Dan Tecuci and Bruce Porter
Intelligent systems need to store their experience
so that it can be reused. A memory for such systems
needs to efficiently organize and search previous
experience and to retrieve items relevant to
the situation at hand. It needs to be content addressable
while providing flexible matching. Previous
approaches to building such memories suffered
from being overly tied to a task or domain.
We propose to separate memory functionality from
that of the system as a way to reduce the complexity
of the overall system and to allow research to focus
on the generic aspects of memory organization and
retrieval in isolation of a specific domain and task.
We built such a generic memory for events. It employs
a representation of generic episodes, uses a
multi-layer indexing scheme and provides a generic
API to external systems.
We tested this memory module on three different
tasks in the logistics domain and showed that it performs
as well as serial search in terms of accuracy,
while beingmuch more efficient and more scalable.
@InProceedings{tecuci-flairs07, author = {Dan Tecuci and Bruce Porter}, title = {A {G}eneric {M}emory {M}odule for {E}vents}, note = {{P}roceedings to the 20th Florida Artificial Intelligence Research Society {C}onference ({FLAIRS}20) }, year = {2007}, address = {Key West, {FL}} }