machine teaching -...
TRANSCRIPT
![Page 1: Machine Teaching - pages.cs.wisc.edupages.cs.wisc.edu/~jerryzhu/career/pub/MachineTeachingAAAI.pdfMachine Teaching A(D) q* A (q )-1 * A A-1 D Q DD Tina wants Steve to learn a target](https://reader034.vdocument.in/reader034/viewer/2022042310/5ed8b2c46714ca7f476867d3/html5/thumbnails/1.jpg)
Machine Teaching
An Inverse Problem to Machine Learningand an Approach Toward Optimal Education
Jerry ZhuDepartment of Computer SciencesUniversity of Wisconsin-Madison
AAAI 2015
Zhu (Wisconsin) Machine Teaching AAAI 2015 1 / 12
![Page 2: Machine Teaching - pages.cs.wisc.edupages.cs.wisc.edu/~jerryzhu/career/pub/MachineTeachingAAAI.pdfMachine Teaching A(D) q* A (q )-1 * A A-1 D Q DD Tina wants Steve to learn a target](https://reader034.vdocument.in/reader034/viewer/2022042310/5ed8b2c46714ca7f476867d3/html5/thumbnails/2.jpg)
Example One
Steve the student runs a linear SVM:
Given a training set with n items xi ∈ Rd, yi ∈ −1, 1
Steve learns w ∈ Rd
Tina the teacher wants Steve to learn a target w∗
x>w∗ = 0
What is the smallest training set Tina can give Steve?
Zhu (Wisconsin) Machine Teaching AAAI 2015 2 / 12
![Page 3: Machine Teaching - pages.cs.wisc.edupages.cs.wisc.edu/~jerryzhu/career/pub/MachineTeachingAAAI.pdfMachine Teaching A(D) q* A (q )-1 * A A-1 D Q DD Tina wants Steve to learn a target](https://reader034.vdocument.in/reader034/viewer/2022042310/5ed8b2c46714ca7f476867d3/html5/thumbnails/3.jpg)
Example One
Steve the student runs a linear SVM:
Given a training set with n items xi ∈ Rd, yi ∈ −1, 1
Steve learns w ∈ Rd
Tina the teacher wants Steve to learn a target w∗
x>w∗ = 0
What is the smallest training set Tina can give Steve?
Zhu (Wisconsin) Machine Teaching AAAI 2015 2 / 12
![Page 4: Machine Teaching - pages.cs.wisc.edupages.cs.wisc.edu/~jerryzhu/career/pub/MachineTeachingAAAI.pdfMachine Teaching A(D) q* A (q )-1 * A A-1 D Q DD Tina wants Steve to learn a target](https://reader034.vdocument.in/reader034/viewer/2022042310/5ed8b2c46714ca7f476867d3/html5/thumbnails/4.jpg)
Example One
Steve the student runs a linear SVM:
Given a training set with n items xi ∈ Rd, yi ∈ −1, 1
Steve learns w ∈ Rd
Tina the teacher wants Steve to learn a target w∗
x>w∗ = 0
What is the smallest training set Tina can give Steve?
