PROGRAM SCHEDULE (Download PDF schedule)

This all-day short course is organized along two major axes -- namely, theoretical foundations, and vision problems. The theoretical foundations of SIGs will be discussed in terms of a unified representation of spatial, temporal and causal AND-OR graphs, and their inference and learning. Vision problems will also be discussed in terms of applications of AND-OR graphs to answering What, Where, When, and Why queries about objects, scenes, and events.

Vision Problems
Objects Scenes Events



Theoretical
foundations

Spatial

Temporal

Causal

Inference

Learning

Time Speaker

Topic

 Download
08:30 -- 08:45 Song-Chun Zhu
  • Historic remarks on Stochastic Image Grammars
  • Demo: Joint spatial, temporal, and causal inference by SIG

               Answering queries about what, who, when, Where and why

 Lecture 1
08:35 -- 09:20 Song-Chun Zhu
  • Representation: Spatial AND-OR graph for representing the scene-object-part-primitive hierarchy
  • Demo: answering queries What and Where about objects and scenes in images
 Lecture 2
09:20 -- 10:00 Sinisa Todorovic
  • Inference of SIGs: traditional algorithms
    • Dynamic programming, CYK and Earley parsers
    • Inside-outside algorithm
 Lecture 3

Lecture 4
  • Representation: Spatial AND-OR graph for representing the stationary event-activities-primitive actions hierarchy
10:00 -- 10:30 --

Coffee break
10:30 -- 11:50 Aleš Leonardis
  • Representation: compositional representations of object structure
  • Inference
  • Learning
  • Demo
 Lecture 5
11:50 -- 12:30 Sinisa Todorovic
  • Representation: Temporal AND-OR graph (T-AOG) for representing the event-action-motion primitive hierarchy
  • Demo
 Lecture 6
12:30 -- 01:30 --

Lunch
01:30 -- 02:15 Sinisa Todorovic
  • Inference of S-AOG and T-AOG: Cost-sensitive and goal guided
  • Scheduling top-down/bottom-up processes (alpha, beta, gamma)
  • Demo
 Lecture 7
02:15 -- 03:00 Song-Chun Zhu
  • Representation: Causal AND-OR graphs (C-AOG) for representing the causal-action recursion for reasoning
  • Demo
 Lecture 8
03:00 -- 03:30 --

Coffee break
03:30 -- 04:00 Sinisa Todorovic
  • Representation: Probabilistic logic CNF for reasoning
  • Demo
 Lecture 9
04:00 -- 04:50 Song-Chun Zhu
  • Learning of SIGs
  • Information projection for learning S-AOG, T-AOG, C-AOG
  • Bounds of inductive learning rates
Lecture 10
04:50 -- 05:30 all speakers
  • Evaluation by querying about WHAT, WHERE, WHEN, WHY
  • Open problems
  • Q&A session