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

headache

cold cough

vomiting

fever

skin rash

Human decision making

How do humans make decisions from past experiences?

(c) One Fourth Labs

fever

headache

cold cough

vomiting

Rash

Inputs

Output

Human decision making

How do humans make decisions from past experiences?

(c) One Fourth Labs

fever

headache

cold cough

vomiting

Rash

inputs

output

1	0	0	0	0
1	0	0	1	0
1	1	1	0	1
1	0	1	1	0

0
0
1
0

Human decision making

Is this applicable in multiple domains ?

(c) One Fourth Labs

LBW out

Impact

Height

no ball

shot attempted

Pitching in line

Inputs

Output

Human decision making

Is this applicable in multiple domains ?

(c) One Fourth Labs

LBW out

Impact

Height

no ball

shot attempted

Pitching in line

Inputs

Output

1	1	0	1	0
1	1	1	0	0
1	1	1	0	1
1	0	1	1	0

0
1
1
0

Features and Rules

What is the semantics of decision making?

(c) One Fourth Labs

Output

Inputs

headache

cold cough

vomiting

fever

skin rash

Fever

Rash

Cold

(c) One Fourth Labs

if HIGH_FEVER and COLD:
    return no_dengue
elif HIGH_FEVER and VOMITING and not RASH:
    return no_dengue
elif HIGH_FEVER and VOMITING and RASH:
    return dengue

Output

Inputs

Fever

Rash

Cold

Expert Systems

How do we outsource this to a machine?

headache

cold cough

vomiting

fever

skin rash

Use-cases

Where have expert systems been used - past and recent applications

(c) One Fourth Labs

(1990)

(1965)

(1995)

(2018)

Hanane et. al., Expert Systems, May 2018.

Limitations

Do we need to look beyond expert systems?

(c) One Fourth Labs

Task of hiring someone for a job

10th marks

12th marks

graduate

CGPA

projects

if 10th>90 and 12>90 and GRADUATE:
    return hired
elif 10th>80 and 12th>80 and CGPA>7:
    return hired
elif 10th<70 and 12th<70:
    return not_hired

10th

12th

graduate

Output

YES / NO

Inputs

Limitations

Do we need to look beyond expert systems?

(c) One Fourth Labs

Task of hiring someone for a job - involving many factors

10th marks

if 10th>90 and 12>90 and GRADUATE:
    return hired
elif 10th>80 and 12th>80 and CGPA>7:
    return hired
elif 10th<70 and 12th<70:
    return not_hired

10th

12th

graduate

12th marks

graduate

CGPA

projects

awards

.....

10th marks	12th marks	graduate	CGPA	projects	...........	awards
92	82	yes	9.2	3	...........	2
83	75.2	yes	7.2	1	...........	0
75	70	no	6.4	0	...........	1
96	95	yes	9.5	5	...........	4
90	89	yes	8.8	2	...........	1
78	82	yes	7.6	0	...........	0
86	88	yes	8.4	1	...........	1

Lots of data to make sense from

Output

YES / NO

publications

Limitations

Do we need to look beyond expert systems?

if 10th>90 and 12>90 and GRADUATE:
    return hired
elif 10th>80 and 12th>80 and POST_GRADUATE:
    return hired
elif 10th<70 and 12th<70:
    return not_hired
elif awards>2 and GRADUATE:
    return hired
elif not POST_GRADUATE and publications>1:
    return hired
elif projects>10 and publications>2:
    return hired
elif competitions>3 and publications>3:
    return hired
elif awards>5:
    return hired
elif GRADUATE and not POST_GRADUATE and awards>5:
    return hired

The rules can be very complex

(c) One Fourth Labs

10th marks	12th marks	graduate	CGPA	projects	...........	awards
92	82	yes	9.2	3	...........	2
83	75.2	yes	7.2	1	...........	0
75	70	no	6.4	0	...........	1
96	95	yes	9.5	5	...........	4
90	89	yes	8.8	2	...........	1
78	82	yes	7.6	0	...........	0
86	88	yes	8.4	1	...........	1

Limitations

Do we need to look beyond expert systems?

(c) One Fourth Labs

The rules can be sometimes inexpressible

10th marks

12th marks

graduate

post graduate

projects

Output

YES / NO

Inputs

I saw honesty in his eyes

Limitations

Do we need to look beyond expert systems?

The rules can sometimes be unknown:

- task of predicting Ebola

Output

high fever

headache

sore throat

weakness

nausea

YES / NO

Inputs

(c) One Fourth Labs

Say Hi to Machine Learning

How to move from writing rules to learning rules?

def f(x):
    if 10th>90 and 12>90 and GRADUATE:
        return hired
    elif 10th>80 and 12th>80 and POST_GRADUATE:
        return hired
    elif 10th<70 and 12th<70:
        return not_hired

10th

12th

graduate

Output

YES / NO

def f(x):
    i=0
    max_epochs=100
    w = rand()
    while(i<max_epochs):
        dw = grad(w,w*x)
        w = w - lr*dx
        i += 1
    return w*x

10th marks	12th marks	graduate	CGPA	projects	...........	awards
92	82	yes	9.2	3	...........	2
83	75.2	yes	7.2	1	...........	0
75	70	no	6.4	0	...........	1
96	95	yes	9.5	5	...........	4
90	89	yes	8.8	2	...........	1
78	82	yes	7.6	0	...........	0
86	88	yes	8.4	1	...........	1

	hire
	don't hire

\( \hat{y} = f(x_1,x_2)\)

Data, Democratization, Devices

Why has Machine Learning been so successful ?

(c) One Fourth Labs

Democratized model and Learning Algorithms

Abundant data

Relatively fast and cheap cloud/computing

Different roles in the ML world

How this relates to you ?

(c) One Fourth Labs

Collect, curate data

ML Engineering

ML Research

Sources are mentioned in the last slide.

Fast

Efficient

Better

Define Tasks

Takeaway

How do we contrast expert systems from machine learning?

(c) One Fourth Labs

if 10th>90 and 12>90 and GRADUATE:
    return hired
elif 10th>80 and 12th>80 and POST_GRADUATE:
    return hired
elif 10th<70 and 12th<70:
    return not_hired
elif awards>2 and GRADUATE:
    return hired
elif not POST_GRADUATE and publications>1:
    return hired
elif projects>10 and publications>2:
    return hired
elif competitions>3 and publications>3:
    return hired
elif awards>5:
    return hired
elif GRADUATE and not POST_GRADUATE and awards>5:
    return hired