Difference between revisions of "SC²S Colloquium - June 4, 2013"

From Sccswiki
Jump to navigation Jump to search
Line 13: Line 13:
  
 
This thesis investigates new loss functions in connection with  
 
This thesis investigates new loss functions in connection with  
classi�cation and regression with sparse grids.
+
classication and regression with sparse grids.
 
Thereby the main focus lies on loss functions better suited for  
 
Thereby the main focus lies on loss functions better suited for  
 
classification than the standard choice, the
 
classification than the standard choice, the
squared diff�erence. A further topic is a new density based approach  
+
squared difference. A further topic is a new density based approach  
 
to regression and gradient boosting,
 
to regression and gradient boosting,
 
a meta learning method that also makes use of loss functions, and  
 
a meta learning method that also makes use of loss functions, and  
 
their application to sparse grids.
 
their application to sparse grids.
 
 
  
 
== Torsten Koller: Employing Policy Search Techniques for Learning Throttle Valve Control ==
 
== Torsten Koller: Employing Policy Search Techniques for Learning Throttle Valve Control ==

Revision as of 17:31, 30 May 2013

Date: June 6, 2013
Room: 02.07.023
Time: 2:30 pm, s.t.


Florian Zacherl: New Loss Functions for Sparse Grid Classifiers

This thesis investigates new loss functions in connection with classication and regression with sparse grids. Thereby the main focus lies on loss functions better suited for classification than the standard choice, the squared difference. A further topic is a new density based approach to regression and gradient boosting, a meta learning method that also makes use of loss functions, and their application to sparse grids.

Torsten Koller: Employing Policy Search Techniques for Learning Throttle Valve Control

The aim of this thesis is to demonstrate the viability of reinforcement learning (RL) in challenging real-world problems such as throttle valve control. The throttle valve is a mechanical device which regulates the flow of a fluid or gas. Numerous applications of throttle valves can be found in industrial processes and systems such as refrigerators, semiconductor manufacturing and waterworks. One of the most important applications however is the gasoline throttle valve system regulating the amount of air entering the combustion engine. Due to high dynamics of the system including nonlinearities and discontinuances, throttle valve control is an especially challenging control task. In this thesis the Policy Search algorithm Pilco is modied to learn throttle valve control from data. In contrast to modelfree PID control or model-based controllers no sophisticated knowledge of the system nor tedious manual parameter tuning is required. To adress task-specic challenges and requirements such as high frequency control, no-overshoot and the control of arbitrary desired trajectories, several theoretical aspects are taken into consideration. To evaluate the theoretical results, the RL controller is applied to a throttle valve simulation aswell as a physical throttle valve system and tested on several control tasks.