Past Lunch Bunches

2008 | 2007 | 2006 | 2005

Tuesday, January 22, 2008
Jason Marden and Adam Wierman
74 Jorgensen

A Game Theoretic Formulation of the Sensor Allocation Problem
Interdisciplinary research at the junction of information sciences, economics and game theory offers new solutions to several problems. This talk looks at applying game theory to solve a problem in cooperative control in distributed systems: the optimal locations of sensors in a sensor network.

This talk is introductory. The problem domain and basic concepts will be described. Detailed proofs will be given in other seminars and are also found in papers.

This talk presents a view of cooperative control using the language of learning in games. Specifically, we look at the cooperative control problem of dynamic sensor allocation. We formulate the sensor allocation problem as a noncooperative game where the decision makers are the sensors. In this setting, each sensor is assigned a local objective or utility function and is given the ability to autonomously alter it's position and sensing activity in real time. The distributed nature of the decision making allows for robustness to communication failures, sensor failures, and environmental changes. In this talk we will discuss several methods for designing the sensors' local utility functions. We measure the efficacy of a particular utility design in two ways:

(i) Does a Nash equilibrium exist
(ii) How efficient is a Nash equilibrium when compared to the optimal allocation?

Tuesday, January 29, 2008
Joel W. Burdick, Caltech
74 Jorgensen

Engineering Interfaces to Damaged Nervous Systems
This talk will introduce some of the challenges involved in trying to develop technology that can partially restore some functionality to people with severe neural deficiencies. The first part of the talk (which is based on joint work with Prof. Richard Andersen and Prof. Y.C. Tai of Caltech) will focus on neural prostheses. A neuroprosthetic is a brain-machine interface that can potentially enable a paralyzed human, via the use of surgically implanted electrode arrays, and associated computer decoding algorithms, to control external electromechanical devices. The problem of decoding the prosthetic user's intent from the recorded neural sisgnals is a problem in inference, or estimation. A few of the estimation techniques that have been used to solve this problem will be reviewed. Progress on making adaptive electrodes that autonomously optimize the quality of the brain-machine interface will also be reviewed.

The second half of the talk (which is based on joint work with Prof. Reggie Edgerton of UCLA and Prof. Y.C. Tai of Caltech) will focus on the problems involved in partially restoring locomotion after a severe spinal cord injury. While there are an enormous number complex issues surrounding a spinal injury, this talk will focus on the use of drug therapy, automated machines for physical therapy, and a new class of epidural spinal cord stimulators to help with the locomotion rehabilitation process. Each of these therapeutic components also have associated computational and algorithmic problems. For example, we are currently trying (in collaboration with Prof. Yaser Abu-Mostafa of Caltech) to formulate the problem of selecting the correct electrode stimulation protocols as a problem in kernel learning.

Tuesday, February 5, 2008
Paul Upchurch, Caltech & JPL
74 Jorgensen

Game Engines for Interactive Visualizations of Space Exploration
The number of people playing online games continues to increase each year. Game engine technology is generating innovations in hardware and software. This project investigates the use of game engines at JPL for interactive visualizations of explorations of the solar system.

This talk describes GoView a scriptable game engine which is tailored for rendering spacecraft trajectories in the solar system. Rendering trajectories across the solar system poses challenges that are not encountered in popular online games such as Doom. In particular, the game engine must be able to handle spatial scenes larger than six orders of magnitude. The talk describes the problems encountered and how they are solved in GoView.

Interactive visualizations of explorations of the solar system are being developed at JPL for public outreach, mission planning and mission operations.

Tuesday, February 12, 2008
Erik Winfree, Caltech
74 Jorgensen

Learning to Program Chemistry
Biological organisms are extremely sophisticated self-organized chemical systems. Their complexity dwarfs anything produced by modern chemical industries. The difference can be ascribed in large part to biological systems being information-processing machines: DNA, RNA, and proteins carry molecularly-encoded messages that direct all of life's processes. Biology is programmable chemistry. Can we learn how to program chemistry? How will this expand the capabilities of synthetic chemistry? Might non-biological chemistries also be programmable? What kinds of models of computation are needed for understanding how to program chemistry and the limits to doing so? I cannot give conclusive answers to these questions, but I will present examples of what we have learned while creating self-assembling structures, molecular machines, and logic circuits built out of DNA, RNA, and the occasional enzyme. I will argue that the key to unleashing the revolutionary potential of programmable molecular systems lies in understanding the design space and managing their complexity as information processing systems -- issues that are fundamental to computer science.

