Preservation of Conditioned Behavior Based on
UV Light Sensitivity in Dissected Tail Halves of
Planarians
-
a Proof by DNN
Kensuke
Shimojo
.
1, 2
,
Eiko Shimojo
.
2
,
Reiya Katsuragi.
3
,
Takuya Akashi
.
4
Shinsuke Shimojo
.
2
1
Harvard
-
Westlake
S
chool,
Studio
City
,
CA
2
California Institute of Technology, División of Biology and Biological
Engineering/Computation and Neural Systems, Pasadena, CA
3
Iwate University
, Graduate school of Arts and
Sciences,
Division of Science and
Engineering,
Morioka
, Iwate
,
Japan
4
Iwate University, Faculty of Science and Engineering, Department of Systems
Innovation Engineering
, Morioka, Iwate,
Japan
Abstract
Planarians are aquatic worms with powerful regenerative and
memory retention
abilities
.
This paper examines whether a
dissected
tail
half of a Planarian (Dugesia Dorotocephala) can retain and exhibit a
previously
-
conditioned
response, possibly before the regeneration of the
head and the ganglia.
We
conditioned intact Planarians in a Pavlovian
procedure with an electric
shock
(ES)
as the unconditioned stimulus and
weak
ultraviolet (UV)
light as the conditioned stimulus
. Then,
we dissected
their bodies into halves
, keeping the
dissected
tail halves.
Starting from the
2nd day after dissection, we presented the
same UV light
3 times daily
while video
-
recording the responses. The recorded responses were then
classified by
a
DNN
: a
VGG16
model
was pre
-
trained by ImageNet for
extracting features from images
and additionally trained
with 211
responses to ES and 118 to UV light be
fore conditioning/dissection
to
categorize
planarian
s’
reactions into
“UV
-
induced” or “ES
-
induced”
reactions
. The cross
-
validated accuracy
in categorization
was 83.6%. We
then let
t
he DNN analyz
e
99 recorded responses to UV from 20 individual
conditioned
tail halves
. 96.8 %
of their reactions
were classified as “ES
-
induced” (against 22.0% wrongly classified as “ES
-
induced” for
unconditioned samples under UV), indicating they have shown the
“Conditioned Response” (p<3.06E
-
30). This provides evid
ence that
planarians can conserve and reveal a learned response even without the
head/ganglia, as it takes approximately 7 days for the head/ganglia to
regenerate versus the given 2
-
3 days.
Although similar findings have been
.
CC-BY-NC 4.0 International license
available under a
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
The copyright holder for this preprint
this version posted November 1, 2022.
;
https://doi.org/10.1101/2022.10.30.514395
doi:
bioRxiv preprint
reported repeatedly in the lit
erature, this is the first positive evidence with
automated procedures and DNN classification. The result implies the
presence of a
decentralized
nervous structure outside of its head/ganglia
that allows a tail half to retain memory and execute
motion accordingly
,
despite their cephalization.
Introduction
Planarians
, also known as flatworms,
are
a type of
freshwater
worms
k
nown for abilities of bidirectional regeneration and basic learning such as
habituation
1,2
.
They have a cephalized nervous system,
with
a distinct
head region
that houses
the
ganglia (
the
ir
brain
-
equivalent structure)
3
.
They can also asexually reproduce by “fission” (splitting off a part of its tail),
thus having the tail regenerate a new head and ganglia within a
week
4
.
However, for this
way of reproduction
to be effective, the tail must be able
to survive in nature duri
ng the week of regeneration without its head
, which
houses
many systems that play important roles in survival, such as the
ganglia, the eyespots (visible light detection and aversion), and the auricles
(food particle and touch detection).
Indeed, the full
regeneration of the
ganglia requires
up to
10
days
4
.
As such, the tail half must be equipped
with survival mechanisms that can function without its head, such as having
ultraviolet light (UV) sensitive opsins distributed across the body
. With this,
planarian tail halves can
detect/avoid sunlight to maintain moisture and
stay hidden from predators
3
.
Another significant ability found in the tail is its ability to
hold onto
learned
effects from habituation
it experienced
as a whole planarian
2
.
This is
a powerful
tool for its
survival
as it offers higher familiarity in
its given environment
. However,
it is
a relatively simple form of learning
compared to other learning
types, such as Skinnerian or Pavlovian
conditioning
.
Such regeneration and survival abilities of the species have made it a
perfect model to investigate the possibility of memory retention without the
brain.
