Science at the Speed of Thought - Ambient Intelligence for Scientific Discovery - page 13

the simulation needn't be written as a graphical collaboration. All it needs to
do is update shared memory.
Our system is designed to use as little network bandwidth as possible. For
example, the position of an object is given by six floating point numbers repre-
senting its location (X,Y,Z) and orientation (H, P, R Euler angles), and these
six numbers are updated only if the object actually moves. Six numbers can also
be used to specify the orientation of a node in a scenegraph (seven if a uniform
scale factor is needed), so complex animations can be achieved with minimal
data transfer.
Simulations also use shared memory as a data-passing mechanism, and multi-
ple simulations can read and write between themselves and the VE. Participants
in the VE will communicate with the simulations using shared memory too, so
the difference between modifications from a participant and from a simulation
is completely transparent.
Permissions and file loading/unloading will be handled using the Jini/Java-
Space distributed computing environment. This system will keep track of who
is allowed to do what, transfer and update data as needed, and handle problems
of resource allocations and conflicts. This functionality will be incorporated into
a DSO that will work in conjunction with the central server.
Additional DSOs will implement awareness tools that reflect the status of
the collaborative VE. For example, one DSO can display the relative position
of all participants in the VE on a two-dimensional radar display. DSOs can be
written to work best in an immersive environment or on a desktop, and users
can choose the DSOs that best suit their needs.

2.3 Machine Learning Tools

Representation is critical to understanding and learning. This is true for im-
mersive visualization, where productivity is dependent upon how well the data
is represented in the 3D world. Representation is also important for machine
learning. In that context, representation means the feature space of the data,
and the formalism for describing the data [35].
Finding the appropriate representation is important, but more than one rep-
resentation may be beneficial. A scientific paper uses multiple representations:
equations, plots, images, tables, diagrams, charts, drawings, and text. A scien-
tific presentation may also contain movies or animations of various types. All
of these are in support of communication of ideas. But these are also used by
scientists as they work towards their own understanding.
With the VL, different representations of data can exist side by side. They
can also be created interactively. Whatever aids understanding can be placed in
the virtual space.
Very large data sets present a challenge, even for visualization. Data sets
can be large in terms of the number of dimensions or the size of the dataset, or
both. While there exist techniques for viewing high dimensional data sets, such as
parallel coordinate plots [36], this may not be the best way to acquire knowledge
from the data. A different approach is to look for those dimensions that are most