Saturday, August 4, 2018

Introduction

History

      In 1950

              - Samuel 's checker- playing program

      In 1960

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

      In 1970

               - Symbolic concept induction
               - Expert system & knowledge acquisition bottleneck
               - Quinlan's ID3
               - Natural language processing (symbolic)

     In 1980

            Advance decision tree & rule learning 
            Learning & planning and problem solving 
            - Resurgence of neural network
             Valiant's PAC learning theory
             Focus on experimental methodology

    In 1990   - ML & Statistics

            - Support Vector Machines
            - Data Mining 
            - Adaptive agents & 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 & the reasons behind that

          - New software/algorithms
                 ➡️Neural networks
                 ➡️Deep learning
          -  New hardware
                 ➡️GPU's
          - Cloud Enabled
          - Availability of Big Data

    In 2009   -  Google builds self driving car

    In 2011  - Watson wins Jeopardy

    In 2014  - Human vision surpassed by ML systems

    In 2015  -  Machine translation system driven by Neural networks

Machine Learning   : Definition

     " Learning is the ability to improve  one's behaviour based on experience."

  • Build computer systems that automatically improve with experience.
  • What are the fundamental laws that govern all learning processes?
  • Machine Learning explores algorithms that can  
            -   learn from data/build a model from data
            -   use the model from prediction, decision making or solving some tasks. For Example - Model can be used for prediction, decision making or solving any particular task.

 Tom Mitchell's definition

 "A computer program is said to be learn from experience E with respect to some class of tasks T performance measure P, if it's performance at tasks in T, as measured by P, improves with experience E".


 Components of a learning problem

  • Task : The behaviour or task that's being improved.   
             -  For example : classification, Prediction, acting in an environment.
  • Data : The experiences that are being used to improve performance in the task.
  • Measure of improvement
            -   For example : increasing accuracy in prediction, new skills that were not present initially, improved speed.
 














So this is our learning system. It is a box to which  we feed the  experiences  or the data and there is a problem or a task that require solution & you can also give background knowledge which will help the system & for problem/task the learning program comes up with a answer or a solution & its corresponding performance can be measured .Inside there are two components Learner and Reasoner. See the Learner takes experiences and from it can also takes the background knowledge and from this the Learner builds model and this models can be used by the Reasoner which given a task find the particular solution  to the task.

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

Many domains and applications

Medicine:

  • Diagnose a disease

              ーInput : symptoms,lab measurements, test result, DNA test,.....
              ー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

Financial:

  • predict if a stock will rise or fall
             - in the next few milliseconds.
  • predict if a user will click on an advertisement or not
              - in order to decide which advertisement to show.

Some other applications

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