Scribd
Upload
9 views

CSC 462 AI Week 07 ML

Uploaded by

abdurrehman.12345678912345
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
9 views

CSC 462 AI Week 07 ML

Uploaded by

abdurrehman.12345678912345
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/ 29

CSC 462: Artificial

Intelligence
Instructor: Dr. M. Umar Khan
Focusing on ML

Its the ability to use


previous data
to perform future
actions

Generalize

2
Difficulties in teaching concepts to computers
Traditional Programming AI Winter and Underwhelming Results
Complexity of the world made it hard to encode all rules

Problem space grew too quickly

In Arthur Samuel’s classic 1962 essay "Artificial Intelligence:


A Frontier of Automation", he wrote: “Programming a
computer for such computations is, at best, a difficult task, not
primarily because of any inherent complexity in the computer
itself but, rather, because of the need to spell out every minute
step of the process in the most exasperating detail.
Computers, as any programmer will tell you, are giant
morons, not giant brains.”
Let’s try Machine Learning
Machine Learning
Input Learning Program/
Output Algorithm Model

Essence of Machine Learning:


– There is a pattern in data, however it is complex and difficult to articulate
– We cannot pin it down mathematically (i.e., it is too complex for that)
– Let the computer learn a model for the concept itself using algorithms.
“Learning” in Machine Learning
How do we learn the right weights (model parameters)?

Arthur Samuel Use of mathematical


optimization to reduce
the discrepancy between
predictions and actual
labels (which is computed
by loss function)
Hands-on Activity
Can you train Juha to recognize digits?

https://machinelearningforkids.co.uk/
Supervised learning
Supervised learning: the machine experiences a series of inputs: x1, x2,
x3, x4, … along with the correct labels y1, y2, … and it aims to learn a
mapping so that it can make a correct prediction for a new input

Binary Class Classification Multi-Class Classification

{0, 1, 2, … 9}

Fraud Detection

Toxic Comment Detection


Deep Learning and Deep (Neural) Networks

M. Mitchell Waldrop PNAS 2019;116:4:1074-1077


Various use cases of ML predictions
Unsupervised learning
Unsupervised learning: the goal of the machine is to build a
model of x that can be used for various tasks such as reasoning,
decision making, predicting things, communicating, etc.

Applications: Clustering, anomaly detection, etc.


Generative vs. Discriminative Models
Generative Adversarial Networks

https://thispersondoesnotexist.com/
Reinforcement Learning

Basic idea:
Receive feedback in the form of rewards
Agent’s utility is defined by the reward function
Must (learn to) act so as to maximize expected rewards
All learning is based on observed samples of outcomes!
Different tribes in AI/ML
Big Ideas in Artificial Intelligence
Big Ideas in Artificial Intelligence (Expanded)
Caveat Emptor
Let the Buyer Be Aware
Despite the AI optimism, ML is no panacea

ple fi n d th e lack of rigor and sound


Many pe o
ta n d in g o f th e in ternal working
analytical unders of ML troubling
Open questions related to ML/AI
Bias (Can the ML model learn our biases?)

Interpretability (Can we understand ML models’ “reasoning”?)

Privacy (Is the ML model protecting our privacy?)

Security (Is the ML model secure against adversarial attacks?)

Trustworthiness/Reliability
(Can I trust the prediction of an ML model?)
Overfitting and the importance of generalization
Clever Hans (pictured) “solved”
arithmetic problems by simply
following the cues that inadvertently
emanated from his trainer’s body
language.
The problem of bias: racial bias
Can ML/AI models be Islamophobic?
AI and Leadership
Looking at technology critically
“Every new technology will bite back.
The more powerful its gifts, the more
powerfully it can be abused.”
Kevin Kelly

Ledger of Harms
Unanticipated (possible inadvertent) harms
Ledger of Harm

REWARD FUNCTION GAMING


Developing Safe & Trustworthy AI/ML systems

“ THERE ARE ETHICAL


CHOICES IN EVERY SINGLE
ALGORITHM WE BUILD
Ability To Act Ethically and With Wisdom

An ounce of information is worth a pound of data.


An ounce of knowledge is worth a pound of information.
An ounce of understanding is worth a pound of knowledge.
An ounce of wisdom is worth a pound of understanding.

Wisdom

Understanding

Good for your self/tribe is Knowledge


not necessarily good for
common good of humanity Information
Data
Credits

• https://ieeexplore.ieee.org/abstract/document/9139969
• https://www.researchgate.net/publication/341478640_An_Overview_of_Privacy_in_Machine_Learning
• https://towardsdatascience.com/generative-adversarial-networks-gans-a-beginners-guide-f37c9f3b7817
• https://ojs.aaai.org//index.php/AAAI/article/view/5053

3
Thank you!

You might also like