Indeed,
there are many classical studies employing
conditioning
5
procedures to indicate such memory retention beyond regeneration.
The
general procedure
and logic
go
as such
: the intact animals are trained to
learn something, eithe
r in the conditioning or the habituation/addiction
procedures
. Then
,
they are dissected
in
to halves, and the tail halves are
tested for memory before the full regeneration of the neural ganglia.
.
CC-BY-NC 4.0 International license
available under a
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
The copyright holder for this preprint
this version posted November 1, 2022.
;
https://doi.org/10.1101/2022.10.30.514395
doi:
bioRxiv preprint
Although there
is
ample classical
evidence
for such memory
retention, these early studies classified the animals’ responses to the
conditioned stimuli subjectively by the eyes, thus potentially vulnerable to
bias and noise, providing no objective evidence. Also, most of
the
classical
studies were
not automated
8
,
9
,
even though
classical
conditioning is known
to
be highly
sensitive to timing, especially between the unconditioned and
the conditioned stimuli. The current study aims to provide more objective
evidence,
employing
a fully automated conditi
oning and testing protocol (by
an
Arduino
microcontroller
) and relying on
a Deep Neural Network (DNN)
to classify
planarian
behavior.
More specifically
,
the
current study
examines whether a
dissected
tail
half of a
p
lanarian (Dugesia
Dorotocephala)
can re
tain
learned
behaviors
from Pavlovian conditioning
before
its
dissection to
exhibit
a previously
conditioned response
before
the
full
regeneration of its ganglia. Pavlovian
conditioning
is beneficial
to
survival but
is a
relatively more complex
and
fundamentally different
form of conditioning;
studies on the cockroach
showed
that Skinnerian
(operant)
conditioning is possible without the
central nervous system, but
it is controversial if
Pavlovian Conditioning
is
possible or
not
without the central nervous system
10,11
.
It
has
been
indicated
that
planarian
tail halves
can
retain effects from Pavlovian
conditioning
and
demonstrate
their
conditioned behavior before
regenerating
their
head/ganglia
2
.
I
f our hypothesis
is correct in
that
planarian tails can
remember
conditioned behavior for
a
better chance of survival and reproduction, we
expect
the
dissected
tail halves of the conditioned planarians to exhibit the
conditioned response
prior to the regeneration of the ganglia (<7 days after
dissection)
.
In this study, we put them under
a
Pavlovian Conditioning
procedure
with the following parameters:
the
condi
tioned stimulus is weak
UV light, the
unconditioned stimulus
is
an electric shock
.,
and the
response
to the electric shock is a sharp contortion reaction.
W
hen
this contortion
response is
triggered
with
only
the conditioned stimulus
(weak UV light
)
, it
would have become the conditioned response
.
Note that the intrinsic
(pre
-
conditioning)
response to this strength of UV light is very little and
qualitatively different (
i.e.,
stationary for the first several seconds, then
slowly start
s
swimming) from that
to the electric shock
. Thus, we can train
the DNN network with the responses
of
whole
-
body individuals
to the
electric shock and the weak UV
light, such that it can
classify a
planarian’s
reaction as either
electric shock
-
like
or UV
-
like
.
Then, we
have it classify
.
CC-BY-NC 4.0 International license
available under a
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
The copyright holder for this preprint
this version posted November 1, 2022.
;
https://doi.org/10.1101/2022.10.30.514395
doi:
bioRxiv preprint
the tail halves’ responses
to the UV light
to see if
the DNN mostly classifies
them
as
responses
to the electric shock
, since t
his would
imply
that the tail
halves showed
the conditioned response and memory of conditioned
behavior
was retained
.
Result
s
After dissecting
planarians
that
had been
previously
trained with Pavlovian
conditioning
, their dissected tails
were able to exhibit their learned behavior
prior to the
regeneration of new ganglia
.
We trained 20 individual planarians
with a Pavlovian conditioning
procedure such that they would start to associate
exposure to
weak UV
with
an electric shock (ES)
. As they were trained, they
began
reacting to
the weak UV light with a contortion response similar
to
that toward
the
ES
.
(See the Methods for more details of the procedures.)
After the conditioning
was successful
, t
hey were dissected, and the tail
halves
, on the
2
nd
and 3
rd
day
s
after dissection
(
after a 1
-
day resting
period
)
, were
exposed to the same weak UV
to see if they exhibited the
conditioned response (
ES
-
like
contortion reaction).