Zhu (Wisconsin) Machine Teaching AAAI 2015 2 / 12
![Page 5: Machine Teaching - pages.cs.wisc.edupages.cs.wisc.edu/~jerryzhu/career/pub/MachineTeachingAAAI.pdfMachine Teaching A(D) q* A (q )-1 * A A-1 D Q DD Tina wants Steve to learn a target](https://reader034.vdocument.in/reader034/viewer/2022042310/5ed8b2c46714ca7f476867d3/html5/thumbnails/5.jpg)
Example One
Tina’s non-iid training set with n = 2 items
Zhu (Wisconsin) Machine Teaching AAAI 2015 3 / 12
![Page 6: Machine Teaching - pages.cs.wisc.edupages.cs.wisc.edu/~jerryzhu/career/pub/MachineTeachingAAAI.pdfMachine Teaching A(D) q* A (q )-1 * A A-1 D Q DD Tina wants Steve to learn a target](https://reader034.vdocument.in/reader034/viewer/2022042310/5ed8b2c46714ca7f476867d3/html5/thumbnails/6.jpg)
Example Two
Steve estimates a Gaussian density:
Given x1 . . .xn ∈ Rd
Steve learns µ =1
n
∑xi, Σ =
1
n− 1
∑(xi − µ)(xi − µ)>
Tina wants Steve to learn a target Gaussian with (µ∗,Σ∗)
Zhu (Wisconsin) Machine Teaching AAAI 2015 4 / 12
![Page 7: Machine Teaching - pages.cs.wisc.edupages.cs.wisc.edu/~jerryzhu/career/pub/MachineTeachingAAAI.pdfMachine Teaching A(D) q* A (q )-1 * A A-1 D Q DD Tina wants Steve to learn a target](https://reader034.vdocument.in/reader034/viewer/2022042310/5ed8b2c46714ca7f476867d3/html5/thumbnails/7.jpg)
Example Two
Steve estimates a Gaussian density:
Given x1 . . .xn ∈ Rd
Steve learns µ =1
n
∑xi, Σ =
1
n− 1
∑(xi − µ)(xi − µ)>
Tina wants Steve to learn a target Gaussian with (µ∗,Σ∗)
Zhu (Wisconsin) Machine Teaching AAAI 2015 4 / 12
![Page 8: Machine Teaching - pages.cs.wisc.edupages.cs.wisc.edu/~jerryzhu/career/pub/MachineTeachingAAAI.pdfMachine Teaching A(D) q* A (q )-1 * A A-1 D Q DD Tina wants Steve to learn a target](https://reader034.vdocument.in/reader034/viewer/2022042310/5ed8b2c46714ca7f476867d3/html5/thumbnails/8.jpg)
Example Two
Tina’s minimal training set of n = d+ 1 tetrahedron vertices
Zhu (Wisconsin) Machine Teaching AAAI 2015 5 / 12
![Page 9: Machine Teaching - pages.cs.wisc.edupages.cs.wisc.edu/~jerryzhu/career/pub/MachineTeachingAAAI.pdfMachine Teaching A(D) q* A (q )-1 * A A-1 D Q DD Tina wants Steve to learn a target](https://reader034.vdocument.in/reader034/viewer/2022042310/5ed8b2c46714ca7f476867d3/html5/thumbnails/9.jpg)
Machine Teaching More Powerful Than Active Learning
θ θ
O(1/2 )n
θ
O(1/n)
passive learning "waits" active learning "explores" teaching "guides"
Sample complexity to achieve ε error
passive learning 1/ε
active learning log(1/ε)
machine teaching 2: Tina knows θ
Zhu (Wisconsin) Machine Teaching AAAI 2015 6 / 12
![Page 10: Machine Teaching - pages.cs.wisc.edupages.cs.wisc.edu/~jerryzhu/career/pub/MachineTeachingAAAI.pdfMachine Teaching A(D) q* A (q )-1 * A A-1 D Q DD Tina wants Steve to learn a target](https://reader034.vdocument.in/reader034/viewer/2022042310/5ed8b2c46714ca7f476867d3/html5/thumbnails/10.jpg)
Machine Teaching More Powerful Than Active Learning
θ θ
O(1/2 )n
θ
O(1/n)
passive learning "waits" active learning "explores" teaching "guides"
Sample complexity to achieve ε error
passive learning 1/ε
active learning log(1/ε)
machine teaching 2: Tina knows θ
