PICTURE

Ian S. Eslick

I am a doctoral candidate at the MIT Media Lab co-advised by Prof. Frank Moss and Prof. Deb Roy with Marvin Minsky serving as my intellectual godfather. I am a member of the Cognitive Machines and Society of Mind research groups and collaborate closely with the Media Lab's Software Agents group.

Research Vision

My research contains two dominant themes: the construction of highly interactive, intelligent computer interfaces and their ability to augment human problem solving and communications skills. Following the approach championed by Marvin Minsky, and with the technical guidance of Deb Roy, I am building experimental programs to explore these themes.

Three central research questions guide my current work. Is there a practical representation of a user's naive model of procedural behavior, such as conditionality, sequencing, and parameterization? Can this representation enable incremental creation and "debugging" of machine behaviors by end-users using natural language? How do we leverage the collective intelligence of communities with common interests to evolve a system that accurately captures and responds to user intent across the domain of interest?

Research and Other Projects

Social Computation

This is an evolving perspective on my research that seeks to empower end-user communities to collaborate in the customization of application platforms. This requires a significant flattening of the learning curve and the ability to efficiently exploit users of all skill levels. Social Computation is an approach to this problem that combines collective intelligence, social networking, machine learning and natural language understanding to enable communities to access and extend computational functionality without programming.
PICTURE

ScratchTalk

Enabling a restricted form of natural language understanding is central to the Social Computation vision. ScratchTalk is an experimental platform that enables users to describe animations and simple games in natural language that are rendered as Scratch programs. This will yield insights into the semantics necessary to connect natural language to procedural behaviors. (active)

ConceptMiner

An information extraction project that extends the relational knowledge base, ConceptNet, using the Web. The project annotated the OpenMind sentence corpus with ConceptNet relation labels to enable automatic extraction of lexico-syntactic patterns that capture those relations for other words. Using a web search engine and a partially filled-in template returns a large set of candidate concept pairs. My MS thesis, Searching for Commonsense, identifies filtering techniques that improve the accuracy of these candidates so they can be combined with ConceptNet to improve the quality and scope of inference. 		PICTURE
PICTURE

GraphStep

 Working with long-time collaborator Andre DeHon, I assisted in the development and evaluation of a novel architecture for performing computations over large graph-oriented data structures. Specifically, we characterized the performance of GraphStep on the inference algorithms in ConceptNet, demonstrating significant improvements in performance, particularily in resource-efficiency. This research was supported by DARPA's ACIP program.

ConceptNet

Joint work led by Hugo Liu and Push Singh. The OpenMind corpus was created by volunteer users who characterized the relationships between English phrases using simple natural language templates. ConceptNet is a semantic network that uses hand-built lexico-syntactic templates to extract 20 semantic relation types between short english phrases (noun phrases and predicate-argument structures) in the corpus. 		PICTURE
PICTURE



PICTURE

Other Projects

I currently have strong formal or informal affiliations with a number of organizations: CM Capital (executive affiliate), New Enterprise Associates (venture advisor), MIT Venture Mentoring Service, The Experimental Study Group, Stretch Inc., ActivityInfo Inc., and Design that Matters.

As an MIT student in the early 90's I was a member of the AI Lab's Transit Project and the Reinventing Computing group. Our group's work on a parallel computing architecture called Matrix inspired Silicon Spice (The Edge - CNBC), a telecommunications company I founded with Ethan Mirsky, Robert French and Robert Ryan in 1996. We were acquired in 2000 by Broadcom Corporation for our VoIP product line. I served as a Director of Engineering at Broadcom through 2003 in the Carrier Access and Mobile Communications Business Units.

I returned to MIT in 2003 as a Research Affiliate with Patrick Winston's Genesis Group (2003-2004). I then enrolled at MIT's Media Lab to work with the newly formed Commonsense Computing group (2004-2006) on cognitive architectures inspired by Marvin Minsky. I co-led the early phase of a collaboration between these two groups in DARPA's Biologically Inspired Cognitive Architectures program (2005-2006).

Common Lisp

I am an avid fan of the Common Lisp language and use it in most of my research work. I am a lead developer on two open-source projects: the Elephant Persistent Object Metaprotocol and Database and the Langutils Natural Language Toolkit. I have developed additional systems and libraries at the Media Lab, including high-performance web-mining, and hope to release some of them as open source in the future. 		PICTURE

Bibliography

Contact Information

Ian S Eslick
MIT Media Lab, E15-320R
20 Ames Street
Cambridge, MA 02139
phone: 617-452-5634
eslick @ media > mit > edu