Episodic Memory for Virtual Humans and Virtual Humans for Episodic - - PowerPoint PPT Presentation
Episodic Memory for Virtual Humans and Virtual Humans for Episodic Memory Cyril Brom et al. Faculty of Mathematics and Physics, Charles University in Prague, CZ brom@ksvi.mff.cuni.cz Team Jakub Gemrot, Michal Bda, Ond ej Burkert,
Episodic Memory for Virtual Humans and Virtual Humans for Episodic Memory
Cyril Brom et al.
Faculty of Mathematics and Physics, Charles University in Prague, CZ brom@ksvi.mff.cuni.cz
Team
Jakub Gemrot, Michal Bída, Ondřej
Burkert, Klára Pešková, Tomáš Korenko, Jan Vyhnánek, Rudolf Kadlec, Jakub Kotrla, Lucie Kučerová, Lukáš Zemčák
Funding agencies: IS, MSMT
Neuro/psychological perspective
Memory for personally
personally experienced past events (episodes)
Recall of an event in the
spatio-temporal context in which it was experienced
H.M. Tulving (1983) Baddley (2000, 2001) Forensic psychology
Place cells, grid cells,
head
irection cells...
Morris water pool, ...
Types of research
Animal research
spatial cognition episodic-like memory
Casuistries “Laboratory” cognitive psychology
world-list memorising etc.
Forensic psychology Computational modelling
robotic artifacts
State of the art
A couple of computational models
fundamental research vs. engineering
These models:
tend to focus only on a narrow range of
phenomena,
most importantly: space vs. events exceptions are rare
typically oriented on short-term intervals,
do not support representation of complex plausible human complex plausible human
like episodes.
Main methodological proposal
This state of affairs can be improved
using the technology of virtual virtual environments environments inhabited by intelligent intelligent virtual agents virtual agents as a test test-
bed for developing, investigating, and integrating various neuro-/psychologically plausible models of episodic and/or spatial memory.
Complex episodes
Interaction with multiple objects
e.g. cooking
Large spaces
e.g. a city
Long periods of time
e.g. months, years
Main advantages of virtual worlds
Features:
more technically accessible and cheaper then robots low-level (e.g. distances, pixels) vs. high-level (e.g. affordances) ecologically plausible models of real worlds (to some extent) speed-up time
Robots or virtual agents?
trade-off: crippled worlds with “natural” noise vs. large worlds the “robotic risk” : you see the phenomena through the hardware limitations the “virtual agents risk”: you see the phenomena without any limitations
Main Pros:
integration of “single-phenomena” models combination of levels of abstraction large environments longitudinal studies in virtual reality settings with human subjects
Pogamut
Integrated Development Environment
Integrated Development Environment
Development Debugging Support for experimenting
Education
Our episodic memory research
Purpose: Hybrid
agents for serious games neuro-/psychologically plausible methodological: a vehicle for other models
Main focus
complex episodes large spaces long periods of time integration of perception of space and
events
Our episodic memory research
Main issues
Representations of complex episodes,
large spaces, life-time intervals
Plausible forgetting
blending of episodes, false memories perception of time, time as a cue
Context abstraction
Techniques
symbolic connectionist
Our agent architecture
PF TF MF
ENV
Attention Filter Goals Structure Task select ion Conflict resolution mechanism <invoke a subgoal> ACTION <search> <remember episode> <recall>
VERBAL OUTCOME
Emotions
<influence>
DRIVES
ENV
tasks
STM
LTEM LTSM
<remember location of an object>
Example Example – – a virtual shaman a virtual shaman
“I was doing SearchRandom for smokeability because of Smoke. I was doing go from room 1 to room 2 because of SearchRandom. I was doing look in environment because of SearchRandom. I was doing go from room 2 to room 5 because of SearchRandom. I was doing pick up Calumet1 because of Smoke. I was doing Smoke.” Autobiographic
memory with forgetting
5
ays
Dynamic scenario Debriefing
Forgetting – example
(no “emotions”)
Forgetting – example
(with “emotions”)
Positions of objects in large worlds
Searching strategies:
where are the glasses?
Could be at your bedside table. If not, could be next to the TV. If not, scrutinise your study room. If not, scrutinise the whole flat.
concrete abstract