Semi-Supervise and Active Learning

jabgoldfishAI and Robotics

Oct 19, 2013 (3 years and 11 months ago)

67 views

Semi
-
Supervise and Active Learning

Meeting 1


1/15/2013

CSCE 6933

Rodney Nielsen


10/19/2013

Rodney Nielsen

2

Machine Learning


What are we going to study?

10/19/2013

Rodney Nielsen

3

Some Main Types of Learning


Supervised Learning



Unsupervised Learning



Semisupervised Learning



Active Learning


10/19/2013

Rodney Nielsen

4

Goal of Machine Learning


Develop algorithms that:


Discover patterns in data (learn from experience)


Use these discoveries to make useful observations
or predictions about that data or unseen data


Allow user (human or system) to improve
performance on associated tasks


Well defined learning tasks make all three
components clear


What is the task


What is the performance metric to be improved


What data is available as input for the learner

10/19/2013

Rodney Nielsen

5

Supervised Learning


Given

a set of training exs
X

and corresponding
outputs
Y


Want

the function
f
(
x
) =
y


Select

a hypothesis space
H
and learning
algorithm
A


Search

H

for a hypothesis
h

that approximates
f

10/19/2013

Rodney Nielsen

6

Some Main Types of Learning


Supervised Learning



Unsupervised Learning



Semisupervised Learning



Active Learning


10/19/2013

Rodney Nielsen

7

Machine Learning Tasks


Human Language Technologies


Optical character recognition


Speech recognition


Natural language processing


Machine translation


Web search engines


Text mining


Information extraction


Document classification


Recommender systems


Spam filtering


10/19/2013

Rodney Nielsen

8

Machine Learning Tasks


Systems and Software


OS optimization / tuning


DB performance tuning


Network intrusion detection


Programming by example


Software engineering




10/19/2013

Rodney Nielsen

9

Machine Learning Tasks


Vision and Image Processing


Optical character recognition


Object recognition


Activity recognition


Computer vision


Image processing


10/19/2013

Rodney Nielsen

10

Machine Learning Tasks


Business


Stock market analysis


Insurance fraud


Credit assessment


Credit card fraud detection


Advertisement placement


Product management



10/19/2013

Rodney Nielsen

11

Machine Learning Tasks


Biomedical


Medical diagnosis


DNA sequencing


Bioinformatics


Clinical informatics


10/19/2013

Rodney Nielsen

12

Machine Learning Tasks


Robotics


Locomotion


Manipulation


Machine vision


10/19/2013

Rodney Nielsen

13

Machine Learning Tasks


Game Playing


AIs





10/19/2013

Rodney Nielsen

14

Semi
-
Supervised and Active Learning


What are we going to study?

10/19/2013

Rodney Nielsen

15

Learning from Unlabeled Data


Is it Possible to Learn from Unlabeled Data

10/19/2013

Rodney Nielsen

16

All about You


Who (Name)


What (M.S., Ph.D., …)


Where (Area of Study within C.S. & Advisor)


When (Do you graduate)


Why (are you taking the course / what do you
hope to get out of it)

10/19/2013

Rodney Nielsen

17

Survey Experience


Overview of common Machine Learning (ML)
algorithms

Research Areas

Natural Language Processing,

Machine Learning, and

Cognitive Science

Intelligent Tutoring Systems

Classroom Technology

Educational Data Mining

Spoken
Dialogue

Companionbots

Clinical

Text Processing

Clinical QA &

Data Mining

10/19/2013

Rodney Nielsen

19

Companionbots

Perceptive, emotive, conversational, healthcare, companion robots

NSF CISE Smart Health & Wellbeing, PI

Collaborators: UC Denver, CU Boulder, BLT, U Denver

Consultants: Columbia, Worcester Polytechnic Institute, UCD Depression Center

10/19/2013

Rodney Nielsen

20

Elderly and Depression

Depression

Leading cause
of disability

M/F

All ages

Worldwide
(WHO)

Doubles cost of
care for chronic
diseases

Stats for
65
+

Double in

number by 2030

12


20%

50
-
58% of
hospital patients

36
-
50% of
healthcare
expenditures

10/19/2013

Rodney Nielsen

21

Goals


Research platform for


Monitoring
physical and
emotional conditions


Education
(
depression
, diet…)


Training/Coaching
(
explanatory style
, cognitive…)


Motivation (exercise, drug abstinence…)


Therapy (cognitive behavioral therapy…)


Clinical trials monitoring and adherence

10/19/2013

Rodney Nielsen

22

Goals


Participant Benefits


Maintain independence / Age in place


Improve quality of life


Reduce cost of care

10/19/2013

Rodney Nielsen

23

Scenario


Ruth: Oh for crying out loud! Why do I always
have to drop everything all the time?


Daisy: Oh that’s a shame, but Ruth, everybody
drops something from time to time. Hmm, when
was the last time you dropped a glass?

10/19/2013

Rodney Nielsen

24

Requirements


Daisy: … everybody drops something from time to time.
Hmm, when was the last time you dropped a glass?


Detect
Pessimistic Explanatory Style


Recognize the glass and the dropping event


Decide what if anything to reframe and what to challenge?


