Michael H. Coen

ICML Conference 2010 Conference Paper

Comparing Clusterings in Space

• Michael H. Coen
• M. Hidayath Ansari
• Nathanael Fillmore

AAAI Conference 2007 Conference Paper

Learning to Sing Like a Bird: The Self-Supervised Acquisition of Birdsong

• Michael H. Coen

This paper presents a new framework for self-supervised sensorimotor learning. We demonstrate this framework with a system that learns to mimic a zebra finch, directly modeled on the dynamics of how male fledglings acquire birdsong from their fathers. Our system first listens to the song of an adult finch. By listening to its own initially nascent attempts at mimicry through an articulatory synthesizer, the system organizes motor maps generating its vocalizations. Our approach is founded on the notion of cross-modal clustering, introduced in (Coen 2005, 2006a), and is unusual for its recursive reuse of perceptual mechanisms in developing motor control. In this paper, we outline this framework, present its results on the unsupervised acquisition of birdsong, and discuss other potential applications.

AAAI Conference 2005 Conference Paper

Cross-Modal Clustering

• Michael H. Coen

This paper presents a self-supervised algorithm for learning perceptual structures based upon correlations in different sensory modalities. The brain and cognitive sciences have gathered an enormous body of neurological and phenomenological evidence in the past half century that demonstrates the extraordinary degree of interaction between sensory modalities during the course of ordinary perception. This paper presents a new framework for creating artificial perceptual systems inspired by these findings, where the primary architectural motif is the cross-modal transmission of perceptual information to enhance each sensory channel individually. The basic hypothesis underlying this approach is that the world has regularities – natural laws tend to correlate physical properties – and biological perceptory systems have evolved to take advantage of this. They share information continually and opportunistically across seemingly disparate perceptual channels, not epiphenomenologically, but rather as a fundamental component of normal perception. It is therefore essential that their artificial counterparts be able to share information synergistically within their perceptual channels, if they are to approach degrees of biological sophistication. This paper is a preliminary step in that direction.

AAAI Conference 2000 Conference Paper

Non-Deterministic Social Laws

• Michael H. Coen

The paper generalizes the notion of a social law, the foundation of the theory of artificial social systems developed for coordinating Multi-Agent Systems. In an artificial social system, its constituent agents are given a common social law to obey and are free to act within the confines it legislates, which are carefully designed to avoid inter-agent conflict and deadlock. In this paper, we argue that this framework can be overly restrictive in that social laws indiscriminately apply to all distributions of agent behavior, even when the probability of conflicting conditions arising is acceptably small. We define the notion of a non-deterministic social law applicable to a family of probability distributions that describe the expected behaviors of a system’s agents. We demonstrate that taking these distributions into account can lead to the formulation of more efficient social laws and the algorithms that adhere to them. We illustrate our approach with a traffic domain problem and demonstrate its utility through an extensive series of simulations.

AAAI Conference 1998 Conference Paper

Design Principles for Intelligent Environments

• Michael H. Coen

AAAI Conference 1994 Short Paper

SodaBot: A Software Agent Environment and Construction System

• Michael H. Coen

Much of the work done in the area of software agents can be placed into one of two categories: (1) highly theoretical treatment of agents’ intentions and capabilities; and (2) applied construction of specific agents. However, determining for what (and if) software agents are actually useful requires building many of them, and the agent construction process poses difficult technical challenges.