1.2 Expert Systems
How rules-based systems can be used for binary classification
Human decision making
How do humans make decisions?
(c) One Fourth Labs
Dengue
Should I do a PhD
LBW
Image of
doctor
Thermometer
headache
cold cough
vomiting
Now show various combinations of input which will lead to a yes/no for a dengue and show tick/cross for these inputs
* 1) Use google image search
2) Click on Tools--> Usage rights --> label for reuse (no copyright issues)
3) Use cartoon images or very HD images
4) Use images which match the white background of the slide or have transparent background
Human decision making
How do humans make decisions?
(c) One Fourth Labs
Dengue
Should I do a PhD
LBW
Image of
doctor
headache
cold cough
vomiting
Now show various combinations of input which will lead to a yes/no for a dengue and show tick/cross for these inputs
* 1) Use google image search
2) Click on Tools--> Usage rights --> label for reuse (no copyright issues)
3) Use cartoon images or very HD images
4) Use images which match the white background of the slide or have transparent background
Human decision making
How do humans make decisions from past experiences?
(c) One Fourth Labs
Image* of Mosquito
Image of
doctor
Thermometer
headache
cold cough
vomiting
Image of Rash
You will now show the same data that you showed on the previous slide but now in a matrix format, one row at a time
inputs
output
Human decision making
Is this applicable in multiple domains ?
(c) One Fourth Labs
Image of Lbw
Image of
umpire
Impact
height
no ball
shot attempted
Pitching in line
You will now show some data for which the player was lbw and some data for which
inputs
output
Features and Rules
What is the semantics of decision making?
(c) One Fourth Labs
A human sitting inside this box with rules (if else code) in his head
Thermometer
headache
cold cough
vomiting
Image of Rash
if (high_fever == false) and rash == false :
return no_dengue
//show this as a function which takes x as the input
// y = f(x)
//This is a code block you can edit it by
//double clicking it
// Also change bg color etc as you want
// and highlight syntax by specifying
// language in settingsOutput
Inputs
Expert Systems
How do we outsource this to a machine?
(c) One Fourth Labs
A computer inside this box
Show an animation that the if-else rules get transferred from human to machine as decision.c
Thermometer
headache
cold cough
vomiting
Image of Rash
if (high_fever == false) and rash == false :
return no_dengue
//This is a code block you can edit it by
//double clicking it
// Also change bg color etc as you want
// and highlight syntax by specifying
// language in settingsOutput
Inputs
Human here
Use-cases
Where have expert systems been used - past and recent applications
(c) One Fourth Labs
Preterm birth risk assessment*
(1994)
Space Shuttle Launch (1990)
Mortgage Loan Advisor (year)
* The titles will go away, the image should convey the title, only year will remain
See https://en.wikipedia.org/wiki/Expert_system#Applications for the applications listed above
- Confirm that these are indeed Expert systems and not ML based systems
Recent Application1 (preferably chat system)
Recent Application 2
Limitations
Do we need to look beyond expert systems?
(c) One Fourth Labs
Think of a decision system which involves many many parameters (say hiring someone for a job)
if (high_fever == false) and rash == false :
return no_dengue
//This is a code block you can edit it by
//double clicking it
// Also change bg color etc as you want
// and highlight syntax by specifying
// language in settingsOne-by-one animation (this blue box is the machine)
1) There will be too many features here (show animation that instead of 5 now you have 100 features)
2) Lots of data to make sense from (replace the LHS by a large matrix)
3) show that in the human head the rules are very very complex making it hard to enumerate (very big if-else code) ... show the google example of broom v/s a dog breed which has broom like hair
4) now show that sometimes the human brain itself does not have rules... show example of ebola
5) and sometimes the rules are inexpressible (for example, I saw truth in his eyes or I did not like the way he sat down)
Human here
(c) One Fourth Labs
if (high_fever == false) and rash == false :
return no_dengue
//This is a code block you can edit it by
//double clicking it
// Also change bg color etc as you want
// and highlight syntax by specifying
// language in settingsHuman here
On LHS show the large data matrix from the previous slide (show x and y)
Now put a cross on the human and the rules in his head and the explicit C program fed to the computer, delete the human and the rules in his head and make the c program empty
now feed this data to the machine and show that it will discover patterns from it (show a lens revolving over the data)
(slowly y = f(X) should be learnt by the machine ... the code inside f(x) should just be ....)
( now show that inside f(x) you could have differernt functions sin(x) , sigmoid(x), some complex function similar to the one that we have in the tower-building sldies)
Say Hi to Machine Learning
How to move from writing rules to learning rules?
Data, Democratization, Devices
Why has Machine Learning been so successful ?
(c) One Fourth Labs
Democratized Algorithms
Show instagram, pinterest icon
Show facebook, twitter icons
Show social media graphs
Show youtube, vimeo icons
show facebook, amazon, myntra and othe ricons
icons of populatr blogs, news papers, techcrunch, medium, etc
Show GPUs, tablets, phones, compute devices
Abundant data
Fast and cheap compute
Different roles in the ML world
How this relates to you ?
(c) One Fourth Labs
Show the gears and data on LHS
Show a pyramid:
- On the bottom most level show all the icons from the previous slide [Collect, curate data]
- Second level show all the bolts and nuts from previous slide [ML Engineering]
-Adding new nuts and bolts [ML Research]
Takeaway
How do we contrast expert systems from machine learning?
(c) One Fourth Labs
LHS show a diagram for expert systems with rules in head transferred to code on computer [a subset of the diagram that you drew earlier]
// show a collage of applications of expert systems
RHS show a diagram of ML [a subset of the diagram shown on say Hi to ML]
// show a collage of applications of ML algorithms many many as compared to ES