Ai ML DL App

You might also like

Download as ppt, pdf, or txt
Download as ppt, pdf, or txt
You are on page 1of 24

INTRODUCTION

Artificial intelligence (AI)

 refers to theory and development of


computer systems to perform tasks of
human beings: such as visual
perception, speech recognition, decision
making, translation between languages
etc.
Machine Learning (ML)

Slides:Kilian Weinberger
Program is written taking the data as input (say MP3 file) and output sound
Learn from input and output - Training
How machine learns? By using large amount of data

Cars Motorcycles

Andrew Ng
What is Machine Learning (ML)
Formal Definition by Mitchell, 1997:

A computer (machine) program A is said


to learn from experience E (more the data
better experience) with respect to some
task T and performance measure P if it’s
performance improves with E

Informally: Algorithms that improve on


task performance with experience.
Machine learning (ML)
 Set of approaches that can detect patterns
(infer useful information) in data (pattern
recognition) and then use the revealed
information to predict what may happen in
future or to perform other kinds of decision
making under uncertainty. Availability of big
data has made ML very popular [Kevin
Murphy book]

 It is a branch of artificial intelligence (AI)


technology
Deep Learning (DL)

 New area of Machine Learning


research, which has been introduced
with the objective of moving Machine
Learning closer to one of its original
goals: Artificial Intelligence.”
 That is: Deep learning is a necessary tool
to achieve machine learning for artificial
intelligence. The name Deep learning has
originated from artificial Neural Network
(ANN) which can be seen as a multilayer
ANN
ML/DL applications – In use

 Your email: Separate Spam mails, email


categorization, smart reply – Gmail
 Ridesharing Apps Like Uber – How they
determine price? Optimally match you
with other passengers to minimize long
route
 Self driving cars
 Plagiarism check
 Banking: Mobile check deposit (OCR - Mitek),
Fraud prevention
 Investment advice (Wealthfront, Betterment)
 Social Networking: Facebook: automatically
highlighting faces and tagging friends – face
recognition
 Instagram:
 Snapchat
 Online shopping: Amazon
 Voice to text: Alexa
Few inspirational applications!

Jason Brownlee on July 14, 2016 in Deep Learning


Colorization of black and white movies
Adding sounds to silent movies

 System must synthesize sounds to match


a silent video.

 Work done on sound of a drum stick


striking different surfaces and creating
different sounds.
Paper-https://arxiv.org/pdf/1512.08512.p
df
Autometic translation of text
Automatic image caption generation
 Short listing of CVs of candidates for
interview from multimedia data

Short list based on text, audio and video


contents

Every candidate sends the above to an


organization as an applicant
 Predictive maintenance for industry

Predict the health of automobile parts,


electrical equipments based on past data
from sensors. Don’t go by periodical
servicing.
Optical character recognition

OCR – Given a scanned document detect the


Language and perform translation (say
Hindi/English) for Indian languages.
Very useful when you want to write an application
in your own language
Adversarial Machine Learning: Big Picture

Madry and Schmidt , online


Recent progress in Adv. Robu
Chen, IBM Research – Talk-2020
What are adversarial attacks

Inputs to Machine Learning models that the


attacker designs and creates intentionally
to make the model to make mistake

If the model (machine) has been taught to


learn (trained) to identify some input (say
Pig), the attacker makes small changes
(spoils the input) so that it is identified as
(Airliner!)
Practical attacks
 Fake news (attack): Deep Fakes – Use
(Generative Adversarial Network) GAN to
generate fake news – trolling!

 Video of Obama saying things he never


said, Automated reviews, tweets
indistinguishable from human generated
messages
Why worry about attacks on ML algorithms?

Security in critical tasks – self driving


cars!

Huge money invested in AI applications–


So, revenue loss if adversary fools the
system!
How to handle attacks?

Develop defence algorithms


 More on this in next semester!! In
Adversarial Machine Learning course

You might also like