IEEE Technology and Society Magazine - Fall 2014 - 40

is above a certain threshold. The integrated chance constraints ensure that
the expected value of a constraint
function is above a certain threshold,
or interpretations based on distributional robustness and conditional
value-at-risk. In the context of traffic
management problems, stochastic
optimization methods are most appropriate when handling performance
constraints that are "soft" in a probabilistic sense, that is, the traffic management system is tasked with respecting
the constraint with a high likelihood.
Such constraints include expected
transit time and traffic flow targets.
Stochastic optimization methods
are also most useful for problems in
which large amounts of historical data
can be accessed to provide example
"scenarios" for modeling purposes.
In traffic management, historical data
relating to traffic inflows and congestion supply exactly these scenarios.
To support proactive event-driven
decision making, it is required to determine which aspects of the application
under consideration should be treated
in each way. More importantly, it is
necessary to develop real-time proactive planning tools using the aforementioned optimization methods within
an event-based planning framework.
These methods may then be employed
at a variety of levels of autonomy,
ranging from simple decision support
functions for human operators to fully
autonomous decision making.

Visual Analytics
While the aim of the methodology
is to automate much of the decisionmaking process, key points will
require people to make choices, and
the system realizing the methodology will require human monitoring.
For example, in traffic management,
determining the trade-off between
minimizing average journey times
and setting acceptable thresholds
on maximum wait times requires
human monitoring. Other tasks such
as communicating the traffic flow
and advising road users and road
planners, require operators to maintain a good mental model of the
40

dynamics of the road system, and
also of the decision-making system
itself (see the previous section). The
effectiveness of human decisions
will be enhanced to the extent that
the dynamics of the entire system
can be made transparent.
We need to address these issues
through visualization technologies
that are tuned to what is known about
human decision-making processes.
This can build on work regarding
online information foraging for decision making [20], and in the time
signature of the human cognitive
architecture, to drive new designs for
visualization. Subtle changes in the
time costs of making comparisons
can lead to macroscopic changes in
decision strategy [13] and, indeed,
we contend it is this regularity that
provides the key opportunity for visualization technologies. For example,
it is known that requiring users to
mouse-over icons in order to reveal
decision critical information reduces
the amount of information that users
retrieve, despite the fact that it only
adds hundreds of milliseconds to
the interaction. More interestingly,
mouse-over designs can shift users
from using non-compensatory to
more compensatory strategies. Conversely, presenting too much information all at once leads to visual
"crowding" and the potential for feature swap, e.g., numerical transposition errors, and therefore error.
Visualization technologies work,
not simply because they are visual, but
because, by enhancing the efficiency
with which people can compare
results, visualization can fundamentally modify the processes by which
decisions are made. In the proposed
system for proactive decision making, visualization design will emphasize comparison, as others have done,
but will do so as directed by recent
theory in the cognitive sciences [21].
We also need to push beyond the
individual. While much research on
visualization has focused on understanding the performance of individuals engaged in diagnosis tasks,
we contend that there is considerable

potential for new insights for the
design of collaborative visualization
technologies. Visual Analytics is not
simply the visualization of the output from analysis processes, but the
creation of insight in the decisionmakers working with these visualizations. The analysts are essentially
active participants in constructing
the manner in which these data are
to be processed, creating and revising associations between parts of the
dataset by manipulating the graphical
user interface [7].
To develop visual analytics for
decision support in Big Data applications, there is a need to apply concepts and principles from Ecological
Interface Design [18]. "Ecological
Interfaces" are designed to visualize
the manner in which physical components of the system map onto the,
more abstract, functions that the system performs. So, they are views of
the process that are not simply maps
of how physical components connect
to each other but are abstractions that
show how types of physical components affect particular functions. The
purpose of such designs is to improve
operator decision making and diagnosis when dealing with faults relating to those specific functions. This
means that the visualization will not
only display the model's input and
output, but also the relationships
between elements in the decision
space. One element of Ecological
Interface Design is simply the reflection of the constraints in the work
domain through constraints in the
user interface. In this way, the "ecologies" of the work domain, environment, and the organization become
reflected in the user interface through
the definition and management of
these constraints. Added to these
ecological constraints are constraints
from the analyst/modeller, such as
expectations and mental models.

Proactive Decision Making
Passively waiting until an unexpected event occurs and an opportunity is missed is an expensive way
to solve a problem. This method

IEEE TECHNOLOGY AND SOCIETY MAGAZINE

FALL 2014

Table of Contents for the Digital Edition of IEEE Technology and Society Magazine - Fall 2014