Modular Monolith - a Practical Alternative to Microservices @ Devoxx UK 2024
Recognition at end of Year 1
1. Cogni&ve
contents
Franco
Bagnoli
and
Andrea
Guazzini
University
of
Florence
RECOGNITION
year
1
review
1
10th
November
2011
2. MoGvaGon
and
Background
• Pervasive
compuGng
devices
– Mobility,
Portability
– Wireless
connecGvity
– Sensors
– MulGmedia
capabiliGes
• Cheap
and
portable
hardware
with
processing,
storage
and
communicaGon
capability
– FacilitaGng
new
ways
to
provide
and
share
content
– CreaGng
more
and
more
diverse
content
RECOGNITION
year
1
review
2
10th
November
2011
3. Content-‐centric
approach
• Content
is
generated
everywhere
– IntegraGon
human
acGvity
and
mobility
– Greater
user
parGcipaGon
(e.g.,
web
2.0)
• Content
is
diverse
– Pictures,
data
from
sensors,
news,
caching
from
the
Internet,
messages
– Unleashed
from
tradiGonal
Internet
• Content
can
be
shared
&
forwarded
– Short
range
wireless
technology
for
forwarding
and
sharing
– Awareness
of
locaGon
and
context
–
a
spaGal
context
RECOGNITION
year
1
review
3
10th
November
2011
4. RECOGNITION
mission
• Seeking
to
capture
the
behavioural
characterisGcs
of
the
most
intelligent
living
species,
namely
human
beings
• Fundamental
approaches
to
cogniGon
that
are
grounded
in
the
organ
responsible
for
the
most
sophisGcated
autonomic
behaviour
–
the
brain…
• PotenGally
begin
to
represent
the
needs
and
characterisGcs
of
the
individual
users
inside
the
network
itself
and
inside
content.
• Include
fundamental
characterisGcs
of
human
cogniGve
behaviour,
such
as
the
ability
to
infer,
believe,
understand,
and
assert
relevance,
interact
and
respond
in
the
face
of
massive
amounts
of
informa&on.
RECOGNITION
year
1
review
4
10th
November
2011
5. The
Approach…
• Developing
models
of
cogni&ve
behaviour
from
psychology
that
are
transferable
to
the
ICT
domain;
– Key
psychological
principles
to
facilitate
self-‐awareness
• ExploiGng
models
of
cogniGve
behaviour
for
a
content-‐centric
Internet
– self-‐awareness
can
provide
new
levels
of
cogniGve
behaviour
to
enhance
content
acquisiGon.
RECOGNITION
year
1
review
5
10th
November
2011
6. Human
Awareness
Behaviours
• Approach:
Capture
&
exploit
key
behaviours
of
the
most
intelligent
living
species
– Human
capability
is
phenomenal
in
navigaGng
complex
&
diverse
sGmuli
– Filter
&
suppress
informaGon
in
“noisy”
situaGons
with
ambient
sGmuli
– Extract
knowledge
in
presence
of
uncertainty
– Exercise
rapid
value
judgment
for
prioriGsaGon
– Engage
a
social
context
and
mulG-‐scale
learning
RECOGNITION
year
1
review
6
10th
November
2011
7. Project
ObjecGves
1. To
iden&fy
and
engage
a
robust
psychological
basis
for
self-‐
awareness
in
ICT.
– This
will
involve
engaging
cogniGve-‐based
processes
from
the
human
brain
that
enable
understanding,
inference
and
relevance
to
be
established
while
suppressing
irrelevant
informaGon
in
the
context
of
massive
scale
and
heterogeneity.
2. To
exploit
the
psychological
basis
for
self-‐awareness
in
a
content
centric
Internet.
• This
will
involve
engaging
the
spaGal
dimension,
interacGons
and
intelligent
processes
that
reflect
cogniGve
behavioural
heurisGcs
to
provide
content
and
knowledge
flow
to
other
parGcipants
and
network
components.
RECOGNITION
year
1
review
7
10th
November
2011
8. RECOGNITION
approach
CogniGve
psychological
basis
For
awareness
and
understanding
Defining
key
principles
for
exploitaGon
by
technology
components
Embedding
these
principles
for
self-‐awareness
in
autonomic
content
acquisiGon
in
pervasive
environments
PotenGal
change
in
behaviour
due
to
self–awareness
in
ICT
RECOGNITION
year
1
review
8
10th
November
2011
9. Minimal
self-‐awareness
cogniGve
agent
Self-‐awareness
can
be
classified
on
the
basis
of
three
criteria:
Gmescales,
cogniGve
costs
and
evoluGonary
features.
Timescales
-‐(Reac&on
&mes)
• Unconscious
Knowledge
(PercepGon
and
Pre-‐ahenGve
acGvaGons)-‐>
Fast
(<.500
ms)
• Conscious
knowledge
(reasoning)
-‐>
medium
(from
seconds
to
hours)
• Learning/development
-‐>
slow
(from
minutes
to
month)
Cost
(Cogni&ve
Economy
Principle
-‐
Amount
of
neural
ac&va&on)
• Unconscious
knowledge
-‐>
light
(small
and
local
acGvaGons)
• Conscious
knowledge
-‐>
heavy
(large
and
diffused
acGvaGons)
•
Learning/development
-‐>
very
heavy
(diffused
acGvaGons)
Evolu&onary
features
(Cogni&ve
development)
• Unconscious
knowledge
-‐>
criGcal
period
and
Hebbian
learning
only
(ACTr)
• Conscious
knowledge
-‐>
trial
and
error,
observaGon/imitaGon
and
inducGon.
• Learning/development
-‐>
fixed
hard
wired
rules.