Zhu (Wisconsin) Machine Teaching AAAI 2015 6 / 12
![Page 11: Machine Teaching - pages.cs.wisc.edupages.cs.wisc.edu/~jerryzhu/career/pub/MachineTeachingAAAI.pdfMachine Teaching A(D) q* A (q )-1 * A A-1 D Q DD Tina wants Steve to learn a target](https://reader034.vdocument.in/reader034/viewer/2022042310/5ed8b2c46714ca7f476867d3/html5/thumbnails/11.jpg)
Machine Teaching More Powerful Than Active Learning
θ θ
O(1/2 )n
θ
O(1/n)
passive learning "waits" active learning "explores" teaching "guides"
Sample complexity to achieve ε error
passive learning 1/ε
active learning log(1/ε)
machine teaching 2: Tina knows θ
Zhu (Wisconsin) Machine Teaching AAAI 2015 6 / 12
![Page 12: Machine Teaching - pages.cs.wisc.edupages.cs.wisc.edu/~jerryzhu/career/pub/MachineTeachingAAAI.pdfMachine Teaching A(D) q* A (q )-1 * A A-1 D Q DD Tina wants Steve to learn a target](https://reader034.vdocument.in/reader034/viewer/2022042310/5ed8b2c46714ca7f476867d3/html5/thumbnails/12.jpg)
Machine Teaching
A(D)
θ∗
Α (θ )−1 ∗
A
Α−1
D
Θ
DD
Tina wants Steve to learn a target model θ∗
I not machine learning: Tina already knows θ∗
Tina knows Steve’s learning algorithm A
Tina seeks the best training set within A−1(θ∗) for Steve
I best = the smallest (Teaching Dimension [Goldman Kearns 1995]), orother criteria
Zhu (Wisconsin) Machine Teaching AAAI 2015 7 / 12
![Page 13: Machine Teaching - pages.cs.wisc.edupages.cs.wisc.edu/~jerryzhu/career/pub/MachineTeachingAAAI.pdfMachine Teaching A(D) q* A (q )-1 * A A-1 D Q DD Tina wants Steve to learn a target](https://reader034.vdocument.in/reader034/viewer/2022042310/5ed8b2c46714ca7f476867d3/html5/thumbnails/13.jpg)
Machine Teaching
A(D)
θ∗
Α (θ )−1 ∗
A
Α−1
D
Θ
DD
Tina wants Steve to learn a target model θ∗
I not machine learning: Tina already knows θ∗
Tina knows Steve’s learning algorithm A
Tina seeks the best training set within A−1(θ∗) for Steve
I best = the smallest (Teaching Dimension [Goldman Kearns 1995]), orother criteria
Zhu (Wisconsin) Machine Teaching AAAI 2015 7 / 12
![Page 14: Machine Teaching - pages.cs.wisc.edupages.cs.wisc.edu/~jerryzhu/career/pub/MachineTeachingAAAI.pdfMachine Teaching A(D) q* A (q )-1 * A A-1 D Q DD Tina wants Steve to learn a target](https://reader034.vdocument.in/reader034/viewer/2022042310/5ed8b2c46714ca7f476867d3/html5/thumbnails/14.jpg)
Machine Teaching
A(D)
θ∗
Α (θ )−1 ∗
A
Α−1
D
Θ
DD
Tina wants Steve to learn a target model θ∗
I not machine learning: Tina already knows θ∗
Tina knows Steve’s learning algorithm A
Tina seeks the best training set within A−1(θ∗) for Steve
I best = the smallest (Teaching Dimension [Goldman Kearns 1995]), orother criteria
Zhu (Wisconsin) Machine Teaching AAAI 2015 7 / 12
![Page 15: Machine Teaching - pages.cs.wisc.edupages.cs.wisc.edu/~jerryzhu/career/pub/MachineTeachingAAAI.pdfMachine Teaching A(D) q* A (q )-1 * A A-1 D Q DD Tina wants Steve to learn a target](https://reader034.vdocument.in/reader034/viewer/2022042310/5ed8b2c46714ca7f476867d3/html5/thumbnails/15.jpg)
Machine Teaching
A(D)
θ∗
Α (θ )−1 ∗
A
Α−1
D
Θ
DD
Tina wants Steve to learn a target model θ∗
I not machine learning: Tina already knows θ∗
Tina knows Steve’s learning algorithm A
