Scribd
Upload
27 views21 pages

2.concept Learning

Uploaded by

aditya332006
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
27 views21 pages

2.concept Learning

Uploaded by

aditya332006
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 21

Chapter 1:

Concept Learning

Adopted from Machine Learning, Mitchell


Logistics
• Course Instructor: Tanmoy Chakraborty (NLP)
https://tanmoychak.com/

• Guest Lecture: TBD (possibly from the industry)


• TAs: Sahil, Aswini, Palash, Prottoy, Vaibhav, Soumyodeep, Anand

• Course page: https://lcs2-iitd.github.io/ELL409-2401/


• Discussion forum: Piazza (https://piazza.com/iitd.ac.in/fall2024/ell409)
Access Code: ell409mli

• For assignment submission: Moodle


• Group Email: [email protected]
Course Directives
• Class Time: Mon and Thu, 8:00 AM - 9:20 AM
• Office Hour: as per requirement (email me to schedule an
appointment)
• TA Hour: (Please email at least an hour before to confirm the meeting location)
• Monday 4 PM to 5 PM: Vaibhav ([email protected])
• Tuesday 4 PM to 5 PM: Soumyodeep ([email protected])
• Wednesday 4 PM to 5 PM: Sahil ([email protected])
• Wednesday 3 PM to 4 PM: Aswini ([email protected])
• Thursday 4 PM to 5 PM: Palash ([email protected])
• Friday 3 PM to 4 PM: Anant ([email protected])
• Room: LH114
Timeline
• Project Finalization: 10/08/2024
• Quiz 1: 12/08/2024
• Assignment 1: 13/08/2024
• Quiz 2: 05/09/2024
• Mid-Term: 12/09/2024 - 18/09/2024
• Assignment 2: 20/09/2024
• Assignment 3: 17/10/2024
• Quiz 3: 21/10/2024
• Quiz 4: 11/11/2024
• Major: 16/11/2024 - 23/11/2024
• Project assessment: Before endsem
Some announcements

• Coding practice twice a month (led by the TA) – outside the regular
lecture hours
Some announcements

• Coding practice twice a month (led by the TA) – outside the regular
lecture hours

• Sample questions for practice before midterm and major


Some announcements

• Coding practice twice a month (led by the TA) – outside the regular
lecture hours

• Sample questions for practice before midterm and major

• Quiz every class - 8:00 AM to 8:05 AM


Outline
Training Examples for EnjoySport
Concept Learning
Inferring a Boolean-valued function from training
examples of its input and output
Representing Hypotheses
Notations
• Instances: The set of items over which the concept is defined

• Target concept (c): The concept to be learned

• Hypothesis (h): A supposition or proposed explanation made on the


basis of limited evidence (training set)

• Hypotheses Space (H): The set of all possible hypotheses


Prototypical concept learning task
Concept Learning as a search
• The task of searching through a large space of hypotheses
• Goal: Find the hypothesis that best fits the training example
• How many distinct instances are possible?

• How many systematically distinct hypotheses are possible?


• How many semantically distinct hypotheses are possible?
General-to-specific ordering of hypotheses
• h1= <Sunny, ?, ?, Strong, ?, ?>
• h2= <Sunny, ?, ?, ?, ?, ?>

• For any instance x in X and hypothesis h in H, we say that x satisfies h


if and only if h(x) = 1.
Hasse diagram
Find-S Algorithm
Find-S Algorithm

At each step, h is the most/least specific/general hypothesis consistent with the training examples observed
to this step
Find-S Algorithm – ignore negative instances
• Ignores every -ve training instances!
• However, the current hypothesis is already consistent with the -ve
example
• As long as we assume that H contains a hypothesis that describes
target concept and the training data is correct, it never requires to
consider –ve examples

• Why?
Complaints about Find-S
• Has the learner converged to the current target concept?
• No way to determine if it has found the only hypothesis that is consistent with
the target concept
• Or there are many other consistent hypotheses as well

• Why prefer the most specific hypothesis?


• In case of multiple hypotheses consistent with the target concept, why to
consider the most specific one?
• Are the training example consistent?
• What if a few training instances are corrupted?

• What if there are several maximally specific consistent hypotheses?


• Find-S should be backtracked to generalize the hypothesis

You might also like