|
Agenda
|
||
|
Thursday
August 15, 2002
|
||
| 8:30 am | Breakfast | Courtyard outside 306 Soda Hall |
| 9:00 am |
Welcome
& Introduction to EECS |
306 Soda Hall |
| 9:15 am |
Center
for Intelligent Systems (CIS) Overview |
306 Soda Hall |
| 9:35 am |
Keynote:
Developing Cognitive Systems |
306 Soda Hall |
| 10:20 am | Break | |
| 10:50 am |
From Images to Objects, People and Activities Jitendra Malik |
306 Soda Hall |
| 11:25 am | Words,
Pictures & People David Forsyth |
306 Soda Hall |
| 12:00 pm | LUNCH, with Poster Session | Wozniak Lounge (430 Soda Hall) |
| 2:00 pm | Open
Problems in Speech Recognition Nelson Morgan |
306 Soda Hall |
| Speech recognition has been a significant research topic for at least 50 years. During the last 5-10 years, a significant number of commercial applications have appeared, including dictation programs that are available or less than $100. Additionally, speech-driven telephone query systems are an effective replacement for touch-tone entry, both due to being easier to use and permitting much more interesting features. Despite these successes, a significant number of research challenges remain. Speech recognition systems typically fail for moderate amounts of noise or room reverberation, when the speaker speaks quickly or with an accent, or in general for unexpected or casual articulation. In one standard government sponsored task, for instance, the best error rate for conversational speech is still roughly 25-30% (in a system that is 400x real time with a fast PC), for a case where there is little background noise, environmental acoustics, or speakers with a foreign accent. One secret of the current systems' success is that they have found niches for which these limitations can be sidestepped. To use speech more generally as input for intelligent systems, it will be necessary to overcome such limitations. I will describe some of these problems and a set of approaches that are current being explored to achieve some measure of solution. Some of these will be developed as part of the new DARPA EARS program that is now part of the Information Awareness Office. | ||
| 2:30 pm | Graphical
Models for Pattern Recognition and Probabilistic Reasoning Michael Jordan |
306 Soda Hall |
| 3:00 pm | From
Data to Knowledge Stuart Russell |
306 Soda Hall |
| 3:30 pm | Break | |
| 4:00 pm | Digital
Libraries: Re-inventing Scholarly Information Dissemination and Use Robert Wilensky |
306 Soda Hall |
| 4:30 pm |
Embodied
Language Understanding, |
306 Soda Hall |
| Natural
Language Understanding (NLU) is necessary for cognitive systems to meet current and anticipated requirements. Several of the theoretical and practical barriers to scalable NLU have been overcome and its time for renewed experimentation on what can be achieved. Recent theoretical innovations include inference with active and probabilistic knowledge and general conceptual primitives. These are compatible with large scale knowledge management efforts such as the semantic web, with connectionist language models and with computational learning methods, providing a promising substrate for large scale NLU applications. |
||
|
5:00 pm
|
Panel:
Information, Reasoning, and Awareness Tom Armour (DARPA IAO) Ron Kohavi (Blue Martini) R.V. Guha (Alpiri) Peter Norvig (Google) John Prange (ARDA) |
306 Soda Hall
|
| 6:30 pm | CIS Dinner for faculty and visitors | Women's
Faculty Club Dining Room |
|
Friday
August 16, 2002
|
||
| 8:30 am | Breakfast | Courtyard outside 306 Soda Hall |
| 9:00 am | Intelligent
Systems Work at NSF Peter Freeman, Director, NSF CISE |
306 Soda Hall |
| 9:20 am | Research
on Flying Robots at UC Berkeley Shankar Sastry |
306 Soda Hall |
| 9:50 am | Millirobotics
for Remote Sensing and Presence Ron Fearing |
306 Soda Hall |
| Millimeter-scale
robots will create the opportunity to sense and explore a wide range of
environments with greatly reduced cost. We are developing one tenth gram
flying robots, which will work safely, non-disruptively, and cooperatively
with humans. The work of the Center for Intelligent Systems will provide
the perceptual, planning, learning, and reasoning capabilities necessary for these small robots to survive in diverse indoor environments. In addition, these small robots provide a unique capability for dynamic, reconfigurable sensor placement for learning about environments. |
||
| 10:20 am | Break | 306 Soda Hall |
| 10:50 am | Making
Silicon Walk Kris Pister |
306 Soda Hall |
| MEMS technology allows the creation of sensors and actuators on a sub-millimeter scale. Since the late 1980s, there has been speculation that this may allow fully autonomous micro robots. Such robots are on the verge of become reality. | ||
|
|
Panel:
Autonomy Moderated by Shankar Sastry Daniel Clancy (NASA) Rakesh Gupta (Honda) Mark Swinson (Sandia Labs) |
306 Soda Hall
|
| 12:00 am | LUNCH | Wozniak Lounge (430 Soda Hall) |
| 1:30 pm | Kernel
Methods in Supervised and Unsupervised Learning Michael Jordan |
306 Soda Hall |
| 2:00 pm |
Optimization and Control vs Statistics : Merge or Fail? Laurent El Ghaoui, Optimization Research Group |
306 Soda Hall |
| Recent
progresses in optimization and control, statistics and estimation begin
to blur the boundaries between these fields. The traditional separation
theorems that fostered these boundaries simply break down in the face of
complex, uncertain systems, making some of the classical tools obsolete
or inefficient. Instead, we find new similarities that offer both challenges
and promises. In this talk, we describe how the ubiquituous concept of "convex
duality" governs the most efficient algorithms, for automatic control
of so-called hybrid systems, optimization of networks, classification of
gene expression levels in bioinformatics. We also emphasize the pressing
need for a more integrated perspective of statistics and optimization, both
in the research and teaching domains. |
||
| 2:00 pm | Statistical
Learning: pattern classification, prediction and control Peter Bartlett |
306 Soda Hall |
| The
talk reviews advances in the theory of pattern classification, focusing
on large margin classifiers, a family of techniques that have proven successful
in many applications. It also presents some speculations on how these successes
might be extended to more complex decision problems, such as prediction
problems that arise in natural language processing and bioinformatics, and
control problems. |
||
| 3:00 pm |
Adaptive
Intelligent Agents |
306 Soda Hall |
| 3:30 pm | Break | |
|
4:00 pm
|
Panel:
The Future of Intelligent Systems Moderated by Stuart Russell Peter Freeman
(NSF CISE) |
306 Soda Hall
|
| 4:45 pm | Closing remarks | 306 Soda Hall |
|
6:00 PM
|
Optional Dinner:
|
Lalimes
Restaurant 1329 Gilman Street Berkeley, CA (510) 527 9838 |