The document discusses how deep learning and neural networks can be used to quantify and classify complex types of input like images of dogs and cats. It explains that objects can be decomposed into characteristic pieces that can then be represented numerically, and a neural network can learn the patterns in these pieces to distinguish between categories like dogs and cats. Diagrams and examples are provided to illustrate how neural networks work by choosing weights to minimize error in classifying input data.
10. Where is the maximum of this function?
A.
B.
C.
D.
x
y
11. Where is the maximum of this function?
A.
B.
C.
D.
x
y
12. hello,
world
EKR
If you can do that kind of
math, you can learn to
harness the power of deep
learning.
That’s Just Math!
13. Going deeper with
Convolutions
Google used a new variant of
convolutional neural network called
“Inception” for classification, and for
detection the R-CNN [5] was used. The
results and the approach that Google’s
team took are summarized here [2, 3].
Google’s team was able to train a much
smaller neural network and obtained
much better results compared to results
obtained with convolutional neural
networks in the previous year’s
challenges.
1: Computers can recognize objects
14. Show and Tell: A
Neural Image Caption
Generator
Automatically describing the content
of an image is a fundamental problem
in artificial intelligence that connects
computer vision and natural language
processing. In this paper, we present a
generative model based on a deep
recurrent architecture that combines
recent advances in computer vision
and machine translation and that can
be used to generate natural sentences
describing an image.
2: Computers can write image captions
15. Human-level control
through deep
reinforcement learning
We tested this agent on the challenging
domain of classic Atari 2600 games. We
demonstrate that the deep Q-network
agent, receiving only the pixels and the
game score as inputs, was able to surpass
the performance of all previous algorithms
and achieve a level comparable to that of a
professional human games tester across a
set of 49 games, using the same algorithm,
network architecture and hyperparameters.
3: Computers can play video games
16. 4: Computers can evolve better models
Mar I/O: Evolving
Neural Networks
through Augmenting
Topologies
Neural networks can be combined
with other machine learning
techniques to solve complex
problems like model selection. Mar
I/O is a program made of neural
networks and genetic algorithms
that kicks butt at Super Mario
World.
17. 5: Computers can dream
Twitch Deep
Dream
Visualization
A convolutional neural
network trained to recognize
objects in images can be
run backwards to dream.
Over the spring of 2015, this
dreaming neural network
was streamed online.
19. Simple Example: Draw a line that
separates blue dots from green dots. Do
this by choosing a slope.
x
y y y y
A. B. C. D.
y = 0.25 * x y = 0.50 * x y = 1.00 * x y = 3.00 * x
20. x
y y y y
A. B. C. D.
y y y = 1.00 * x y
Simple Example: Draw a line that
separates blue dots from green dots. Do
this by choosing a slope.
21. x
y y y y
A. B. C. D.
y y y = 1.00 * x y
Option C. minimizes the classification
error of the example. This value of the
slope “learns” how to separate green
dots from blue ones!
22. x
y y y
A. B. C. D.
y y y = 1.00 * x y
Option C. minimizes the classification
error of the example. This value of the
slope “learns” how to separate green
dots from blue ones!
y
23. y The same simple
problem statement of
“quantifying” the input
and separating it
“optimally”
Can allow you to
differentiate between
complex types of input
29. Objective: Choose w00, w01, w02 so that is
y00 close to y. Minimize the
“approximation error.” By doing that,
you’ve learned to reproduce the original
function.
y00 = φ(w00 * x00 + w01 * x01 + w02 * x02)
30. It’s ruff, but it’s possible,
let me explain.
How do you
quantify a
dog?
31. Dogs are composed of
pieces of things that are
characteristically dog.
Cats are composed of cat
pieces.
You can quantify the way
the pieces look and
distinguish between
them.
32.
33. air & hope air & hope head head head air & hope air & hope
air & hope floppy ear cute eye
furrowed
brow
cute eye floppy ear air & hope
floppy ear floppy ear dog cheek
snout
piece
dog cheek floppy ear floppy ear
floppy ear
fur &
stuff
dog cheek wet nose dog cheek
fur &
stuff
air & hope
air & hope
fur &
stuff
mouth mouth mouth
fur &
stuff
air & hope
air & hope
fur &
stuff
fur &
stuff
fur &
stuff
fur &
stuff
fur &
stuff
air & hope
air & hope
fur &
stuff
fur &
stuff
fur &
stuff
fur &
stuff
fur &
stuff
air & hope
34.
