Information Technology Reference
In-Depth Information
In Fig.
6.1
, the analysis parameters
p
and
q
are marked alongside the appropri-
ate links. In principle parameters could be passed in either direction, but at a simple
level we may consider that
f
receives
p
as input, and emits
q
in a continuous pro-
cess. Moments of silence between events are then represented by streams of constant
parameters. The task of
P
is to deliver a parameter stream
{
p
}
to
f
, and that of
Q
is the sonification of an output stream
.
P
and
Q
act as transducers that enable
f
to interact with the external environment. The process can be formally represented
as
{
q
}
Ψ
out
=
Q
f
x,P(Ψ
in
)
≡
F(Ψ
in
)
where
x
is any internal state of
f
and it is assumed that
does not include any
of the system's own outputs.
F
is the
observable
function of the Live Algorithm.
f
itself is hidden. Performers only have access to each other's
F
; inferences about
private desires, goals, etc. in other words, performers
f
's, are made on the basis of
these observations.
The Live Algorithm must therefore decide when and what to transmit with regard
to the history of inputs and outputs,
{
Ψ
in
}
, the internal dynamic state
x
and any static parameterisation of
f
,
P
or
Q
(which may include data on previous
performances).
The task of finding an
f
that might satisfy the requirements of a Live Algorithm
is formidable. One way forward is to explore possibilities on an ad hoc basis (as is
common) and in the lack of any formalised objective this is the only available means
of development. The development of a performance measure for a Live Algorithm,
however, would suggest a more systematic methodology. The search for a perfor-
mance quantifier could commence by investigating human practice. If such a metric
could be established, it would guide development of
f
s; there is even the possibility
that a Live Algorithm could become its own critic and learn from experience.
Although the input streams are continuous, we may suppose for the purpose of
this discussion that inputs and outputs can be coarse-grained. For example, audio
samples can be grouped into a phrase of notes. The continuous streams could then
be considered as sequences of discrete events; the precise definition of what might
be a meaningful event, and the duration of the sequences is not important for the
following analysis. The streams can then be split into past, current and future and
comparisons can be made between them:
{
Ψ
in
}
and
{
Ψ
out
}
Ψ
out
contains elements
1. of past
Ψ
in
's (referential)
2. of current
Ψ
in
(reactive)
3. of future
Ψ
in
's (pre-emptive)
4. of past
Ψ
out
's (consistent)
5. of future
Ψ
out
's (planning)
6. not found in past or current
Ψ
in
's (independence)
7. not found in past
Ψ
out
's (exploratory).