A total of 99
responses
from these 20 tail halves
were recorded on video.
To objectively
determine whether this recorded reaction is the conditioned
response or not, we trained a
deep neural network (DNN
) to classify
the
recorded responses
into “electric shock triggered responses” vs. “UV light
triggered
responses” (
See the Methods section for more details)
.
Th
e
DNN
was
trained
with 211
control
responses to ES and 118 to UV light
with
a
separate, randomly chosen
set of 20
whole
-
body
planarians.
Its
cross
-
validated
classification
accuracy was 83.6%.
Then,
this DNN categorized
the 99 UV
-
stimulated responses from
the
tail
halves recorded on day
s
2 and 3 after the dissection
. 96.8% were classified
as “ES
-
induced
,”
against 22.0% wrongly classified as “ES
-
induced” for
unconditioned samples under UV
, relative to 22.0 % as “ES
-
induced’ vs.
78.0% classified as “UV
-
induced” before the training
(
Fig. 1 shows the
pooled data
from
day
s
2 and 3; see Supplementary Tables 1 an
d 2 online
for original data).
.
CC-BY-NC 4.0 International license
available under a
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
The copyright holder for this preprint
this version posted November 1, 2022.
;
https://doi.org/10.1101/2022.10.30.514395
doi:
bioRxiv preprint
Fig.1
Classification of Responses to UV
% of Planarians’ Responses to UV Classified (as UV or ES) Before and After
Conditioning.
**x
2
=107.61,
P=3.28E
-
25<.0000
Thus, as expected from
the
retention of the
Pavlovian conditioning effect,
the DNN classified a vast majority of the tails’ responses
to UV as those to
ES,
relative to the latent 22% error on the DNN’s false ES categorization in
the
pre
-
conditioning responses
. Statistical probability (p<3.28E
-
25<.00
001)
strongly suggests that the tail halves were able to retain information.
Discussion
One of the biggest advantage
s
of DNN
-
based classification is that it
enables a classification
that
is purely based on objective features instead of
subjective impressions by eye examination. As such, it opens a wide
window to re
-
examine classical
“
findings
” by eyes
in an
i
mal behavior,
neuroscience, and human psychophysics. In the current paper, we
have
applied a DNN
-
based classification algorithm
(VGG16)
to
see
if, after
p
lanarians have been conditioned and then dissected,
their
tail halves can
77.96610169
6.060606061
22.03389831
93.93939394
0
10
20
30
40
50
60
70
80
90
100
Before Conditioning
After Conditioning/dissection (tail halves)
Percent (%)
% of Planarians’ Responses to UV Classified (as UV or ES) Before
and After Conditioning
UV
ES
.
CC-BY-NC 4.0 International license
available under a
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
The copyright holder for this preprint
this version posted November 1, 2022.
;
https://doi.org/10.1101/2022.10.30.514395
doi:
bioRxiv preprint
reveal the learned behavioral responses even before regeneration of the
neural ganglia.
The majority (>80%) o
f UV
-
induced responses on day
s
2 and 3
after
the dissection
(
which was
long
before the 7th
-
10th day for the maturation of
the ganglia) were classified by the DNN as “ES
-
induced,” confirming that
the planarian can conserve and reveal a learned response without the
head/ganglia
. By employing the DNN a
nd fully automated procedures (
by
the Arduino), we provided much more
objective, convincing evidence than
any in the classical
p
lanarian learning literature, essentially free from any
bias and noise
from
eye
examination.
The results were consistent with the
distributed nature of neural
pathways and UV sensors in the animal
3
.
The result suggests the existence of a structure outside its ganglia
that retains learned effects. This is consistent with the latest evidence on
habituation,
but
it also shows a more sophisticated capacity of memory
in
this structure,
as Pavlovian conditioning tends to be more complex than
habituation. The results were also consistent with the distributed nature of
neural pathways and UV sensors
3
.
Further, they were able to execute this function before the
regenerati
on of their new ganglia, given the 1
-
2 day regeneration period
they had before the tails were tested in our study vs. the 7
-
10 days needed
for full regeneration.
This line of investigation is key to understanding the intricate interplay
between the regene
ration mechanism and the nervous system
12
.
Along the
line, it would be very beneficial to compare the biological mechanisms of
regeneration and memory retention in the
p
lanarians with those in partly
regenerative and non
-
regenerative species.
