1. Research Circle – PHD – 2016
By, Amr Kamel Ahmed
Supervised by, Dr. Alaa Hamdi
2. Cognitive architecture is an engineering approach for
modeling cognitive systems.
Cognitive systems could be bug, bird, animal or human
But in most cases it is human
It doesn’t model static structure only
It models cognitive behavior also
3. A single set of mechanisms that account for all of cognition
(using the term broadly to include perception and motor
control). For example,
Language
Problem solving
Dreaming.
Some of the things a UTC must explain are:
How intelligent organisms respond flexibly to the
environment
How they exhibit goal-directed behavior and choose goals
rationally (and in response to interrupts: see previous point)
How they use symbols
How they learn from experience
4. Psychology
Evaluating cognitive models
Studying learning techniques and methods
Understanding cognition models of specific tasks i.e.
Vehicle driving
Computer Programming
Robotics
Transformation from systems that do limited low level tasks
extremely well to systems that perform wide range of tasks
with acceptable results
Artificial Brain Projects
7. The roots of the distinction of human memory returns to Atkinson-
Shiffrin model proposed in 1968
They assert that human memory has three separate components:-
Sensory register, where sensory information enters memory.
Short-Term store, also called working memory or short-term memory,
which holds inputs from sensory register and long-term store.
Long-Term store, where information which has been rehearsed in the
short-term store is held indefinitely.
8. Each sensor has its own sensory register
Do not process information carried by stimulus
Detect and hold information for use by Short-Term memory
Information only transferred to short-term memory when
attention is given to it, otherwise decays rapidly and forgotten
Iconic Memory:-
Sensory memory associated with visual system.
It was experimentally shown that it is separated to short-term and
long-term memory
Decays after 0.5 – 1.0 seconds
Echoic Memory
Associated with auditory system
Holds superficial aspects of sound (e.g. pitch, temp or rhythm)
Having duration between 1.5 and 5 seconds
9. Attended information is transferred to Short-Term Memory (STM)
Information that enters STM decays and lost as well as sensory
memory
However it usually stays longer than sensory memory (18 – 20 seconds)
Information can stay at STM for much longer time through rehearsal
Information in STM doesn’t have to be of the same modality
Example, written text enters as visual can be held as auditory and vise
versa
There is limit to the amount of information that can be held in STM to
7 +/- 2 chuncks
10. It is a process involves repeating information over and over
in order to get the information processed and stored in
memory
There are two types of rehearsal
Maintenance rehearsal
It is useful in maintaining information in STM. However; it is not an
effective way of having information processed and transferred into
Long-Term Memory
Example, repeating a phone number out loud until put the number
onto the phone to make the call.
Elaborative Rehearsal
This type of rehearsal if more effective in transfering information
from STM to Long-Term Memory
It involves thinking about the meaning of the information and
connecting it to other information already stored in memory
11. Is the memory stage where information can be stored for long periods
of time.
Information can be maintained in Long-Term Memory (LTM)
indefinitely
Two types:-
Explicit Memory (Declarative) refers to all memories that are
consciously available, includes:-
Episodic Memory:-
Refers to memory for specific events in time as well as supporting their formation
and retreival
Semantic Memory
Refers to memory of factual information such as meaning of word
Autobiographical Memory
Refers to knowledge about events and personal experience from an individual’s
own life.
Similar to episodic but it is related to individual lifespan
Implicit Memory (Procedural Memory)
Refers to the use of objects or movements of the body
Such as how to use a pencil, drive a car or ride a bicycle
13. Soar is a general cognitive
architecture for developing
systems that exhibit
intelligent behavior
It has been in use since 1983
evolving through many
different versions to where it
is now Soar, Version 9
http://soar.eecs.umich.edu/
14. Adaptive Control of
Thought – Rational (ACT-
R)
The ACT theory has origins
in the Human Associative
Memory (HAM) theory of
human memory
A production system was
proposed that procedural
knowledge was
implemented by
production rules
15. includes three major brain
systems:
Posterior cortex
Perceptual and semantic
processing
Using slow, integrative learning
Hippocampus
Specialized for rapid encoding
of novel information
Using fast, arbitrary learning
Frontal cortex / basal ganglia
complex
Active and flexible maintenance
of goals and other context
information,
Serves to control or bias
processing throughout the
system
16. Action-Centered Subsystem (ACS)
Control actions
i.e., to maintain and apply procedural
knowledge
Non-Action-Centered Subsystem
(NACS)
to maintain general knowledge
i.e., declarative knowledge
Motivational Subsystem (MS)
to provide underlying motivations
for perception, action, and
cognition, in terms of providing
impetus and feedback
Meta-Cognitive Subsystem (MCS)
to monitor, direct, and modify the
operations of the other
subsystems for better
performance
17. Based primarily on Global Workspace Theory
LIDA simplified cognitive cycle constitutes a unit of sensing,
attending and acting
LIDA cognitive cycle divided into three phases:-
Understanding phase
Attention / conscious phase
Action selection phase
18.
19. Cognitive computing aims to develop a coherent, unified,
universal mechanism inspired by the mind’s capabilities
it seeks to implement a unified computational theory of
the mind
2007, IBM developed C2 mammalian scale near real-time
cortical simulator
A notable C2 innovation is the memory efficient
representation of synaptic state significantly increasing
model scale.
In 2013, IBM introduced the TrueNorth neuromorphic
CMOS chip.
It consists of 4096 HW cores
Each one simulating 256 programmable neurons
Each neuron has 256 programmable synapses
20.
21. Although there are hybrid cognitive architectures which
includes both symbolic and sub-symbolic subsystems
however they are separated islands
There are no integration between symbolic and sub-
symbolic subsystems that enables the usage of symbolic
representations based on sub-symblic platform.
All symbolic representations relies only on Von Neumann
architecture
There is no sub-symbolic architecture that allow causality
22. How to create symbolic cognitive architecture relies on
neural-based platform
How to build neural-based hardware platform suitable for
cognitive architectures:-
Nano scale neurons
Very low power consumption
Low cost of implementation
Scalable to include massive number of neurons (10^8 neurons
scale)
How to integrate both paradigms into single cognitive
architecture that realizes Allen Newell aspects for Unified
Theory of Cognition