped!
I am a forever learner and this is a copy meant for sharing, learning and collaborating under School of AI mission as one of the AI deans. To know more check out here SchoolOfAI