Tina seeks the best training set within A−1(θ∗) for Steve
I best = the smallest (Teaching Dimension [Goldman Kearns 1995]), orother criteria
Zhu (Wisconsin) Machine Teaching AAAI 2015 7 / 12
![Page 16: Machine Teaching - pages.cs.wisc.edupages.cs.wisc.edu/~jerryzhu/career/pub/MachineTeachingAAAI.pdfMachine Teaching A(D) q* A (q )-1 * A A-1 D Q DD Tina wants Steve to learn a target](https://reader034.vdocument.in/reader034/viewer/2022042310/5ed8b2c46714ca7f476867d3/html5/thumbnails/16.jpg)
Machine Teaching
A(D)
θ∗
Α (θ )−1 ∗
A
Α−1
D
Θ
DD
Tina wants Steve to learn a target model θ∗
I not machine learning: Tina already knows θ∗
Tina knows Steve’s learning algorithm A
Tina seeks the best training set within A−1(θ∗) for SteveI best = the smallest (Teaching Dimension [Goldman Kearns 1995]), or
other criteria
Zhu (Wisconsin) Machine Teaching AAAI 2015 7 / 12
![Page 17: Machine Teaching - pages.cs.wisc.edupages.cs.wisc.edu/~jerryzhu/career/pub/MachineTeachingAAAI.pdfMachine Teaching A(D) q* A (q )-1 * A A-1 D Q DD Tina wants Steve to learn a target](https://reader034.vdocument.in/reader034/viewer/2022042310/5ed8b2c46714ca7f476867d3/html5/thumbnails/17.jpg)
Machine Teaching Harder Than Machine Learning
Special case: teaching the exact parameters, minimizing training set size
minD∈D
|D| Tina’s problem
s.t. θ∗ = argminθ∈Θ
1
|D|∑zi∈D
`(zi, θ) + Ω(θ) Steve’s algorithm A
Bilevel optimization, NP-hard in general
Zhu (Wisconsin) Machine Teaching AAAI 2015 8 / 12
![Page 18: Machine Teaching - pages.cs.wisc.edupages.cs.wisc.edu/~jerryzhu/career/pub/MachineTeachingAAAI.pdfMachine Teaching A(D) q* A (q )-1 * A A-1 D Q DD Tina wants Steve to learn a target](https://reader034.vdocument.in/reader034/viewer/2022042310/5ed8b2c46714ca7f476867d3/html5/thumbnails/18.jpg)
General Machine Teaching Framework
[poster 819 Wednesday]
minD∈D,θ∈Θ
RT (θ) + λET (D)
s.t. θ = A(D)
RT (): teaching risk function e.g. ‖θ − θ∗‖22
ET (): teaching effort function e.g. different item costs
Tina’s search space D: constructive or pool-based, batch or sequential
Tractable solutions when Steve runs linear regression, logisticregression, SVM, LDA, etc. [Mei Z 2015a, Mei Z 2015b]
Zhu (Wisconsin) Machine Teaching AAAI 2015 9 / 12
![Page 19: Machine Teaching - pages.cs.wisc.edupages.cs.wisc.edu/~jerryzhu/career/pub/MachineTeachingAAAI.pdfMachine Teaching A(D) q* A (q )-1 * A A-1 D Q DD Tina wants Steve to learn a target](https://reader034.vdocument.in/reader034/viewer/2022042310/5ed8b2c46714ca7f476867d3/html5/thumbnails/19.jpg)
General Machine Teaching Framework
[poster 819 Wednesday]
minD∈D,θ∈Θ
RT (θ) + λET (D)
s.t. θ = A(D)
RT (): teaching risk function e.g. ‖θ − θ∗‖22ET (): teaching effort function e.g. different item costs
Tina’s search space D: constructive or pool-based, batch or sequential
Tractable solutions when Steve runs linear regression, logisticregression, SVM, LDA, etc. [Mei Z 2015a, Mei Z 2015b]
Zhu (Wisconsin) Machine Teaching AAAI 2015 9 / 12
![Page 20: Machine Teaching - pages.cs.wisc.edupages.cs.wisc.edu/~jerryzhu/career/pub/MachineTeachingAAAI.pdfMachine Teaching A(D) q* A (q )-1 * A A-1 D Q DD Tina wants Steve to learn a target](https://reader034.vdocument.in/reader034/viewer/2022042310/5ed8b2c46714ca7f476867d3/html5/thumbnails/20.jpg)
