home

Ian Fasel

Assistant Professor
School of Information: Science, Technology and Arts,
and Department of Computer Science,
Executive Committee, Cognitive Science Program
University of Arizona

Gould-Simpson Building
1040 E. 4th Street
Tucson, AZ 85721

(520)477-7626

ianfasel@sista.arïzona.edu

Research

In the CACTUS Lab (Cognitive Agents: Curious, Thinking, Understanding Systems), we study human cognition through (a) computational analysis of the microstructure of human behavior, such as facial expressions and body movements, and (b) development of robots that interact with humans and objects, both through intrinsically-motivated, autonomous exploration, and through active engagement with human teachers. Below are some of the research areas:

Autonomous, language learning robots:

A current focus in the CACTUS Lab is “active learning” agents, i.e., agents that have to make intelligent decisions about how to move about space and orient their sensors over time in order learn about their environment as quickly as possible. Not only are our robots “born” not knowing where or what the objects in the room are, but they often don’t even know the specific response characteristics of their own sensors. These all must be learned from experience. Thus for these robots, sensing is not a passive process, in which someone presents a stimuli which must then be classified, rather it is an active information gathering process that requires intelligent decision making. Key to this work is the ability to learn and reason about how moving one's own body might change one's perceptions, and how different types of objects might behave when acted upon. Visual, haptic, and other sensory information are all combined to develop much richer representations of concepts than just whether a patch of pixels looks like a face, a car, a bike, or some other object. Ultimately our aim is for agents to be able to learn and truly understand a primitive set of language-like concepts about space and physical properties that can serve as the basis for natural interaction with humans and rapid adaptation to new environments.

Understanding Articulatory Speech:

This line of work involves analysis of the human articulatory apparatus during speech. Using a variety of sensors in addition to audio, including ultrasound, electroglottograpy, nasal airflow, and 3D video of the face, our aim is to develop a more complete understanding of the process by which humans produce sounds of all types, including language and music, and how we understand them. This work has broad potential applications, not only in phonology and linguistics, but also in speech recognition, language documentation, language learning, speech therapy, and music. This work is in collaboration with Linguistics here at UofA.

“Mind-Reading” for social decision-making:

Social decisions are not driven solely by money – emotions have important effects on our choices. For example, a decision-maker can decide to punish the person he is interacting with even at the cost of losing money. In this line of work, we use automatic analysis of facial expressions and of fMRI brain data during social decision making to predict the behavior of decision-makers. In particular we are interested in finding specific facial actions and brain areas associated with behaviors such as punishing unfair treatment, exploration versus exploitation decisions, and moral and social disgust. Our computational analysis of decision-making under uncertainty is unique in the field of neuroeconomics, and has wide application in social robotics and other interactive devices.

Human-Robot Interaction:

Many of the projects in the CACTUS Lab fall under the umbrella of “human-robot interaction” (HRI). In addition to the projects mentioned above, other HRI sppecific projects involve e.g., a large distributed multi-player game, several different “Wizard of Oz” experiments involving humans teaching robots, and recognition of users' identity and specific touch gestures on touch-sensitive tablets. Several collaborators in cognitive science, psychology, and economics are involved in this work. Despite the apparent diversity in these research topics, they all share a common methodology of applying modern machine learning methods to extract deeper, hidden information about human internal states from temporal sequences of data.

Together, the behavioral science and robotics work in the CACTUS Lab may be seen as two, convergent approaches to understanding intelligence, by respectively analyzing how nature's best example of intelligent agents – namely, people – behave, and by formulating hypotheses for how to produce intelligent behavior in real-world autonomous agents.

Students:

Below are students who are doing research projects in the CACTUS lab. Note that several of these students also have their main advisor in another department – in particular, Jeff Berry’s advisor is Diana Archangeli in Linguistics, and Filippo Rossi’s advisor is Alan Sanfey in psychology.

Grad Students

Farnaz Abtahi

Jeff Berry (linguistics)

Nassim Mafi

Mohsen Malmir

Anton Rebguns

Filippo Rossi (psychology)

Raquel Torres Peralta

Undergraduates

Peter Allen Brown

Matt DePorter

Postdocs