COGNITIV

E
ARCHITE
ACT-R & SOAR

CTURES
FLOW 1 INTRODUCTION

2 COMMON MODEL OF COGNITION

3 ARCHITECTURE PROCESSING

4 TYPES OF AGENT DATA

5 SIMILARITIES & DIFFERENCES

INTRODUCT
ION
• The most mature & general common model
architectures
• Freely available
• Been applied to 100’s of tasks
• Supports real time performance (it is possible to
have complex agents that achieve real time
performance)
• Online incremental learning while they do real
time performance
• ACT-R: model human behaviour
• SOAR: model complex cognitive agents
 Community
COMMON MODEL OF Guide research on
consensus: an
abstract
COGNITION individual components

specification of
human-like
cognitive
architecture Testable theory for
cognitive structures
and functions

Baseline for
development and
cumulative progress Fixed set of
communicating
modules: Memories,
perception, motor and
architectural learning
associated with these
Not a prescription for The modules are all task independent different components
not all cognitive
architectures No task-specific learning modules
No NL, planning, navigation, attentional modules
Basic cycle: Selection
ARCHITECTURE
and execution of an
internal action from
PROCESSING • Procedural memory
drives the basic cycle
Procedural Memory
to change Working of the system
Memory • Both ACT-R and
SOAR have the same
architecture
• Basic Cycle runs for
Internal Action: 50 ms approx
• Internal reasoning
• Retrieval from
• Declarative
memory retrieval declarative memory
• Motor action not necessarily
execution asynchronous
• Perception
modulation
 TYPES OF AGENT DATA
Internal Agent Module Commands
Module Status Data
Data
• WM elements created by a
• WM elements created by PM
• Working memory, module in its WM buffer.
actions in WM buffers.
procedural memory, • Provide feedback on the
• Sent to a module (P, DM, M)
declarative memory. module's processes:
to initiate its processes.
• Unconstrained success/failure.
• Innate/fixed set of symbols
content. • Meta-process data: data about
that define module
architectural processing.
commands.
• Innate/fixed set of symbols.
 WORKING MEMORY COMMONALITIES
Relational graph structures
AND DIFFERENCES SOAR ACT-R
• Element is a triple: symbolic node, • Fixed number of buffers • Unlimited graph in
labeled edge, value. - no other WM elements. breadth and depth.
• Declarative chunk: Elements that • Exact number varies by • Rooted in state node
share node agent/model • Substates are created at
• Single level chunks in impasse
Buffers to other modules: DM, PM, P, M
buffers - no substructure. ⚬ Meta-process data for
• Chunks - data transfer to module
• Additional modules procedural memory.
• Module commands (fixed semantics)
include goal module, ⚬ Allows recursive
• Module status (read only - fixed
imaginal module,etc, "universal*
semantics) • Meta-process data subgoaling =
which is task dependent.
metagoals.
 WORKING MEMORY META DATA
COMMON
SOAR
Source (relational):
• Connection from WM chunk in DM chunk • All mentioned in common
• Used for updating DM chunks and DM • Activation (numeric):
metadata ⚬ Base-level activation of element - recency
Derivation (relational): and frequency of creation and access
• Instantiation of procedural memory element ⚬ Used in forgetting and for biasing DM
used to create WM elements
retrievals
• Used in procedural learning
• Highest substate (relational):
⚬ Used in returning results from substates
and procedural learning

ACT-R
• All mentioned in common
 PROCEDURAL MEMORY META DATA
ACT-R
COMMON
• All mentioned in common
• A single PM element is selected on • PM element = individual rule
cognitive cycle
• Rule-like elements with conditions and
SOAR
actions • All mentioned in common
⚬ Conditions test WM elements (not • PM element = operator = collection of rules.
metadata) • • Elaboration, proposal, selection, and
⚬ Actions modify WM elements: application rules.
⚬ Updates metadata (PM utility, WM • • Impasse if failed selection.
derivation, ...)
PROCEDURAL MEMORY ELEMENT: RULE
VS OPERATOR
ACT-R SOAR
• Individual rule • Operator
• Conditions match chunks in buffers • Elaboration rules: create WM structures tested in
• Select rule based on utility and degree of proposal and selection

