Breakout session on applications:
=================================
Several of the students remarked that applications are closely coupled
to the sensors that support them. We discussed weather and traffic
monitoring, "smart" mines and virtual data gloves. Medical apps were
felt to be very compelling, for home vital signs monitoring, drug
delivery etc. But none of these applications are currently in the
scope of Endeavour. The application that do fit well are collaboration
via telepresence, and classrooms. 

Output Devices:
It was noted that we have a lot of emphasis on sensors and monitoring
the world, but not much on affecting the world. i.e. Lots of sensors
but not many actuators. e.g. smart paper is a new kind of output
device. Building control (temperature, air flow) is another good app.

Existing Pieces:
The most closely related existing projects are 
  Iceberg, which has notions of personal preference lists and
     personal activity tracking
  Digital libraries, which has a notion of user annotations, personal
     view, and customization

Data Filtering/Compression
There followed an interesting discussion (mostly between Mike Jordan 
and Anthony Joseph) about data filtering: discovering interesting data
using interest metrics; deriving interest metrics by data modeling;
prediction as a form of data compression; 
Data fusion will be an important part of the project. It can be
formulated as a problem of combining probabilistic evidence. 

Vertical Slices
---------------
We felt that we should have some bottom-to-top slices through the
project as well as horizontal layers. Several themes were heard
throughout the project. Two of them were:

Moving Data: 
At the storage level, data should move as close as possible to the
agents that will access it. Applications determine what data needs to
be used. Collaborative applications in particular can make predictions
about where the data is likely to be needed, so that extensive
prefectching can be done. Some applications (but not all of them) need
to be aware of data location, and must be able to influence it. 

Context: Awareness of context (any part of the world state or user
state) is important for ubiquitous computing applications. The more
the system can infer context, the less explicit interaction is needed,
and the more useful work can be done by the computing
infrastructure. Context includes the location of people: on the road,
indoors, which room, are they in a conversation circle? Activity:
are they talking, working on a task, which task? Time of day etc. 

Milestones for Year 1
=====================

o People tracker for classrooms. Jitendra will develop a person tracker
  from multiple cameras in a classroom, based on Chris Bregler's work. 
  We will fit out one of the classrooms (310 or 320) for this.

o We will deploy a PRoP telerobot in a class setting. One of the 
  TAs will alternate discussion section presentations face-to-face
  and via the PRoP.


Universal Collaboration (Presentation)

Motivation
Interaction with people is still a killer app. 
Interaction thru ubiquitous computing makes many things harder:
Finding the right person at the right time.
Sensemaking for opaque documents.
Keeping track of distributed projects.
The Personal Touch: communication, persuasion, trust.


Project 1
PRoPs: Wireless robot appliances that act as proxys or avatars. 
What they could achieve:
 Mobility and access to remote workplaces: factories, offices, warehouses
 A better level of interpersonal interaction through non-verbal
communication.
 Recreation when its too far to go. 

Project 2
Tacit knowledge mining: use logs from single or multiple servers to 
mine information about:
 Personal expertise.
 Document authority.
 Document history and context.
 User activity and document use in a project.
 Perspectives on a document or meeting. 

Where this fits
 PRoPs: Are a class of internet appliance. They require ubiquitous 
    access and QoS. Complementary to other internet appliances: 
    phones and wearables. 

 Tacit knowledge mining: Use log data from single or multiple servers. 
    Could run as an agent that shadows a user, the resources 
    they use and the people they interact with.