Open Problems Theoretical understanding and Algorithms

Download Report

Transcript Open Problems Theoretical understanding and Algorithms

Open Problems
Theoretical understanding; Algorithms; Programming
I Understanding:
Interaction:
• If there is a modeling choice – model interaction at feature level or output
level?
• When is there a need to train with the inference, when not?
Interacting classifiers vs. non-interacting classifiers. (Vertical vs. Horizontal)
Need to fully train; ability to fully train
II Learning:
Training with an Inference Mechanism
• How to train classifiers (given that feedback is not direct)?
(Propagation via Constraints; implications to hierarchies)
• Relations to Constraint Classification using a general cost function?
Open Problems
Theoretical understanding; Algorithms; Programming
III Bootstrapping:
Labeling using Inference
IV Programming:
How to program a system of that sort?
•
•
•
•
•
Information sources  Features;
Feature and previously learning classifiers
Hierarchies;
Dependencies
External Resources