Please enable JavaScript.
Coggle requires JavaScript to display documents.
Learning Deep Physiological Models of Affect (Deep Learning (can…
Learning Deep Physiological Models of Affect
human-computer interaction (HCI)
Affective computing (AC)
detecting and modelling emotion
artificial intelligence (AI)
key challenges :red_flag:
vague definitions
#
fuzzy boundaries
#
model processing
#
input (emotional manifestations + annotated affective states)
a) behavioural responses
b) bodily response (physio-logical)
c) interaction
model (processing signal + express manifestations)
#
output (affective annotation)
input -> model
#
computational model
feature extraction / selection
automatic
ad-hoc
semi-automatic
Deep Learning
can
artificial neural network (ANN)
#
raw data
discrete / continues
#
reduce signal resolution
#
preference-based (ranking-based)
preference deep learning (PDL)
#
emotional manifestations
#
relaxation
anxiety
excitement
fun
phsiological signals
skin conductance
blood volume pulse
36 participants
feature extraction
raw signal
skin conductance sensor
microphone
camera
blood volume pulse
accelerameter
Training Models of Affect
traditionally in AC
#
limitations
manual feature extraction
expert
inappropriate detectors
cannot capture he manifestations
#
large degree
principal component analysis (PCA)
Fisher projection
priori
#
#
discrete signals
dynamic affect modeling approaches
Hidden Markove Models
Dynamic Bayesian Networks
use