General Machine Teaching Framework
[poster 819 Wednesday]
minD∈D,θ∈Θ
RT (θ) + λET (D)
s.t. θ = A(D)
RT (): teaching risk function e.g. ‖θ − θ∗‖22ET (): teaching effort function e.g. different item costs
Tina’s search space D: constructive or pool-based, batch or sequential
Tractable solutions when Steve runs linear regression, logisticregression, SVM, LDA, etc. [Mei Z 2015a, Mei Z 2015b]
Zhu (Wisconsin) Machine Teaching AAAI 2015 9 / 12
![Page 21: Machine Teaching - pages.cs.wisc.edupages.cs.wisc.edu/~jerryzhu/career/pub/MachineTeachingAAAI.pdfMachine Teaching A(D) q* A (q )-1 * A A-1 D Q DD Tina wants Steve to learn a target](https://reader034.vdocument.in/reader034/viewer/2022042310/5ed8b2c46714ca7f476867d3/html5/thumbnails/21.jpg)
General Machine Teaching Framework
[poster 819 Wednesday]
minD∈D,θ∈Θ
RT (θ) + λET (D)
s.t. θ = A(D)
RT (): teaching risk function e.g. ‖θ − θ∗‖22ET (): teaching effort function e.g. different item costs
Tina’s search space D: constructive or pool-based, batch or sequential
Tractable solutions when Steve runs linear regression, logisticregression, SVM, LDA, etc. [Mei Z 2015a, Mei Z 2015b]
Zhu (Wisconsin) Machine Teaching AAAI 2015 9 / 12
![Page 22: Machine Teaching - pages.cs.wisc.edupages.cs.wisc.edu/~jerryzhu/career/pub/MachineTeachingAAAI.pdfMachine Teaching A(D) q* A (q )-1 * A A-1 D Q DD Tina wants Steve to learn a target](https://reader034.vdocument.in/reader034/viewer/2022042310/5ed8b2c46714ca7f476867d3/html5/thumbnails/22.jpg)
Education
Steve may actually be a human!
Tina computes the best lesson D for Steve
Needs Steve’s cognitive model A
I a “good enough” A is fine
Zhu (Wisconsin) Machine Teaching AAAI 2015 10 / 12
![Page 23: Machine Teaching - pages.cs.wisc.edupages.cs.wisc.edu/~jerryzhu/career/pub/MachineTeachingAAAI.pdfMachine Teaching A(D) q* A (q )-1 * A A-1 D Q DD Tina wants Steve to learn a target](https://reader034.vdocument.in/reader034/viewer/2022042310/5ed8b2c46714ca7f476867d3/html5/thumbnails/23.jpg)
Education
Steve may actually be a human!
Tina computes the best lesson D for Steve
Needs Steve’s cognitive model A
I a “good enough” A is fine
Zhu (Wisconsin) Machine Teaching AAAI 2015 10 / 12
![Page 24: Machine Teaching - pages.cs.wisc.edupages.cs.wisc.edu/~jerryzhu/career/pub/MachineTeachingAAAI.pdfMachine Teaching A(D) q* A (q )-1 * A A-1 D Q DD Tina wants Steve to learn a target](https://reader034.vdocument.in/reader034/viewer/2022042310/5ed8b2c46714ca7f476867d3/html5/thumbnails/24.jpg)
Education
Steve may actually be a human!
Tina computes the best lesson D for Steve
Needs Steve’s cognitive model A
I a “good enough” A is fine
Zhu (Wisconsin) Machine Teaching AAAI 2015 10 / 12
![Page 25: Machine Teaching - pages.cs.wisc.edupages.cs.wisc.edu/~jerryzhu/career/pub/MachineTeachingAAAI.pdfMachine Teaching A(D) q* A (q )-1 * A A-1 D Q DD Tina wants Steve to learn a target](https://reader034.vdocument.in/reader034/viewer/2022042310/5ed8b2c46714ca7f476867d3/html5/thumbnails/25.jpg)
Education
Steve may actually be a human!
