ALIHT 2011 - The second annual workshop on Agents Learning Interactively from Human Teachers (ALIHT)

The second annual workshop on Agents Learning Interactively from Human
Teachers (ALIHT)

A two-day workshop at IJCAI-11 in Barcelona, Spain.

http://www.cs.utexas.edu/~bradknox/IJCAI-ALIHT11/H...


Important dates


Paper submission deadline: April 6, 2011
Acceptance notification: May 1, 2011
Preliminary program online: May 6, 2011
Camera ready deadline: May 14, 2011
ALIHT workshop at IJCAI-11: 2 days within July 16-18, 2011


Description


As intelligent robotic and software agents populate everyday human
environments, they will need the ability to adapt. For such agents,
human teachers present a narrow but critical learning environment.
Research in interactive learning seeks to enable agents to learn from
human instruction, harnessing human expertise to learn tasks and
customizing behavior to match human preferences.

Machines cannot be imbued by their designers with all of the knowledge
and skills that they will need to serve useful, long-term roles in our
dynamic world. If humans and machines are to work together in
challenging environments, then machines must be able to learn quickly
from any knowledgeable human, not just expert programmers. The
environments in which humans act are highly diverse and constantly
changing --- whether at home, in the office, or out in the field. Even
a machine with exhaustive background knowledge must be customized for
the particular environment in which it acts. Between humans, the
interactive student-teacher framework is known to be effective and,
furthermore, is familiar to every human; even young children teach one
another games and novel skills. Endowing machines with human-like
learning capabilities allows humans to teach such machines as they
would teach other humans and thus exploits skills that humans already
possess. Another strength of an interactive learning approach is in
the generality of the student-teacher mechanism; namely, that it might
be applied to a wide range of tasks with minimal, ideally no,
modification or specialization.

Research in interactive learning encompasses a wide range of
approaches and is referenced under many names, including teachable
agents, imitation, learning from demonstration, active learning,
interactive shaping, and bootstrapped learning. This workshop aims to
advance research in interactive learning by bringing together
researchers with different approaches and attempting to establish
standards for comparing systems, evaluating their performance,
combining results from different aspects of the problem, and creating
an organizing structure to situate the body of work within.

The potential application domains for interactive learning are many
--- both for autonomous robots and for virtual agents --- and range
across a spectrum of abstractness from fine motor skills to deep,
abstract knowledge such as that acquired at a university. An
interactive learning system therefore may be formulated in a variety
of ways, and we encourage a broad spectrum of applications. In
particular, we welcome contributions from both the robotic and
software agents communities. One of our main goals is to bridge this
divide, to create a single community of research on interactive
learning agents.


Submissions


We invite short and long length papers, reviews, and position papers.
Submissions will be judged on technical merit, the potential to
generate discussion, and their ability to foster collaboration within
the community.

The maximum length is 6 pages and papers should follow IJCAI
formatting guidelines.

Questions should be directed to Brad: bradknox-at-cs.utexas.edu