Self-organizing Hierarchical
Optimal Subspace Learning
and Inference Framework
SHOSLIF is a framework for comprehensive
visual learning for the task of object recognition and segmentation from
2-D images. Motivated by the concept of comprehensive learning, the SHOSLIF
is fully adaptive in the sense that there is little constraint on the shape
of the objects the system can deal with. Visual learning plays a fundamental
rule at various stages of the system. The structure of the network itself
is a function of learning.
In order to systematically develop the task-independent part of intelligent
learning and control, SHOSLIF uses a core-shell model. The core accomplishes
basic functionality of intelligence, such as memory, recall, reasoning,
and inference. The SHOSLIF core C
= (N,L,R) has three components: a network N
as a knowledge base, a learning procedure L,
and a knowledge retrieval procedure R. The
shell serves an an interface between the sensor or actuator and the core.
With this core-shell structure, the core is independent of specific
tasks. The same core can be used for many different tasks. The shell applies
the core to different types of sensory data and different instances of knowledge
base.
For more information on the SHOSLIF, see Laura Blackwood's paper at
/user/prip4003/blackwoo/papers/sail2.ps or
Dr. John Weng's paper at /user/prip4003/blackwoo/papers/weng.ps
which can be reached from any of the Computer Science UNIX machines or remotely
by connecting to the cps.msu.edu domain.