Stochastic Optimization Learning at Universität Graz | Flashcards & Summaries

Select your language

Suggested languages for you:
Log In Start studying!

It looks like you are in the US?
We have a website for your region.

Take me there

Lernmaterialien für Stochastic Optimization Learning an der Universität Graz

Greife auf kostenlose Karteikarten, Zusammenfassungen, Übungsaufgaben und Altklausuren für deinen Stochastic Optimization Learning Kurs an der Universität Graz zu.

TESTE DEIN WISSEN
Characteristics APD
Lösung anzeigen
TESTE DEIN WISSEN
Replace true function with statistical approximation
Move foreward in time
Lösung ausblenden
TESTE DEIN WISSEN
Supervised Learninh, Unsupervised Learning, Reinforcement Learning
Lösung anzeigen
TESTE DEIN WISSEN
SL: Task driven (regression, ranking,...)
Training by correctly labeled data
UL: data driven (clustering)
no solution provided, algorithm finds pattern
RL: decision process, algorithm learns to take actions to macimize reward
Lösung ausblenden
TESTE DEIN WISSEN
Problems involving many states and actions
Lösung anzeigen
TESTE DEIN WISSEN
For small number of states and actions: lookup table
But not realistic - use functions
Lösung ausblenden
TESTE DEIN WISSEN
Learning dimensions
Lösung anzeigen
TESTE DEIN WISSEN
- Model free or model based (model of rewards or transition properties)
- real world or simulator
- active or passive learning (policy given?
- on policy or off policy

Lösung ausblenden
TESTE DEIN WISSEN
Optimal Learning
Lösung anzeigen
TESTE DEIN WISSEN
Exploration vs Exploitation
Best long term strategy may involve sacrifice
Optimal: policy with least number of measurements or lowest sacrifice

Lösung ausblenden
TESTE DEIN WISSEN
Elements of a learning problem
Lösung anzeigen
TESTE DEIN WISSEN
1. how to make measurement?
2. effect of measurement?
3. evaluate result of measurement?
4. offline or online learning?
Lösung ausblenden
TESTE DEIN WISSEN
Nature of measurement decision 
Lösung anzeigen
TESTE DEIN WISSEN
- 0/1: stoppung problems
- Z: discrete set if alternatives (ranking and selection)
- R: continuous set (temperature, speed)
- 0/1/0/0/1/0/1: subset selection 
Lösung ausblenden
TESTE DEIN WISSEN
Effect of measurement 
Lösung anzeigen
TESTE DEIN WISSEN
- Frequentist point of view
- Baysian point of view:
-> start with distribution of belief about true mean -> after observation update distribution
Lösung ausblenden
TESTE DEIN WISSEN
Policies
Lösung anzeigen
TESTE DEIN WISSEN
-Deterministic 
- Sequential optimal: Dynamic programming
-Sequential: next measurement depends on knowledge state
— exploration
— exploitation
— epsilon greedy
— interval estimation
— boltzman exploration
— knowledge gradient policy
Lösung ausblenden
TESTE DEIN WISSEN
Knowledge gradient policy
Lösung anzeigen
TESTE DEIN WISSEN
Choose measurement that would improve best mean the most
Lösung ausblenden
TESTE DEIN WISSEN
Properties of KG
Lösung anzeigen
TESTE DEIN WISSEN
- One step look ahead
- optimal decision with one measurement remaining

Lösung ausblenden
TESTE DEIN WISSEN
Challenges of ADP
Lösung anzeigen
TESTE DEIN WISSEN
Exploration vs Exploitation
Value function approximation
Updating Vtn
Lösung ausblenden
  • 118625 Karteikarten
  • 2140 Studierende
  • 78 Lernmaterialien

Beispielhafte Karteikarten für deinen Stochastic Optimization Learning Kurs an der Universität Graz - von Kommilitonen auf StudySmarter erstellt!

Q:
Characteristics APD
A:
Replace true function with statistical approximation
Move foreward in time
Q:
Supervised Learninh, Unsupervised Learning, Reinforcement Learning
A:
SL: Task driven (regression, ranking,...)
Training by correctly labeled data
UL: data driven (clustering)
no solution provided, algorithm finds pattern
RL: decision process, algorithm learns to take actions to macimize reward
Q:
Problems involving many states and actions
A:
For small number of states and actions: lookup table
But not realistic - use functions
Q:
Learning dimensions
A:
- Model free or model based (model of rewards or transition properties)
- real world or simulator
- active or passive learning (policy given?
- on policy or off policy

Q:
Optimal Learning
A:
Exploration vs Exploitation
Best long term strategy may involve sacrifice
Optimal: policy with least number of measurements or lowest sacrifice

Mehr Karteikarten anzeigen
Q:
Elements of a learning problem
A:
1. how to make measurement?
2. effect of measurement?
3. evaluate result of measurement?
4. offline or online learning?
Q:
Nature of measurement decision 
A:
- 0/1: stoppung problems
- Z: discrete set if alternatives (ranking and selection)
- R: continuous set (temperature, speed)
- 0/1/0/0/1/0/1: subset selection 
Q:
Effect of measurement 
A:
- Frequentist point of view
- Baysian point of view:
-> start with distribution of belief about true mean -> after observation update distribution
Q:
Policies
A:
-Deterministic 
- Sequential optimal: Dynamic programming
-Sequential: next measurement depends on knowledge state
— exploration
— exploitation
— epsilon greedy
— interval estimation
— boltzman exploration
— knowledge gradient policy
Q:
Knowledge gradient policy
A:
Choose measurement that would improve best mean the most
Q:
Properties of KG
A:
- One step look ahead
- optimal decision with one measurement remaining

Q:
Challenges of ADP
A:
Exploration vs Exploitation
Value function approximation
Updating Vtn
Stochastic Optimization Learning

Erstelle und finde Lernmaterialien auf StudySmarter.

Greife kostenlos auf tausende geteilte Karteikarten, Zusammenfassungen, Altklausuren und mehr zu.

Jetzt loslegen

Das sind die beliebtesten Stochastic Optimization Learning Kurse im gesamten StudySmarter Universum

Einführung in die Stochastische Optimierung

Karlsruher Institut für Technologie

Zum Kurs
Modeling and Optimization

TU München

Zum Kurs
Planning & Optimization (HS21)

University of Basel

Zum Kurs
Conditioning and Learning Styles

John Cabot University

Zum Kurs
Query Optimization

TU München

Zum Kurs

Die all-in-one Lernapp für Studierende

Greife auf Millionen geteilter Lernmaterialien der StudySmarter Community zu
Kostenlos anmelden Stochastic Optimization Learning
Erstelle Karteikarten und Zusammenfassungen mit den StudySmarter Tools
Kostenlos loslegen Stochastic Optimization Learning