Learning Revealed Pedro Domingos University of Washington Where Does Knowledge Come From Evolution Experience Culture Where Does Knowledge Come From Evolution Experience Culture Computers ID: 562595
Download Presentation The PPT/PDF document "The Secrets of Machine" is the property of its rightful owner. Permission is granted to download and print the materials on this web site for personal, non-commercial use only, and to display it on your personal computer provided you do not modify the materials and that you retain all copyright notices contained in the materials. By downloading content from our website, you accept the terms of this agreement.
Slide1
The Secrets of MachineLearning Revealed
Pedro Domingos
University of WashingtonSlide2
Where Does Knowledge Come From?Evolution
Experience
CultureSlide3
Where Does Knowledge Come From?Evolution
Experience
Culture
ComputersSlide4
Most of the knowledge in the world in thefuture is going to be extracted by machinesand will reside in machines.– Yann LeCun, Director of AI Research, FacebookSlide5
Traditional Programming
Machine Learning
Computer
Data
Algorithm
Output
Computer
Data
Output
AlgorithmSlide6
So How Do Machines Learn?
1.
Fill in gaps in existing knowledge
2.
Emulate the brain
3. Simulate evolution4. Systematically reduce uncertainty
5. Notice similarities between old and newSlide7
The Five Tribes of Machine Learning
Tribe
Origins
Master Algorithm
Symbolists
Logic, philosophy
Inverse
deductionConnectionistsNeuroscience
Backpropagation
Evolutionaries
Evolutionary
biology
Genetic programming
Bayesians
Statistics
Probabilistic
inference
Analogizers
Psychology
Kernel machinesSlide8
Symbolists
Tom Mitchell
Steve Muggleton
Ross QuinlanSlide9
Inverse DeductionAddition
Subtraction
2
+ 2
―――
=
?
――
2
+
?
―――
= 4
――Slide10
Inverse DeductionDeduction
Socrates is human
+ Humans are mortal
.
―――――――――――
= ?
Induction
Socrates is human + ?
―――――――――――
= Socrates is mortal
――――――――――
――――――――――Slide11
Spot the Biologist in this PictureSlide12
Connectionists
Yann LeCun
Geoff Hinton
Yoshua BengioSlide13
A NeuronSlide14
An Artificial NeuronSlide15
BackpropagationSlide16
The Google Cat NetworkSlide17
Evolutionaries
John Koza
John Holland
Hod LipsonSlide18
Genetic AlgorithmsSlide19
Genetic ProgrammingSlide20
Evolving RobotsSlide21
Bayesians
David Heckerman
Judea Pearl
Michael JordanSlide22
Probabilistic InferenceSlide23
Probabilistic InferenceSlide24
Spam FiltersSlide25
Analogizers
Peter Hart
Vladimir Vapnik
Douglas HofstadterSlide26
Nearest NeighborSlide27
Kernel MachinesSlide28
Recommender SystemsSlide29
The Big Picture
Tribe
Problem
Solution
Symbolists
Knowledge composition
Inverse deduction
ConnectionistsCredit assignment
Backpropagation
Evolutionaries
Structure discovery
Genetic programming
Bayesians
Uncertainty
Probabilistic inference
Analogizers
Similarity
Kernel machinesSlide30
The Big Picture
Tribe
Problem
Solution
Symbolists
Knowledge composition
Inverse deduction
ConnectionistsCredit assignment
Backpropagation
Evolutionaries
Structure discovery
Genetic programming
Bayesians
Uncertainty
Probabilistic inference
Analogizers
Similarity
Kernel machines
But what we really need is
a single algorithm that solves all five!Slide31
Putting the Pieces TogetherRepresentationProbabilistic logic (e.g., Markov logic networks)
Weighted formulas → Distribution over states
Evaluation
Posterior probability
User-defined objective function
OptimizationFormula discovery: Genetic programming Weight learning: BackpropagationSlide32
Toward a Universal LearnerMuch remains to be done . . .We need your ideasSlide33
What a Universal Learner Will EnableHome Robots
Cancer Cures
360
o
Recommenders
World Wide BrainsSlide34
If we used all our technology resources,we could actually give people personalizedrecommendations for every step of your life. – Aneesh Chopra, former CTO of the U.S.Slide35