match • Proposal rules: create proposed operator structures

• Selected rule actions change contents of • Selection rules: create preferences with ratings of
proposed operators
buffers
• Select operator based on preferences
• Selected operator apply rules change working memory
data
 DECLARATIVE MEMORY META DATA
ACT-R
COMMON
• All mentioned in common
• Base-level activation associated with
• Association strength between pairs of elements based
every chunk on co-occurrence
• Computed from recency and frequency of • Used in spreading activation
access
SOAR
• Influences future retrievals from DM
• All mentioned in common
• Base-level activation associated with every element

BASE LEVEL ACTIVATION reflects the recency • Used in spreading activation

and frequency of practice of the chunk i. e: The • Episodic memory has temporäl and sequencing
noise value
metadata for elements.
 DECLARATIVE MEMORY RETRIEVAL
DIFFERENCES
ACT-R SOAR
• Spontaneous retrieval • Direct Retrieval

DECLARATIVE MEMORY LEARNING

SOAR
ACT-R
• Direct storage of WM chunk to Semantic Memory
• Add chunk to DM when any buffer is
• Update DM chunk activation on chunk use and access.
cleared.
• All changes to WM recorded in Episodic Memory.
• Update DM chunk activation on chunk use
and access.
 3 TYPES OF
INFORMATION
AGENT METADATA
AGENT DATA • Associated with Agent Data MODULE STATUS DATA
• Created by perception, • Created by architectural • Meta-process data: created
processing as a side effect of by module processing
retrievals from procedural and
agent reasoning. • Opening for metacognition
declarative memory, and • Influences architectural
architectural learning • Tested by PM by cannot be
processing - retrievals and
learning. modified
mechanisms
• Tested and modified by PM • Does not "mix" with agent data -
cannot be tested or modified by
(except perception)
agent data.
 SIMILARITIES DIFFERENCES
• Agent data represented as graph structures: ACT-R SOAR
elements, chunks, and rules • WM: Fixed set of • WM: Unconstrained
• PM and DM retrievals biased by metadata: Change buffers graph structure:
WIM; update metadata; set module status • PM elements: single Multiple states
• Modules interact through buffers in WM: Provide rules • PM elements: operators:
meta-process data • PM retrieval is all or Multiple rules
• Agent Metadata none and no • PM retrieval can fail in
• Numeric and relational indication of failure multiple ways
• Updated in parallel with retrievals from PM and • Single declarative • Semantic and episodic
DM. memory
memory
• Not readable or modifiable.
• Activation used to focus on important memory
elements for selection and learning.
 Practical uses of act-r
Cognitive Psychology Human-Computer Education and Learning
Research Interaction (HCI) Sciences

• To model and simulate • To design & evaluate user • To model & predict
human cognition. interfaces, providing learning outcomes.
insights into how users • Helps design effective
• Use it to test hypotheses
interact with software & instructional strategies &
about cognitive processes,
systems. understand how students
memory, decision-making, acquire & retain
• Helps in optimizing
problem-solving, and interface design for knowledge.
learning. usability & efficiency.
 Practical uses of soar
Cognitive Modeling and AI and Robotics Air Traffic Control &
Psychology Research • Applied in development of Military Application
intelligent agents & robots.
• Provides framework for • Explored for modeling air
• Extensively used to simulate
creating agents that can traffic control operations,
human problem-solving, reason, plan, & make military command & control
decision-making, & learning. decisions in complex scenarios
• Create computational models environment. • Can simulate decision-making
that help explain & predict • Used in scenarios such as
processes in high-stress
human behavior in these autonomous navigation,
situations, aiding in the design
cognitive tasks. robotic control, and human-
& testing of more effective
robot interaction.
operational procedures.
THANK YOU FOR
LISTENING!