Tuesday, February 26, 2008
Paul De Martini, Southern California Edison
74 Jorgensen

Partnering with our Customers for a Smarter, Cleaner Energy Futu
SmartConnect, a new effort by Southern California Edison, will enable the utility to manage increasing demand for electricity from customers and provide environmental benefits by deploying a smart meter system which could reduce peak demand by as much as 1,000 megawatts - the output of a large power plant - as customers reduce some peak electricity usage and shift some peak usage to off-peak periods of the day when power costs less. Additional savings include lower labor costs due to the use of wireless data transfer from meters to the utility rather than manual meter reading. Paul De Martini, Director of SmartConnect, will describe the program, its development plan, and the technological challenges it faces.

Tuesday, March 4, 2008
Hsuan-Tien Lin, Caltech
74 Jorgensen

From Ordinal Ranking to Binary Classification
Ordinal ranking is an important concept in modeling our preferences. We rank hotels by stars to represent their quality; we give feed-backs to products on Amazon using a scale from one to five; we say that the weather is hot, warm, cool, or cold without referring to the actual temperature. The wide applications of ranking range from social science to behavioral science to information retrieval. For yet another example, in 2006, Netflix (an on-line DVD rental company) announced a million-dollar-prize challenge for building a better automatic personalized movie ranking system, and the prize is heating up the competition in machine learning and related areas.

Many machine learning approaches are designed in recent years to understand ordinal ranking better, but the design process can be time-consuming. Our work presents a novel alternative -- a reduction framework that systematically transforms ordinal ranking to simpler yes/no questions, i.e., binary classification.

Then, well-studied binary classification approaches can be effortlessly casted as new ordinal ranking ones. Furthermore, the reduction framework reveals a strong theoretical connection between ordinal ranking and binary classification, and allows us to easily extend well-known theoretical results for binary classification to new ones for ordinal ranking. In this talk, I will discuss the intuition and the construction of the reduction framework, as well as its theoretical and algorithmic merits, using the Netflix challenge as an example.

Tuesday, March 11, 2008
Jeanne Holm, JPL
74 Jorgensen

Virtual Worlds and Space Exploration
Populations of virtual worlds, such as Second Life, have grown rapidly. This talk discusses using virtual worlds for bringing lots of people into the NASA mission, letting them participate in the day to day work and successes of the U.S. space agency. NASA has established several islands in Second Life. NASA CoLab and Explorer Island are the two main public entrance points. These areas are geared at working with any person who is interested in learning more about NASA or, better yet, participating in a NASA mission for exploration in this virtual world. While NASA CoLab focuses on a place to host meetings and talks, Explorer Island (created by the Jet Propulsion Laboratory) is meant to be an immersive environment for interacting with spacecraft, being at a live launch of the Space Shuttle or a mission to Mars, talking to NASA scientists and engineers, sharing your ideas for space exploration in international workshops, and walking on the surface of another world. Come in world and look for Jet Burns (Charles White) or Devery Barrymore (Jeanne Holm) and we'll give you a tour!

Tuesday, March 18, 2008
Ulrich Pinkall, Berlin University of Technology
12:00 pm, 74 Jorgensen

What Game Engines Can Do For Mathematical Visualization
Surfaces with constant (mean or Gaussian) curvature are a classical topic in Differential Geometry. The shape of such surfaces is usually quite complex and poses a real challenge for Computer Graphics and interactive visualization. We will demonstrate that an optimal environment for doing mathematical experiments with surfaces is to put them in a virtual landscape where one can walk on the surfaces and interact with them in a way modeled after first person computer games.

| top |