Graphics Reference
In-Depth Information
A key feature of LiD4E is the provision for signal processing pipelines (i.e.,
chains of signal processing modules, such as classifiers, digital filters and transform
operators) that can be data-dependent and dynamically changing. LiD4E employs
hierarchical core functional dataflow
(
HCFDF
) semantics as the specific form of
dynamic dataflow [
35
]. HCFDF and the core functional dataflow (CFDF) model [
29
]
that it extends belong to the class of signal-processing-oriented dataflow models
of computation described in Sect.
12.4.1
. HCFDF can be viewed as a hierarchical
extension of CFDF. Through its emphasis on supporting structured, application-level
dynamic dataflow modeling, HCFDF provides a formal, model-based framework
through which stream mining applications can be designed and analyzed precisely
in terms of integrated principles of DDDAS and dataflow.
In HCFDF graphs, actors are specified in terms of sets of processing modes, where
each mode has static
dataflow rates
—i.e., each mode produces and consumes a fixed
number of data values (tokens) on each actor port. However, different modes of the
same actor can have different dataflow rates, and the actor mode can change from one
actor execution (
firing
) to the next, thereby allowing for dynamic dataflow behavior
(dynamic rates). Additionally, HCFDF allows dataflow graphs to be hierarchically
embedded (nested) within actors of higher level HCFDF graphs, thereby allowing
complex systems to be constructed and analyzed in a scalable manner. The design
rules prescribed for hierarchical composition in HCFDF graphs ensure that actors
at each level in a design hierarchy conform to the semantics of HCFDF or some
restricted subset of HCFDF semantics, such as cyclo-static dataflow or synchronous
dataflow (SDF) [
9
,
23
]. For further details on HCFDF semantics, we refer the reader
to [
35
].
As demonstrated in [
35
], HCFDF modeling enables run-time adaptation of signal
processing topologies, including dataflow graphs that are constructed using arbitrary
combinations of classifiers, filters, and transform units. Through the inclusion of
a special HCFDF design component called an
adaptive classification module
,the
designer can invoke multiple operating modes at run-time, and selection of such
operating modes can be driven based on system feedback—e.g., based on instru-
mentation that monitors data characteristics, and guides selection based on desired
trade-offs among performance, accuracy, and energy consumption.
Figure
12.3
provides an illustration of the LiD4E design tool and its application
to DDDAS-enabled, multimedia, stream mining system design. For more details on
LiD4E, we refer the reader to [
35
]. Extensions of the design principles in LiD4E to
handle multi-mode stream mining systems are discussed in [
34
].
12.5 Case Study: Learning Based on Multi-armed Bandits
In this section, we present a case study in data-driven ASM techniques that are
relevant for the emerging class of a ASM-enabled, embedded computer vision sys-
tems introduced in Sect.
12.1
through Sect.
12.4
. The methods presented in the case
study can be viewed as representative of the kinds of advances that are needed to
