Introduction to Machine Learning | History of Machine Learning | nptel Videos of machine Learning

Machine Learning History

In 1950s :
  - Samuel's checker-playing program.

In 1960s :
 - Neural network : Rosenblatt's perceptron
 - Pattern Recognition
 - Minsky & Papert prove limitations of Perceptron


In 1970s :
 - Symbolic concept induction
 - Expert systems and knowledge acquisition bottleneck
 - Quinlan's ID3
 - Natural language processing (symbolic)

In 1980s :
 - Advanced decision tree and rule learning
 - Learning and planning and problem solving
 - Resurgence of neural network
 - Valiant's PAC learning theory
 - Focus on experimental methodology

In 1990s ML and Statistics
 - Support Vector Machines
 - Data Mining
 - Adaptive agents and web applications
 - Text learning
 - Reinforcement learning
 - Ensembles
 - Bayes Net learning
           In 1994 : Self-driving car road test
           In 1997 : Deep Blue beats Gary Kasparov

Popularity of this field in recent time and the reasons behind that-
 New software / Algorithms
    - Neural networks
    - Deep learning
 New hardware
    - GPU's
 Cloud Enabled
 Availability of Big Data
                                                2009 : Google builds self driving car
                                                 2011 : Wastson wins Jeopardy
                                                 2014 : Human vision surpassed by ML systems

Learning :- The ability to improve behaviours based on experience.
Machine Learning :- It is explores algorithm that-
                                 Learn from data
                                 build models from data
                                 Model can be use for prediction, decision making or solving tasks.

By Mitchell:-
                 A computer program is said to learn from experience E with respect to some class of tasks T and performance measure P, if its performance on tasks in T and measured by P to improve with experience E.

There are different type of task-
 -Prediction
 -classification
 -acting an environment
           *  Experience E is also called as Data

So the learner takes experience and background knowledge and learn some order and the reasoner works with model and given a new problem or task it can camp up with the solution to the task and performance measure corresponding to this.

Many domains and applications
Medicine :
  Diagnose a disease
     - Input : symptoms, lab measurements, test results, DNA tests, ....
     - Output : one of set of possible diseases, or "none of the above"
 Data mine historical medical records to learn which future patients will respond best to which treatments.

Vision :
   Say what objects appear in an image
   Convert hand-written digits to characters 0....9
   Detect where objects appear in an image

Robot Control :
   Design autonomous mobile robots that learn to navigate from their own experience

Natural Language Processing (NLP)
  Detect where entities are mentioned in NL
  Detect what facts are expressed in NL
  Detect if a product/movie review is positive, negative, or neutral.
                  Speed recognition
                  Machine translation
 Financial :
    Predict if a stock will rise or fall
     - In the next few milliseconds
    Predict if a user will click on an ad or not
     - In order to decide which ad to show

Application in Business Intelligence
  Robustly forecasting product sales quantities taking seasonality and trend into account.
  Identifying cross selling promotional opportunities for consumer goods.
  Identify the price sensitivity of a consumer product and identify the optimum price point that maximizes net profit.
  Optimizing product location at a super market retail outlet.
  Modeling variables impacting customers churn and refining strategy.

Some other applications
  Fraud detection : Credit card Providers
  Determine whether or not someone will default on a home mortgage.
  Understand consume sentiment based off of unstructured text data.
  Forecasting women's conviction rates based off external macroeconomic factors.

How we can go about creating a learner these are the following steps-
  1. Choose the training experience
  2. Choose the target function (that is to be learned)
  3. Choose how to represent the target function
  4. Choose a learning algorithm to infer the target function