DEEP LEARNING SERIES- PART 1

Have you ever wondered how the brain works? One way of understanding it is by cutting open the brain and analyzing the structures present inside it. This however can be done by researchers and doctors. Another method is by using electricity to stimulate several regions of the brain. But what if I say that it is possible to analyze and mimic the brain in our computers? Sounds quite interesting right! This particular technology is known as deep learning.

Deep learning is the technique of producing networks that process unstructured data and gives output. With the help of deep learning, it is possible to produce and use brain-like networks for various tasks in our systems. It is like using the brain without taking it out.  Deep learning is advanced than machine learning and imitates the brain better than machine learning and also the networks built using deep learning consists of parts known as neurons which is similar to biological neurons. Artificial intelligence has attracted researchers in every domain for the past two decades especially in the medical field; AI is used to detect several diseases in healthcare.

Sl.noNameDescriptionExamples
1DataType of data provided to inputBinary(0,1) Real
2TaskThe operation required to do on the inputClassification(binary or multi) Regression(prediction)
3ModelThe mathematical relation between input and output. This varies based on the task and complexityMP neuron(Y=x+b) Perceptron(Y=wx+b) Sigmoid or logistic(Y=1/1+exp(wx+b)) *w and b are parameters corresponding to the model
4Loss functionKind of a compiler that finds errors between the output and input (how much the o/p leads or lags the i/p).Square error= square of the difference between the predicted and actual output.  
5AlgorithmA kind of learning procedure that tries to reduce the error computed beforeGradient descent
NAG
AdaGrad
Adam
RMSProp
6EvaluationFinding how good the model has performedAccuracy
Mean accuracy

Every model in this deep learning can be easily understood through these six domains. Or in other words, these six domains play an important role in the construction of any model. As we require cement, sand, pebbles, and bricks to construct a house we require these six domains to construct a network.

 Now it will be more understandable to tell about the general procedure for networks.

  1. Take in the data (inputs and their corresponding outputs) from the user.
  2. Perform the task as mentioned by the user.
  3. Apply the specific relation to the input to compute the predicted output as declared by the user in the form of model by assigning values to parameters in the model.
  4.  Find the loss the model has made through computing the difference between the predicted and actual output.
  5. Use a suitable learning algorithm so as to minimize the loss by finding the optimum value for parameters in the network
  6. Run the model and evaluate its performance in order to find its efficiency and enhance it if found less.

By following these steps correctly, one can develop their own machine. In order to learn better on this, pursuing AI either through courses or opting as a major is highly recommended. The reason is that understanding those concepts requires various divisions in mathematics like statistics, probability, calculus, vectors and matrices apart from programming. 

       

HAPPY READING!!