Tina computes the best lesson D for Steve
Needs Steve’s cognitive model AI a “good enough” A is fine
Zhu (Wisconsin) Machine Teaching AAAI 2015 10 / 12
![Page 26: Machine Teaching - pages.cs.wisc.edupages.cs.wisc.edu/~jerryzhu/career/pub/MachineTeachingAAAI.pdfMachine Teaching A(D) q* A (q )-1 * A A-1 D Q DD Tina wants Steve to learn a target](https://reader034.vdocument.in/reader034/viewer/2022042310/5ed8b2c46714ca7f476867d3/html5/thumbnails/26.jpg)
A Case Study
θ =0.5∗0 1
y=−1 y=1
Human categorization [Patil Z Kopec Love 2014]
A: a limited capacity retrieval cognitive model
human training set human test accuracy
D 72.5%iid 69.8%
(statistically significant)
Zhu (Wisconsin) Machine Teaching AAAI 2015 11 / 12
![Page 27: Machine Teaching - pages.cs.wisc.edupages.cs.wisc.edu/~jerryzhu/career/pub/MachineTeachingAAAI.pdfMachine Teaching A(D) q* A (q )-1 * A A-1 D Q DD Tina wants Steve to learn a target](https://reader034.vdocument.in/reader034/viewer/2022042310/5ed8b2c46714ca7f476867d3/html5/thumbnails/27.jpg)
A Case Study
θ =0.5∗0 1
y=−1 y=1
Human categorization [Patil Z Kopec Love 2014]
A: a limited capacity retrieval cognitive model
human training set human test accuracy
D 72.5%iid 69.8%
(statistically significant)
Zhu (Wisconsin) Machine Teaching AAAI 2015 11 / 12
![Page 28: Machine Teaching - pages.cs.wisc.edupages.cs.wisc.edu/~jerryzhu/career/pub/MachineTeachingAAAI.pdfMachine Teaching A(D) q* A (q )-1 * A A-1 D Q DD Tina wants Steve to learn a target](https://reader034.vdocument.in/reader034/viewer/2022042310/5ed8b2c46714ca7f476867d3/html5/thumbnails/28.jpg)
A Case Study
θ =0.5∗0 1
y=−1 y=1
Human categorization [Patil Z Kopec Love 2014]
A: a limited capacity retrieval cognitive model
human training set human test accuracy
D 72.5%iid 69.8%
(statistically significant)
Zhu (Wisconsin) Machine Teaching AAAI 2015 11 / 12
![Page 29: Machine Teaching - pages.cs.wisc.edupages.cs.wisc.edu/~jerryzhu/career/pub/MachineTeachingAAAI.pdfMachine Teaching A(D) q* A (q )-1 * A A-1 D Q DD Tina wants Steve to learn a target](https://reader034.vdocument.in/reader034/viewer/2022042310/5ed8b2c46714ca7f476867d3/html5/thumbnails/29.jpg)
A Case Study
θ =0.5∗0 1
y=−1 y=1
Human categorization [Patil Z Kopec Love 2014]
A: a limited capacity retrieval cognitive model
human training set human test accuracy
D 72.5%iid 69.8%
(statistically significant)
Zhu (Wisconsin) Machine Teaching AAAI 2015 11 / 12
![Page 30: Machine Teaching - pages.cs.wisc.edupages.cs.wisc.edu/~jerryzhu/career/pub/MachineTeachingAAAI.pdfMachine Teaching A(D) q* A (q )-1 * A A-1 D Q DD Tina wants Steve to learn a target](https://reader034.vdocument.in/reader034/viewer/2022042310/5ed8b2c46714ca7f476867d3/html5/thumbnails/30.jpg)
Open Problems for the AI Community
Optimization: solve Tina’s problem for any Steve
Theory: generalize Teaching Dimension [Goldman Kearns 1995]
Cognitive Science and Education: Steve’s A, real education tasks
Computer Security: data poisoning attacks
References:http://pages.cs.wisc.edu/~jerryzhu/machineteaching/
Zhu (Wisconsin) Machine Teaching AAAI 2015 12 / 12
