Information Encoding and Modality

1
Information Encoding and Modality

• Veerendra Allada
• Souvik Ray
• Neurocomputing Seminar

2
Outline

• Information encoding
• Types of encoding
• Engram
• Encoding specificity principle
• Modality
• Information modality
• Spatial organization of neural patterns
• Transformation networks

3
Information Processing in Brain

• Information encoding get information into the
  brain
• Information storage retain the information
• Information retrieval get the information back
  out later

4
Simplified Model of Human Memory(Multi-store
model, Atkinson and Shiffrin, 1968)
(Iconic, Echoic)
(Semantic)
(Acoustic)
5
Memory Models

• Individual memories are stored in distinct
  locations in the brain (Penfield, 1950)
• Engram model (Hubbard, 1950)
• The engram is the neural network representing
  fragments of past experiences which have been
  encoded
• Strong evidence that there are distinct elements
  of memory which involve different parts of the
  brain (e.g. hippocampus - short-term memory,
  amygdala emotional memories, hippocampus,
  temporal lobes and structures of limbic system
  long term memory)

6
(No Transcript)
7
(No Transcript)
8
Encoding

• Active process
• Requires selective attention to the material to
  be encoded
• Memories may be affected by the amount or type of
  attention devoted to the task of encoding the
  material
• Cocktail party phenomenon

9
Types of Encoding

• Shallow encoding
• Short-term memory repeating information to
  memorize it (e.g. phone no.)
• Utilizes the brains phonological loop (relies on
  part of brain designed for short-term memory)
• Elaborative encoding
• Link, associate or connect incoming information
  with something already in memory
• Visualization or guided imagery techniques can
  also be used to recall memories (Mnemonics)

10
Types of Encoding (based on depth of encoding)

• Structural encoding
• Emphasis on the physical structural
  characteristics of the stimulus
• Shallow level
• Phonemic encoding
• Emphasis on the sounds of the words
• Intermediate level
• Phonemic encoding
• Emphasis on the meaning
• Deep processing

11
Engrams How the Brain Stores Memories (Hubbard,
1950)

• Recording an event Strengthening the connections
  between groups of neurons that participate in
  encoding the experience
• Pattern of connections Engram
• Engrams are spread throughout the brain
  (equipotentiability)
• Engrams remain dormant unless retrieved from
  memory using cues
• Engrams weaken and disappear with time
• We tend to remember only important things

12
The Search for the Engram (Unit of Memory)

• Biochemical theories
• Memory storage occurs in biochemical changes at
  the synapse
• People with Alzheimers show a depletion of
  acetylcholine and glutamate
• Neural circuit theories
• There may be specific circuits for specific
  memories

13
Biochemical Theories

• Current view (Kandel theory, 1970)
• At molecular level, new proteins are manufactured
  and these proteins stabilize the changes
  underlying the memory
• Every new memory results in a permanent
  representation in the brain
• Routtenberg-Rekart theory (2005)
• Memory storage is a dynamic, meta-stable process
• Replication of memories across many different
  brain networks

14
Routtenberg-Rekart theory (contd.)

• For example, ones name is represented in
  innumerable neural circuits thus it is extremely
  difficult to forget
• Since no particular neural network lasts a
  lifetime, it is theoretically possible to forget
  ones name (advanced stages of Alzheimers
  disease)
• Brain stores long-term memory by rapidly changing
  the shapes of proteins already present at those
  synapses activated by learning

15
Neural Circuit Theories

• Short-term or immediate memory
• Temporary information is maintained in
  reverberating neural circuits which sustain a
  nerve impulse by channeling it repetitively
  through the same network
• Long-term memory
• Memories are stored in a distributed fashion- as
  synaptic strengths (weights) in a neural network
• No localization
• Graceful degradation
• Issue Difficulty of correlating artificial
  neural networks with biological ones lies in the
  way weights are modified in the former and
  synaptic weights are modified in the latter
• Backpropagation model (unrealistic)

16
Encoding Specificity Principle (Tulving)

• A specific way a person thinks about, or encodes
  an event determines what gets into the engram
• The recollection of an event depends on the
  interaction between the properties of the encoded
  event and the properties of the encoded retrieval
  information

17
Encoding Specificity Principle Context-Dependent
Memory

• Golden and Baddeley (1975)
• some scuba divers learned words on land and
  others underwater
• at test, each group was split in two and half
  were tested underwater and half were tested on
  land

18
(No Transcript)
19
Encoding Specificity Principle State-Dependent
Memory

• Same design as scuba experiment but using
  pharmacological states as context (Eich, 1980)

20
Problems with the Engram Theory

• Conditioning as an alternative
• A small fish in shallow, stale waters is bumped
  and hurt by larger fish trying to eat him
• The small fish gets an engram from this
  occurrence (pain and momentary unconsciousness)
• If attacked again, the engram is reinforced and
  the small fish avoids shallow, stale water in the
  future

21
Conditioning

• The small fish may at a later time figure out
  that shallow, stale waters do not always include
  danger and may return to those areas to feed
• Conditioning can be unlearned whereas engrams
  remain until audited out.

22
Encoding ve and -ve Values

• Biased Encoding
• Neuron firing at intermediate or background rate
  represents the value zero
• Firing at slower rate represents a negative value
  while faster rate represents a positive value
• Two-Cell Encoding
• Two different neurons encode a single numeric
  value
• Firing of neither represents zero firing of only
  one cell represents positive and firing of
  another represents negative value. The magnitude
  is proportional to the firing rate

23
Encoding Vectors and Complex Numbers

• Vectors quantities that represent both
  magnitude and direction
• Velocity, acceleration and surface orientation of
  visually perceived objects
• 6 neurons are logically grouped in to 3 pairs,
  each pair representing one component of a vector
• Complex numbers are represented as a pair of
  complex numbers and hence either by two cells if
  a biased encoding is used or by 4 cells if a two
  cell encoding is used for each real number.

24
Information Modality

• Human brain processes numerous modes of
  information

Modality Qualification Examples
Visual Shape, color, texture, intensity, position etc. Red, green, tree, smooth, dim, bright
Auditory Sound, harmony intensity, location Speech ,chords, loudness
Affect Joy, fear, anger, hunger
25
Spatial Organization of Neural Patterns

• Sensory information is encoded as 2-D patterns
  (in all sensory modalities)
• In a visual system, the retina encodes a 2-D
  representation of the ocular image
• Two aspects of sensory organs are of importance
• - Spatial Organization
• - Temporal Organization

26
Spatial Organisation

• Sensory patterns are composed of several
  independent sub patterns in registration with
  each other.

27
Transformation Networks

• How can alternate representations of a given
  sensory pattern be formed ?
• Are mechanisms for forming alternate
  representations prewired (genetically specified)
  or formed with experience ?
• Are the pathways that convey the different
  representations distinct or same ?

28
Contd..
Control Pattern to be stored
Storage and Access Control
Storage system for control patterns
Information output
Controlled information transformation network
Information Input
The essential components of a sensory processing
system capable of learning
29
Dimensionality of Encoded Patterns

• Dimensionality of a transformed pattern is not
  necessarily the same as that of the original
  pattern
• When the transformed pattern is smaller,
  information is lost

30
Temporal Organization

• Temporal organization of information likewise
  derives either from the external environment or
  from the neural substrate
• With in the time domain, the speed at which the
  sensory signals are transformed from one
  representation to another varies greatly,
  suggesting several different storage systems
• For example speech sounds must be processed at
  the phoneme level( milliseconds), at the syllable
  and word levels( tenths of seconds), at the
  sentence level( seconds) and at the thematic
  level( years).

31

• Thank You .