Subject: RESCUE MEETING
Date: 02/13/04
Loc: URP
Author: AM
Status presentations of the various Rescue subgroups. Prep for upcoming visits and demos.
2:00 - 4:00 Groups
presentations. All students working on projects/demos to present
4:00 - 5:00 Break out into
discussion groups
SM: *Dmitri in charge of meeting, equipment.
March 17 infrastructure, site visit.
Net & video infrastructure at UCI.
Map to an emergency drill.
The equipment will help with the drills.
*Need 2 students/people at 25% time; asking for volunteers.
[Everyone takes two steps back.]
*Will assign volunteers offline.
DK: [CAMAS1 Status Presentation][slide show]
HL: [Status][slideshow]
Run and evaluation of facilities data on original Camas system.
SM: (1) Representation of events in db.
(2) Location – test Room # extraction
(3) Date – granularity
*Don’t artificially create raw report.
Matching to problem type is too easy.
The keywords almost always appear in the “problem type” field.
*Demo extraction using context.
Rule-based, eg, VisualText.
Stanford method.
- Auto extract conent of web pages.
- Non-ML, ad hoc auto extraction papers
ML-based, e.g., Mooney at Austin.
*NEW. Extraction given context, knowledge.
- Looking up in db (e.g., room #)
*Triaging.
*Adaptive filtering.
We’re not showing the research aspects yet.
CB: How related to research, facilities?
*STORY is important at this point.
Start with something simple, available like facilities data.
SM: Raw data -> event.
Voice, video, sensor data, text all combined.
1st year == TEXT. Then on to other sources.
CB: These are “problems” similar to the rescue tasks.
SM: What is new, researchy.
CB: Issue is funding, not 1st year review.
All: Have research funding, need equiemtn.
CB: Event occurrence, given a report.
SM: *Extraction with context and knowledge.
Framework for event IE with context and knowledge.
What analyzer?
Why manual phrase lookup?
AM: [Status]
[Deferred dataset slides]
[Demo of analyzers and their status]
Scramble – process and dumb scramble of facilities data.
Simple scrambling won’t work.
Camaskb – process problem/location tree from Haimin.
Synonym – VisualText version of synonym handling.
Camas1 – process the old Camas input texts [not shown].
Hdesk – process UCI help desk emails.
Crime – process, normalize various police crime logs for Alternate 911.
All: [Discuss] Synonym analyzer – take discussion offline.
KA: Automated ML-based methods for crime logs with given problem type.
SM: Is it a new problem [event] or one of 20 existing problems.
*Interact with use to get to problem.
*Classify problems, differences among problems.
E.g., disambiguate 5 car accidents based on differences.
Reading off the problem list to the user is not research.
Jehan: [Privacy Preserving Video Surveillance][slideshow]
Mahesh: 65m range vs 5m range. Two types.
RFID tags.
Collisions, walls.
CB: Indirect inferences.
Fm anomalies in trajectories.
Eg, movement of people in groups.
SM: *Privacy preservation in terms of data collection.
YM: [Adaptive Filtering]
[This notetaker left about 4:50pm –AM]
DK: In charge of March 17 meeting and equipment.
AM: Extract problem types from facilities reports.
AM: Ask Leslie about CAD data from facilities.