RECOGNITION
year
1
review
9
10th
November
2011
10. External
Tri-‐parGte
model
Data
Reac&on
&me
Module I
Unconscious knowledge
Flexibility
perceptive and attentive processes
Relevance Heuristic
Cogni&ve
costs
Module II
Reasoning
Goal Heuristic
Recognition Heuristic
Solve Heuristic
Module III
Learning
Behavior
Evaluation Heuristic
RECOGNITION
year
1
review
10
10th
November
2011
11. An
applicaGon:
cogniGve
audio
stream
• Many
people
live
inside
an
audio
sphere:
portable
music,
radio,
ambient
music..
• Music
streams
(playlists)
can
be
assembled
manually,
or
by
means
of
automaGc
systems:
– Randomly
(shuffling)
– Based
on
similariGes
among
clips
(Pandora)
– SimilariGes
among
users
(like
amazon)
– Based
on
mood
(moodagent)
– SubscripGon
(podcasts)
– DelegaGon
(radio)
– Direct
suggesGon
(friends)
RECOGNITION
year
1
review
11
10th
November
2011
12. The
“radio”
structure
• The
delegaGon
mode
(i.e.,
classical
radio)
allows
the
discovering
of
new
elements
(informaGon,
entertainment,
new
genres)
• Favours
social
interacGon
(commenGng,
voGng)
and
parGcipaGon
• But
is
hard
to
be
personalized
RECOGNITION
year
1
review
12
10th
November
2011
13. CogniGve
playlist
• Context:
locaGon,
Gme,
weekday,
status
(e.g.,
work,
commuGng,
home..),
network
access/bandwidth,
mood
(user
input),
memory
(played
clips),
feedback
(user
input),
user
profile
• External
data:
sugges&ons
from
a
server,
based
on
user
pahern
similariGes,
clip
similariGes,
user
choices,
direct
suggesGons
from
social
networks/friends
RECOGNITION
year
1
review
13
10th
November
2011
14. SuggesGons
• SuggesGons
contains
the
descripGon
of
the
resource
and
its
availability
(downloadable,
local,
stream,
permission,
cost),
clip
characterisGcs
that
can
be
used
for
context
matching.
• They
originate
the
actual
playlist
according
with
their
score,
assigned
by
methods
(schemes).
• A
dynamical
score
is
assigned
to
suggesGons
by
schemes
(actually,
each
scheme
proposes
a
score).
The
score
is
recalculated
dynamically
since
the
context
and
the
schemes
may
vary.
RECOGNITION
year
1
review
14
10th
November
2011
15. From
suggesGons
to
playlist
• The
goal
is
that
of
building
a
dynamical
playlist
based
by
the
match
(score)
between
suggesGons
and
the
context.
• The
matching
is
performed
by
methods
(schemes)
that
compete/
collaborate
for
assigning
scores
to
suggesGons.
For
instance,
a
method
may
propose
random
scores
(shuffling),
simply
avoiding
repeGGons,
another
may
propose
scores
based
on
status
and
clip
genre.
• Schemes
themselves
have
a
score,
assigned
to
heurisGcs
(meta-‐
schemes),
according
to
user
feedback
(for
instance
clip
skipping,
voGng,
suggesGons).
RECOGNITION
year
1
review
15
10th
November
2011
16. HeurisGcs
• HeurisGcs
are
similar
to
schemes,
and
assign
a
score
to
schemes,
based
on
feedbacks,
performances
of
schemes,
collisions.
• For
instance,
it
may
happen
that
no
schemes
proposes
a
sufficiently
high
score
to
any
suggesGon
in
a
given
context
(this
is
reported
to
the
server),
then
heurisGcs
may
decide
to
import
other
schemes
from
the
server
• It
may
happen
also
that
a
scheme
systemaGcally
proposes
scores
that
are
different
from
others,
or
finally
that
the
clips
selected
by
a
method
receives
negaGve
feedbacks.
The
method
can
be
purged
by
the
pool.
RECOGNITION
year
1
review
16
10th
November
2011
17. The
compeGGve
environment
• HeurisGcs
try
to
maintain
an
assorted
pool
of
schemes
that
cooperates
(proposing
scores
that
are
not
systemaGcally
in
conflict)
and
that
do
not
receive
negaGve
feedbacks.
• The
scores
are
used
to
instanGate
suggesGons
into
a
short
playlist
(since
context
changes),
and
possibly
also
to
build
a
tree
anGcipaGng
context
changes
(for
instance,
switching
from
commuGng
to
work)
• The
feedback
(for
instance
that
a
clip
has
been
listened
or
skipped
or
that
a
suggesGon
is
never
promoted
to
playlist)
is
reported
to
the
server,
together
withe
direct
suggesGons
to
friends.
RECOGNITION
year
1
review
17
10th
November
2011
18. The
server
architecture
• The
server
is
essenGally
a
database
of
user
profiles
and
clip
choices
• From
the
overlap
among
user
profiles
(clip
choices,
messages,
social
informaGon)
one
obtains
the
affinity
among
users,
that
can
be
used
to
infer
suggesGons
based
on
heurisGcs
(weighted,
take
the
best,
etc.)
• It
may
use
also
databases
of
clip
similariGes
like
pandora
• Collects
direct
suggesGons
RECOGNITION
year
1
review
18
10th
November
2011
19. Conclusions
• Three-‐level
cogniGve
system
(server/suggesGons,
schemes,
heurisGcs)
• Related
to
Hypermusic
(context-‐based,
user
input)
• Ecosystem-‐like,
compeGGon/cooperaGon
• Decentralized,
adapGve,
pervasive
• Can
be
exported
to
other
scenarios
(e.g.,
learning
objects).
RECOGNITION
year
1
review
19
10th
November
2011