Tuesday, March 20, 2018

Artificial intelligence and Machine learning

Attended a meetup on AI and Machine learning conducted by Harinder.


Short notes on same:

AI -> artificially created intelligence

Machine learning is an *important* approach to AI. (There are many approaches out there).

Some applications:

Why Candy crush was so famous ?

       According to the player's habits, the game adjusts itself to ensure that the player stays for long.

Facebook, Google maps etc., makes use of AI.

recaptcha -> use this to crowsource old book's scanning.

youtube suggestions

Spam  detection

Deep learning

Natural language processing.

John Mccarthy coined the term AI.  1959.


many developments followed especially games played by AI.

1997 -> Deepblue played chess
2005 -> self driving car
2011 -> watson jeopardy
2016 -> alphago game

Don't take games lightly :)

2005 -> 135 Exabyte human data
Now -> 100,000 exabytes !!
(1 Exabyte -> 1000 PB -> 1PB=1000 TB -> 1TB=1000 GB)

So much of DATA and so  much of computing power, hence good interest now.

Given x and y whose values are linear , AI can figure out
 y = mx+c
// slope and interceptor

Here, y=mx+c is the model figured out by AI.
It can use the model to get output for any input.

what if scatter plot ?

    make use of squares to get smaller numbers.

Fit the DATA -> important term

Fit the DATA to figure out the MODEL.

machine learning -> supervised
                             -> unsupervised

                              supervised -> regression
                                                 -> classification

supervised -> labelled.
unsupervised -> unlabeled.

Regression -> continuous output
classification -> well defined classes

Machine learning:

1. Fit

                  data -> learning -> model

eg: model -> y = mx+c
2. Predict
                 testdata -> model -> New value (prediction)

So, out of 30 test data

Use 20 test data get the model.

input 10 test data -> use model and get output
compare with actual output.

helps to know whether algorithm is good / bad.

model -> forumla / rules.

test data =================
                                                            -> test data accuracy
actual output   -----------------------

train data  =================
                                                               -> train data accuracy
actual output  -----------------------

(test data is part of train data)

Avoid over fitting (100% data accuracy).
(Remembered) data

4 important points in ML:

1. good data - data needs to be cleaned.
2. Algorithm
3. Features.
4. Parameter tuning.

AI framework :


Machine learning in python:


No comments:

Post a Comment