Generate natural language response

10/19/2013

Rodney Nielsen

25

Companionbot Strengths


Proactive real
-
time engagement / intervention


More complete exploration of the patient’s overall state


Ask questions and dialogue at opportune times


Real
-
time feedback and therapy


Vs. passively waiting for participant initiated interaction


Data mining


Adherence to doctor orders


Caregiver notification


Send report to doctor

10/19/2013

Rodney Nielsen

26

Companionbots Architecture


Goal


Manager

Speech
Recognition

Dialogue
Manager

Natural
Language
Generation

Natural
Language

U
nderstanding

Entity
Tracking

Audio

Vision

Location

Force /
Touch

Sonar

Scenario

U
nderstanding

Language

Beliefs

Body &
Motion

Habits,
Hobbies &
Routines

Emotions

Health

Scenario
Prediction

Emotion
Prediction

Dialogue
Prediction

M
anipulation

Manager

Posture
Manager

Expression
Manager

Gesture
Manager

Locomotion
Manager

Sensor 1
Manager

Entity
Recognition

Emotion
Recognition




Sensory


Input




F
undamental


R
ecognition




Situation


U
nderstanding

Emotion

U
nderstanding

E
nvironment

U
nderstanding




Situation


Prediction

E
nvironment

Prediction




User


Modeling


& History


Tracking




Behavior


Manager

Scenario
Goal
Manager

Emotion Goal
Manager

Dialogue
Goal
Manager

E
nvironment

Goal
Manager

Health Goal
Manager

Text to
Speech

Motor
Controls




Mechatronic


Control



Other
M
echatronic

Controls


Natural


Behavior


Generation

Natural
Movement
Generation

Natural
Expression
Generation




Tools

Question
Answering

Information
Retrieval /
Extraction

Document
S
ummarization

Audio

Movement


Mechatronic


Outputs

Visual
Displays



Time



Interpretation

Action

Recognize pessimistic
explanatory style and
suicidal ideation

10/19/2013

Rodney Nielsen

27

Recognizing Suicidal Ideation


Detection of mentions of suicidal ideation


It just isn’t worth continuing anymore.



Contact appropriate healthcare professional and
include in the live conversation


Meanwhile, invoke scripted dialogue to interact
with the participant

10/19/2013

Rodney Nielsen

28

Recognizing Suicidal Ideation


Detection of mentions of suicidal ideation in
clinical notes (AHRQ, sub, PI)


pt has thought about overdose but then she would
not be in heaven with her husband no longer
thinking about it oct 07



Found order of magnitude more incidents than
coded
(Anderson et al., 2011a & b)

10/19/2013

Rodney Nielsen

29

Companionbots Architecture


Goal


Manager

Speech
Recognition

Dialogue
Manager

Natural
Language
Generation

Natural
Language

U
nderstanding

Object
Tracking

Audio

Vision

Location

Force /
Touch

Distance
M
easurement

R
adar
,
IR


Scenario

U
nderstanding

Language

Beliefs

Body &
Motion

Habits,
Hobbies &
Routines

Emotions

Health

Scenario
Prediction

Emotion
Prediction

Dialogue
Prediction

M
anipulation

Manager

Posture
Manager

Expression
Manager

Gesture
Manager

Locomotion
Manager

Sensor 1
Manager

Object
Recognition

Emotion
Recognition




Sensory


Input




F
undamental


R
ecognition




Situation


U
nderstanding

Emotion

U
nderstanding

E
nvironment

U
nderstanding




Situation


Prediction

E
nvironment

Prediction




User


Modeling


& History


Tracking




Behavior


Manager

Scenario
Goal
Manager

Emotion Goal
Manager

Dialogue
Goal
Manager

E
nvironment

Goal
Manager

Health Goal
Manager

Text to
Speech

Motor
Controls




Mechatronic


Control



Other
M
echatronic

Controls


Natural


Behavior


Generation

Natural
Movement
Generation

Natural
Expression
Generation




Tools

Question
Answering

Information
Retrieval /
Extraction

Document
S
ummarization

Audio

Movement


Mechatronic


Outputs

Visual
Displays



Time



Interpretation

Action

Instance selection
for Co
-
training in
emotion recognition

Recognize pessimistic
explanatory style and
suicidal ideation

10/19/2013

Rodney Nielsen

30

Multimodal Emotion Recognition


Vision


Speech


Language

Why does this always have to happen to
me

10/19/2013

Rodney Nielsen

31

Co
-
Training Emotion Recognition

Given

a set
L

of labeled training examples




a set
U

of unlabeled training examples

Create a pool
U'

of examples by choosing
u

examples at random from
U

Loop for
k

iterations:


Use
L

to train a classifier
h
1

that considers only

Use
L

to train a classifier
h
2

that considers only

Use
L

to train a classifier
h
3

that considers only

Allow
h
1

to label
p
1

positive and
n
1

negative examples from
U’

Allow
h
2

to label
p
2

positive and
n
2

negative examples from
U'

Allow
h
3

to label
p
3

positive and
n
3

negative examples from
U'

Add these self
-
labeled examples to
L

Randomly choose examples from
U

to replenish
U’



(Blum & Mitchell, 1998)

Why does this always
have to happen to me

10/19/2013

Rodney Nielsen

32

Semisupervised & Active Learning


Most common strategy for instance selection


Based on class probability estimates


Semisupervised learning


Select
k

instances with highest class probabilities


Active learning


Select
k

instances with lowest class probabilities

10/19/2013

Rodney Nielsen

33

Probability Confidence / Variance?

+

10/19/2013

Rodney Nielsen

34

Probability Confidence / Variance?

+

+

+

+

+

+

+

+

+

+

+

+

+

+

+

+

+

+

+

+

+

+

+

+

+

+

+

+

+

+

+

+

+

+

+

+

+

+

+

+

+

+

-

-

-

-

-

-

-

-

-

-

-

-

-

-

-

-

-

-

-

-

-

-

-

-

-

-

-

-

-

-

-

-

-

-

-

+