Lecture 13 - CNNs and Emotions

Question 1

CNNS (Convolutional Neural Networks)

Answer 1

Image flattening (unrolling)
2d into 2d vector
Each pixel to different input

Convolution
3x3 receptive field
Bottom is the input image
Outcome is the top image
Yellow=receptive field (convolutional kernel)
Sum input x convolutional kernel
(add up all the 1s that make an X shape in a 3x3 box)
For this example in the slide, it is excited by Xs
Applied everywhere, in the entire image

AlexNet
The bigger the convolutional kernel (receptive field), the smaller the output image
Takeaway: bunch of convolutional layers early on, and can have multiple convolutional features learned in every layer

Edge detection
With combinations of edges, can look for facial features
Then combine eyes and noses and other features to form faces
The more convolutional layers, the more specific they get
Happens completely automatically
Similar to pandemonium theory
Similar to Cf v1-v4 in the brain

Question 2

Physical Symbol System

Answer 2

“A physical symbol system consists of a set of entities,
called symbols, which are physical patterns that can
occur as components of another type of entity called an
expression or symbol structure.”
– Newell & Simon

Symbols aka concepts
Relationships
Expressions

Ex: Symbols: S(C)- A(C) – A(P(C))
Also expressed in emojis
Expression: If I see cat, become aware (conscious) of cat, which triggers action to go pet the cat
Symbols, have relations between them, expression
Brain shifts and manipulates symbols

Connectionist representation
Ex: cat (Tabasco) is not a single symbol, but lives as a distribution of others concepts
Cat – 4 legs, hefty weight, long tail, 2 pointy ears

Question 3

Simon’s symbolic model of emotion

Answer 3

His idea: If only do symbolic reasoning, enough to create AI that is human-like
Added the idea of emotion
Higher-level system
Goal competition system (done, start)
When complete?
When satisficed? Ex: maybe just seeing face of cat enough to trigger action of going to pet
Timeout. Ex: if cat runs away when I try to chase it, there is a timeout point where would stop trying to chase it
(this is all part of the old symbolic system, before added emotion)
Realized missing:
-time constraints: if get hungry, might stop chasing cat
-survival: if car down the street when chasing cat to pet, might stop to prioritise your survival
Central nervous system… to interrupt

Simon’s SME, Features
● Auxiliary system monitoring others
● Interrupt ongoing processes
● Generates feelings

Simon’s SME, Criticisms
● Classification?
● Hierarchy of emotions?
● Physiological markers?
● Neuronal basis of emotion?

Question 4

Ekman’s Basic Emotions, 1971

Answer 4

How much is culture vs. innate?

Tested Papua New Guinea people of Fore culture
→ no English, no Pidgin
189 adults + 130 children

Found that it is universal

Ekman’s Basic Emotions, 1971
Picked 6 basic emotions:
● Fear
● Anger
● Sadness
● Surprise
● Disgust
● Happiness
Gotta go FASSDH
Claims:
1. Discrete emotions + distinct physiological features
2. Evo. functions + hardwired

E.g. Fear response
● Eyes widen → gather more information
● Heart rate up, sympathetic NS aroused → ready for action
● Finger temperature lower → blood to central body

Ekman’s Basic Emotions, Criticism
● Link between emotions & physiological response?
● Sociological impact?
-What if child grew up completely alone or with animals? Would they develop those same emotions?

Question 5

Russel’s Circumplex Model of Emotions, 1980

Answer 5

Valence (goodness), … pleasant to unpleasant
Arousal (engaging)… activation to deactivation

Question 6

Modern Alternative: Adolph&Anderson, 2018

Answer 6

● Scalability: can vary in intensity
● Valence: pleasantness
● Persistence: outlast stimulus
● Generalization: specificity to stimulus
● Global Coordination: engage whole organism
● Automaticity: how challenging to control
● Social coordination: social functions

→ Goal: interspecies framework

Question 7

Appraisal

Answer 7

Emotion → change in
perception of environment

Question 8

Emotional Episode and Tools for studying emotions

Answer 8

(see slide)

Tools for Studying Emotions
● Neural responses (EEG, fMRI, …)
● Somatic responses (heart rate, sweat, …)
● Affective responses (verbal automatic, cued)
● Genetic tools
○ Knockout experiments (e.g. Mice serotonin receptors)
○ Optogenetics
○ Pharmacogenetics
● Lesion studies
○ Temporary (TMS, neurotransmitters)
○ Permanent (neurotoxins)

Question 9

Summary

Answer 9

● CNNs:
○ Convolutional Kernels applied to whole image
○ Learns hierarchical features
● Early emotional theory:
○ Herbert: interrupts on symbolic execution
○ Ekman: hardcoded intercultural basic emotions (FASSDH)
● Affective spaces:
○ Circumplex model: Valence + Arousal
○ Adolph&Anderson: multidimensional model
○ Appraisal theory: how does emotion change perception?
● Affective toolkit:
○ Neural, somatic, affective, genetic, lesions