"From Simple Associations to Systematic Reasoning"
We are capable of drawing a variety of inferences effortlessly, spontaneously,
and with remarkable efficiency --- as though these inferences are
response of our cognitive apparatus. This remarkable human ability poses
a challenge for cognitive science and computational neuroscience: How can
a system of slow neuron-like elements represent a large body of systematic
knowledge and perform a wide range of inferences with such speed?
SHRUTI attempts to address this challenge by demonstrating how a
network can encode a large body of semantic and episodic facts, systematic
rule-like mappings, knowledge about entities, and types, and yet perform
a wide range of reflexive inferences within a few hundred milliseconds.
Relational structures (frames, schemas) are represented in SHRUTI by
focal clusters of cells, and inference in SHRUTI corresponds to a transient
propagation of rhythmic activity over such cell-clusters.
Dynamic bindings between roles and entities are represented
within such a rhythmic activity by the synchronous firing of
appropriate role and entity cells.
Rules correspond to high-efficacy links between cell-clusters, and
long-term facts correspond to coincidence and coincidence-failure detector
circuits. In particular, SHRUTI demonstrates that temporal synchrony
in conjunction with structured neural representations suffices to support
rather complex forms of relational information processing in the brain.
SHRUTI's representational machinery has been augmented to
represent beliefs as well as utilities and actions (X-schemas).
The resulting system propagates utilities and beliefs over a single
underlying causal structure to make predictions, seek explanations, and
identify actions that increase expected future utility.
In on going work, neurally plausible mechanisms for control and
decision-making are being developed for SHRUTI. The integration of these
mechanisms with SHRUTI would lead to - SHRUTI-agent - a connectionist
architecture capable of decision making, problem solving, and planning.
The SHRUTI project meshes with the NTL project
on language acquisition and provides connectionist
solutions for representational and computational issues arising
in the project.
- Principal Investigator
- graduate student
V. Ajjanagadde - alumnus
Dean J. Grannes
A list of publication may be found here
The reflexive reasoning abilities of SHRUTI can be integrated with a
reflective (i.e., meta-cognitive) component resulting in a hybrid system
capable of attention shifting, conflict detection, and the reflective evaluation
of assumptions and alternate interpretations. This work was done in
of Cognitive Technologies.
Episodic Memory Formation in the Hippocampal System.
Agents and Inferential Retrieval.
Here are some demos generated using the SHRUTI simulator:
to the postoffice on President's Day.
fell in the hallway.
View the SHRUTI submission to ACNN96
ACNN'96 by clicking
Research on SHRUTI has been supported by the following: NSF grants ECS-9970890,
SBR-9720398 and IRI 88-05465, ONR grant N00014-93-1-1149, and DoD grant
MDA904-96-C-1156. Subcontracts from Cognitive
Technologies Inc. under ONR grant N00014-95-C-0182 and ARI Contract DASW01-97-C-0038.
ARO grants DAA29-84-9-0027 and DAAL03-89-C-0031 to the AI Research Center,
the University of Pennsylvania, DFG grant Schr 275/7-1 to V. Ajjanagadde,
and ICSI general funds. Work on the CM-5 version of Shruti was supported
by NSF Infrastructure Grant CDA-8722788 to L. Shastri.
"SHRUTI" is a Sanskrit word which refers to the oral tradition of communicating
knowledge. In oral communication, knowledge is encoded as a transient and
dynamic pattern of (acoustic) energy. This resonates with the functioning
of SHRUTI where reasoning is the transient propagation of a rhythmic pattern