![Page 31: Machine Teaching - pages.cs.wisc.edupages.cs.wisc.edu/~jerryzhu/career/pub/MachineTeachingAAAI.pdfMachine Teaching A(D) q* A (q )-1 * A A-1 D Q DD Tina wants Steve to learn a target](https://reader034.vdocument.in/reader034/viewer/2022042310/5ed8b2c46714ca7f476867d3/html5/thumbnails/31.jpg)
Open Problems for the AI Community
Optimization: solve Tina’s problem for any Steve
Theory: generalize Teaching Dimension [Goldman Kearns 1995]
Cognitive Science and Education: Steve’s A, real education tasks
Computer Security: data poisoning attacks
References:http://pages.cs.wisc.edu/~jerryzhu/machineteaching/
Zhu (Wisconsin) Machine Teaching AAAI 2015 12 / 12
![Page 32: Machine Teaching - pages.cs.wisc.edupages.cs.wisc.edu/~jerryzhu/career/pub/MachineTeachingAAAI.pdfMachine Teaching A(D) q* A (q )-1 * A A-1 D Q DD Tina wants Steve to learn a target](https://reader034.vdocument.in/reader034/viewer/2022042310/5ed8b2c46714ca7f476867d3/html5/thumbnails/32.jpg)
Open Problems for the AI Community
Optimization: solve Tina’s problem for any Steve
Theory: generalize Teaching Dimension [Goldman Kearns 1995]
Cognitive Science and Education: Steve’s A, real education tasks
Computer Security: data poisoning attacks
References:http://pages.cs.wisc.edu/~jerryzhu/machineteaching/
Zhu (Wisconsin) Machine Teaching AAAI 2015 12 / 12
![Page 33: Machine Teaching - pages.cs.wisc.edupages.cs.wisc.edu/~jerryzhu/career/pub/MachineTeachingAAAI.pdfMachine Teaching A(D) q* A (q )-1 * A A-1 D Q DD Tina wants Steve to learn a target](https://reader034.vdocument.in/reader034/viewer/2022042310/5ed8b2c46714ca7f476867d3/html5/thumbnails/33.jpg)
Open Problems for the AI Community
Optimization: solve Tina’s problem for any Steve
Theory: generalize Teaching Dimension [Goldman Kearns 1995]
Cognitive Science and Education: Steve’s A, real education tasks
Computer Security: data poisoning attacks
References:http://pages.cs.wisc.edu/~jerryzhu/machineteaching/
Zhu (Wisconsin) Machine Teaching AAAI 2015 12 / 12
![Page 34: Machine Teaching - pages.cs.wisc.edupages.cs.wisc.edu/~jerryzhu/career/pub/MachineTeachingAAAI.pdfMachine Teaching A(D) q* A (q )-1 * A A-1 D Q DD Tina wants Steve to learn a target](https://reader034.vdocument.in/reader034/viewer/2022042310/5ed8b2c46714ca7f476867d3/html5/thumbnails/34.jpg)
Open Problems for the AI Community
Optimization: solve Tina’s problem for any Steve
Theory: generalize Teaching Dimension [Goldman Kearns 1995]
Cognitive Science and Education: Steve’s A, real education tasks
Computer Security: data poisoning attacks
References:http://pages.cs.wisc.edu/~jerryzhu/machineteaching/
Zhu (Wisconsin) Machine Teaching AAAI 2015 12 / 12
![Page 35: Machine Teaching - pages.cs.wisc.edupages.cs.wisc.edu/~jerryzhu/career/pub/MachineTeachingAAAI.pdfMachine Teaching A(D) q* A (q )-1 * A A-1 D Q DD Tina wants Steve to learn a target](https://reader034.vdocument.in/reader034/viewer/2022042310/5ed8b2c46714ca7f476867d3/html5/thumbnails/35.jpg)
Open Problems for the AI Community
Optimization: solve Tina’s problem for any Steve
Theory: generalize Teaching Dimension [Goldman Kearns 1995]
Cognitive Science and Education: Steve’s A, real education tasks
Computer Security: data poisoning attacks
References:http://pages.cs.wisc.edu/~jerryzhu/machineteaching/
Zhu (Wisconsin) Machine Teaching AAAI 2015 12 / 12