We as human may have often wondered if the intelligence of human can be copied and machines can work the same way as us.
While it is still a distant dream but we are not very far away. In the path to artificial intelligence lets have an overview of what it really means and how data science is helping us achieve it.
Artificial Intelligence: It is an important science that actually helps in daily activities nowadays. The end goal of any machine learning or deep learning algorithm is achieving artificial intelligence.
Artificial intelligence means intelligence of machines wherein they can be as intelligent as to take decisions, recommend actions, solving problems and get important information for use.
From self-driving cars to Siri in your phone to the Netflix recommendations, everything is driven by artificial intelligence.
Machine Learning: Machine learning is one of the subset of artificial intelligence where main aim of it is to learn from data that is given to it as input and predict the continuous values or classify the set of data into some class.
These problems are called regression and classification problems respectively.
Machine learning can be of 3 types:
1) Supervised: In supervised learning we have labeled data, as in we have data which has predicted value to train from.
There can be 2 types of supervised learning:
In this kind of supervised learning a set of independent data is labeled with a output class.
In above example you can see the independent features like chest pain, blocked arteries and blood pressure can determine if a person has heart disease or no, which is labeled data of classification.
When a model is trained on this kind of data, it can be used to predict the class of heart disease or No.
Examples of algorithms used for classification are logistic, Support Vector Classifier, Decision Tree classifier and many more.
In this kind of supervised learning a set of independent data is labeled with a output continuous predicted value.
In above example house age, total bedrooms, population, households, family income are independent features which determine the value of house, and while data can be trained on it, this model can be used to predict the house prices of various houses.
Examples of algorithms used for classification are linear, Support Vector regression, Decision Tree Regression and many more.
In this kind of learning there is no labeled data, i.e. output label is not present, and we need to determine the relation between the data that is given. Mostly the problems like these are solved by Clustering or Grouping, i.e. that data is grouped to form the clusters which are related to each other.
· Grouping or Clustering
A very useful example of it can be Customer Segmentation in any industry wherein, company tries to classify its customers into future customers of their products based on their income, relationship status, age, and other factors.
Another example of it can be houses preferring a reputed food chain outlet in a vicinity.
The branch which is nearest to the houses will be preferred and houses can be clustered based on the locality and distance from that branch.
3) Reinforcement Learning
Also known as semi-supervised learning. It is the kind of learning which happens in an environment, by doing some actions, if action is correct some rewards are given and if incorrect then some observations are sent to the system to learn from it, the whole idea of the system is to concentrate on more and more awards by taking actions which are learned and better in each iteration.
Example of Reinforcement Learnings can be in robotics or AI app driven car. Where learning happens through feedbacks or observations and model is improved.
There are 2 kinds of reinforcement learning
· Positive: In this type of learning, the actions by which awards are given are improved and strengthened which results in reward every time, hence maximizing the performance.
· Negative: In this type of learning, the actions because of which the observation which are negative are avoided or stopped, which improves the performance every time.
Deep Learning phenomenon was developed when artificial intelligence had to be so deep that it could actually imitate human brains as neural networks.
The machine leaning by which very complex problems can be learnt in a same way as human learns by being in environment and repeatedly seeing or hearing and analyzing things can be termed as Deep Learning.
Human Brain has various functions as in the image below:
For each function and part artificial intelligence programs are getting developed as for:
Occipital Lobe- Convolutional Neural Networks (CNN) or Deep Residual Learning (ResNet) for visual recognition
Temporal Lobe-Recurrent Neural Networks (RNN) or Deep Neural Networks (DNN) for speech recognition, Represent Learning (RL) for dialogue understanding. Text Analysing and NLP(Natural Language processing ) are expanding fields these days.
Motor Cortex- Robotics
A simple neural network is a set of algorithms that tries to recognize the relationships in a set of data through a process that closely resembles the way the human brain operates.
It basically refers to neurons of brains and how they act if given a situation.
The diagram above, has an input layer which resembles as given a set of data or situation in which neuron has to function.
Hidden layer has neuron and its internal working which can be in 2 steps.
Step 1) Analyzing inputs x and giving them weights w. It multiplies each input with weight and adds a bias b to it.
Y =∑ xiwi +b which is very similar to solving a linear equation y=mx+c
Step 2) Activates neuron- This function determines if the neuron has to be active to respond to the given situation. It can use any function like Sigmoid, Tanh, Relu.
The value of it will be between 0 to 1 .
The more the value near to 0 means the neuron has to be inactive and more towards 1 means it has to be active.
I will take this topic of internal functioning of hidden layer in neural