35. cat ear cat ear
air &
plotting
air &
plotting
air &
plotting
cat ear cat ear
cat ear cat ear
top of cat
head
top of cat
head
top of cat
head
cat ear cat ear
air &
plotting
cat brain cat brain cat brain cat brain cat brain
air &
plotting
air &
plotting
brow ridge cat eye cat middle cat eye brow ridge
air &
plotting
air &
plotting
cheek cat eye cat nose cat eye cheek
air &
plotting
whiskers whiskers whiskers
little cat
nose
whiskers whiskers whiskers
whiskers whiskers whiskers chin chin whiskers whiskers
74. Make Autonomous Weapons?
Autonomous weapons select and engage targets without
human intervention. They might include, for example, armed
quadcopters that can search for and eliminate people meeting
certain pre-defined criteria, but do not include cruise missiles
or remotely piloted drones for which humans make all
targeting decisions. Artificial Intelligence (AI) technology has
reached a point where the deployment of such systems is —
practically if not legally — feasible within years, not decades,
and the stakes are high: autonomous weapons have been
described as the third revolution in warfare, after gunpowder
and nuclear arms.
In summary, we believe that AI has great
potential to benefit humanity in many ways, and
that the goal of the field should be to do so.
Starting a military AI arms race is a bad idea,
and should be prevented by a ban on offensive
autonomous weapons beyond meaningful human
control.
Stephen Hawking Director of research at the
Department of Applied Mathematics and Theoretical
Physics at Cambridge, 2012 Fundamental Physics
Prize laureate for his work on quantum gravity
Elon Musk SpaceX, Tesla, Solar City
Steve Wozniak, Apple Inc., Co-founder, member of
IEEE CS
Jaan Tallinn co-founder of Skype, CSER and FLI
Frank Wilczek MIT, Professor of Physics, Nobel
Laureate for his work on the strong nuclear force
Max Tegmark MIT, Professor of Physics, co-founder
of FLI
Daniel C. Dennett, Tufts University, Professor, Co-
Director, Center for Cognitive Studies, member of
AAAI
Noam Chomsky MIT, Institute Professor emeritus,
inductee in IEEE Intelligent Systems Hall of Fame,
Franklin medalist in Computer and Cognitive Science
Barbara Simons IBM Research (retired), Past President
ACM, ACM Fellow, AAAS Fellow
Stephen Goose Director of Human Rights Watch's
Arms Division
Anthony Aguirre, UCSC, Professor of Physics, co-
founder of FLI
Lisa Randall, Harvard, Professor of Physics
Martin Rees Co-founder of CSER and Astrophysicist
Stuart Russell Berkeley, Professor of Computer
Science, director of the Center for Intelligent Systems,
and co-author of the standard textbook “Artificial
Intelligence: a Modern Approach"
Nils J. Nilsson, Department of Computer Science,
Stanford University, Kumagai Professor of
Engineering, Emeritus, past president of AAAI
Barbara J. Grosz Harvard University, Higgins
Professor of Natural Sciences, former president AAAI,
former chair of IJCAI Board of Trustees
Tom Mitchell CMU, past president of AAAI, Fredkin
University Professor and Head of the Machine
Learning Department
Eric Horvitz, Microsoft Research, Managing
director, Microsoft Research, past president of AAAI,
co-chair of AAAI Presidential Panel on Long-term AI
Futures, member of ACM, IEEE CIS
Martha E. Pollack University of Michigan, Provost,
Professor of Computer Science & Professor of
Information, past president of AAAI, Fellow of AAAS,
ACM & AAAI
Henry Kautz, University of Rochester, Professor of
Computer Science, past president of AAAI, member of
ACM
Demis Hassabis, Google DeepMind, CEO
Yann LeCun, New York University & Facebook AI
Research, Professor of Computer Science & Director of
AI Research
Oren Etzioni, Allen Institute for AI, CEO, member of
AAAI, ACM
Peter Norvig, Google, Research Director, member
of AAAI, ACM
Geoffrey Hinton University of Toronto and Google,
Emeritus Professor, AAAI Fellow
Yoshua Bengio, Université de Montréal, Professor
75. Engage with deep learning,
it’s powerful magic;
please don’t make evil
killer robots. #ekr
Chris Friel
@tbd
cfriel@gmail.com