RepLearn 2013
Workshop on Learning Rich Representations from Low-Level Sensors

Motivation and Relevance

A human-level artificially intelligent agent must be able to represent and reason about the world, at some level, in terms of high-level concepts such as entities and relations. The problem of acquiring these rich high-level representations, known as the knowledge acquisition bottleneck, has long been an obstacle for achieving human-level AI. A popular approach to this problem is to handcraft these high-level representations, but this has had limited success. An alternate approach is for rich representations to be learned autonomously from low-level sensor data. Potentially, the latter approach may yield more robust representations, and should rely less on human knowledge-engineering.

Schedule

(PDFs of all accepted papers are available on AAAI's library website.)

(08:00-08:20) Sign-in. Poster setup.

08:30-08:45 Workshop introduction and focus
08:45-09:30 Invited talk: Jeff Hawkins
09:35-10:20 Invited talk: Juergen Schmidhuber [talk overview]

10:30-11:00 Coffee Break

11:00-11:45 Invited talk: Benjamin Kuipers
11:45-12:05 Top-Down Abstraction Learning Using Prediction as a Supervisory Signal, Jonathan Mugan [slides]
12:10-12:30 Two Perspectives on Learning Rich Representations from Robot Experience, Joseph Modayil [slides]

12:30-13:30 Lunch

13:30-13:50 Learning Perceptual Causality from Video, Amy Fire and Song-Chun Zhu [slides, poster]
13:50-14:10 Rates for Inductive Learning of Compositional Models, Adrian Barbu, Maria Pavlovskaia, and Song-Chun Zhu [slides]
14:15-14:30 Poster advertisements:
    Symbol Acquisition for Task-Level Planning, George Konidaris, Leslie Pack Kaelbling, and Tomas Lozano-Perez
    The Construction of Reality in a Cognitive System, Michael S. P. Miller
    Events, Interest, Segmentation, Binding and Hierarchy, Richard Rohwer
    Learning Behavior Hierarchies via High-Dimensional Sensor Projection, Simon D. Levy, Suraj Bajracharya, and Ross W. Gayler
    Building on Deep Learning, Marc Pickett
    Autonomous Hierarchical POMDP Planning from Low-Level Sensors, Shawn Squire and Marie desJardins

14:30-15:30 Poster Session
15:30-16:00 Coffee Break

16:00-16:20 Representation Search through Generate and Test, Ashique Rupam Mahmood and Richard S. Sutton
16:25-17:15 Invited talk: Generative Stochastic Networks trainable by Backprop, Yoshua Bengio [slides]
17:20-18:00 Wrap up

Topics

We are interested in all parts of the bridge between low-level-sensors and rich high-level representations and their use in reasoning tasks. This includes but is not limited to:

Learning concept hierarchies from sensor data.

Representing and learning invariant concepts.

Postulating objects and theoretical entities.

Postulating relations from sensor data, when the data is not explicitly relational.

Learning symbolic representations from numerical sensor data.

High-level reasoning grounded in robotic sensors and effectors.

Sensor-grounded research on cognitive architectures.

Invited Speakers

• Yoshua Bengio, University of Montreal
• Jeff Hawkins, Numenta Inc.
• Benjamin Kuipers, University of Michigan
• Juergen Schmidhuber, Dalle Molle Institute for AI

 

Organizing Committee

• Marc Pickett, Naval Research Laboratory contact
• Benjamin Kuipers, University of Michigan
• Yann LeCun, New York University
• Clayton Morrison, University of Arizona

Program Committee

• David Crandall, Indiana University
• Joseph Modayil, University of Alberta
• James MacGlashan, Brown University
• Clayton Morrison, University of Arizona
• Jonathan Mugan, 21 CT
• Tim Oates, University of Maryland Baltimore County
• Marc Pickett, Naval Research Laboratory

---