How the Quest for the Ultimate Learning Machine Will Remake Our World Pedro Domingos University of Washington Machine Learning Traditional Programming Machine Learning Computer Data Algorithm ID: 418186
Download Presentation The PPT/PDF document "The Master Algorithm" 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 Master AlgorithmHow the Quest for the Ultimate Learning Machine Will Remake Our World
Pedro Domingos
University of WashingtonSlide2
Machine LearningSlide3
Traditional Programming
Machine Learning
Computer
Data
Algorithm
Output
Computer
Data
Output
AlgorithmSlide4
Traditional Programming
Machine Learning
Computer
Data
Algorithm
Output
Master
Algorithm
Data
Output
AlgorithmSlide5
The Five Tribes of Machine Learning
Tribe
Origins
Master Algorithm
SymbolistsLogic, philosophyInverse deductionConnectionistsNeuroscienceBackpropagationEvolutionariesEvolutionary biologyGenetic programmingBayesiansStatisticsProbabilistic inferenceAnalogizers
PsychologyKernel machinesSlide6
Symbolists
Tom Mitchell
Steve Muggleton
Ross QuinlanSlide7
Inverse Deduction
Addition
Subtraction
2
+ 2――― = ?―― 2 + ?――― = 4――Slide8
Inverse Deduction
Deduction
Socrates is human
+ Humans are mortal
.――――――――――― = ?Induction Socrates is human + ?――――――――――― = Socrates is mortal――――――――――――――――――――Slide9
Spot the Biologist in this PictureSlide10
Connectionists
Yann LeCun
Geoff Hinton
Yoshua BengioSlide11
A NeuronSlide12
An Artificial NeuronSlide13
BackpropagationSlide14
The Google Cat NetworkSlide15
Evolutionaries
John Koza
John Holland
Hod LipsonSlide16
Genetic AlgorithmsSlide17
Genetic ProgrammingSlide18
Evolving RobotsSlide19
Bayesians
David Heckerman
Judea Pearl
Michael JordanSlide20
Probabilistic InferenceSlide21
Probabilistic InferenceSlide22
Spam FiltersSlide23
Analogizers
Peter Hart
Vladimir Vapnik
Douglas HofstadterSlide24
Nearest NeighborSlide25
Kernel MachinesSlide26
Recommender SystemsSlide27
The Big Picture
Tribe
Problem
Solution
SymbolistsKnowledge compositionInverse deductionConnectionistsCredit assignmentBackpropagationEvolutionariesStructure discoveryGenetic programmingBayesiansUncertaintyProbabilistic inferenceAnalogizersSimilarityKernel machinesSlide28
The Big Picture
Tribe
Problem
Solution
SymbolistsKnowledge compositionInverse deductionConnectionistsCredit assignmentBackpropagationEvolutionariesStructure discoveryGenetic programmingBayesiansUncertaintyProbabilistic inferenceAnalogizersSimilarityKernel machines
But what we really need is
a single algorithm that solves all five!Slide29
Putting the Pieces TogetherRepresentation
Probabilistic logic (e.g., Markov logic networks)
Weighted formulas → Distribution over states
Evaluation
Posterior probabilityUser-defined objective functionOptimizationFormula discovery: Genetic programming Weight learning: BackpropagationSlide30
Toward a Universal LearnerMuch remains to be done . . .
We need your ideasSlide31
What a Universal Learner Will Enable
Home Robots
Cancer Cures
360
o RecommendersWorld Wide BrainsSlide32