Since the entire body of the planarian is UV
sensitive, it also raises a
more drastic possibility of even learning without the ganglia (establishing
new conditioning without the ganglia). As described earlier, Skinnerian
conditioning is possible without the central nervous system in the
cockroach, bu
t
this
is not confirmed for
Pavlovian conditioning
10,11
.
It may
also be worth seeing if the animal can reveal the learned behavior with the
newly
grown (not directly learned) ganglia.
.
CC-BY-NC 4.0 International license
available under a
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
The copyright holder for this preprint
this version posted November 1, 2022.
;
https://doi.org/10.1101/2022.10.30.514395
doi:
bioRxiv preprint
Procedures
The
overall procedures
can be summarized in the
following three steps:
1.
The selected animals (N=20) underwent the
Pavlovian
conditioning
procedure
, where
the
paired association of a neutral/conditioned
stimulus with an unconditioned aversive stimulus/response was
made
.
W
hen
successful, the conditioned stimulus alone can trigger
the unconditioned response.
Thus here, we employed
an
electric
shoc
k
(ES) as the
U
nconditioned
S
timulus
(US)
and
weak UV light as
the
C
onditioned
S
timulus
(CS).
T
heir
original
,
contortion
-
like
response to the electric shock is
considered the unconditioned response (UR)
.
When
this
contortion
response
is observed
toward
the
weak UV light, or the
CS
,
it would
be considered
the
Conditioned Response (CR)
, a
learned behavior
.
In other words, the tail half showing a
ES
-
like contortion response to
only
weak UV can be considered
evidence for memory retention of the
conditioned behavior
.
2.
The conditioned
planarians are dissected. On the 2
nd
and 3
rd
days
after the
dissection, the tail halves are exposed to the same weak UV
light,
and their responses are recorded
.
3.
A DNN,
pre
-
trained with not
-
yet
-
conditioned, whole
-
body Planarians’
responses
to UV and ES
, then categorizes the dissected tail halves’
responses
to
UV only
as UV
-
like
or ES
-
like
responses
. The majority
of responses categorized as ES
-
like would then be taken as
evidence
for retained memory of conditioned behavior
.
Hardware
Setup:
The setup consists of a
microcontroller
(an Arduino),
an overhanging
projector camera,
a UV
flash
light weakened with UV filters, a
3D printed
stand for the flashlight
,
a small aquarium
with metal sheets
(electrodes)
attached to
the sides,
and a computer
(See Supplementary Methods
online
for more details). The planarians go
into
the
aquarium beneath the UV
flashlight and the camera,
such that
the microcontroller controls the
planarians’ exposure to
UV and ES
and
starts
/stop
s
recording any of the
planarians’ activities on the camera.
DNN
Setup and
T
raining
We developed a video classification model for two planarian behaviors
: the
response
s
to ultraviolet light and electric shock. Many studies of video
.
CC-BY-NC 4.0 International license
available under a
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
The copyright holder for this preprint
this version posted November 1, 2022.
;
https://doi.org/10.1101/2022.10.30.514395
doi:
bioRxiv preprint
classification focused on
a
two
-
network architecture
that extracts
spatial
feature
s
and temporal
features
from
the
video
13
.
The two
-
stream models
yield
high
-
performance
results
in vid
eo classification. However, a large
number of parameters are required in the model because the model fuses
many layers of
the
neural network to extract spatiotemporal features from
video, and
the
issue of
overfitting
can arise with
a
small dataset [2
]
14
.
Therefore, we developed
a
simple video classification model
that
expand
ed
the scope of convolutional neural network (CNN) based image
classification without complicated networks of the video classification
model.
O
ur model is developed by
expanding
an
image classification model with a
CNN architecture such as VGG
15
and
ResNet
16
.
In our method, each frame
of video is classified by the image classification model pre
-
trained by
ImageNet
17
, and then, a mode value is ca
lculated as a result of video
classification with the image classification results. In our case, VGG16 is
adopted as
the backbone for
image classification because it
has
the best
performance.
Moreover, w
e needed to
modify
our original
videos of
planarian
s recorded in our experimental environment and create a
new
dataset of
video
s
as
training data
for
the following
2 reasons:
the
model
’s
training is affected by
the
color difference between
videos for planarians
under
UV
vs
.
those under ES
, and the
re are multiple planarians present in
each video
when the DNN needs to analyze
one
planarian
’s
motions
at a
time
.
To account for
these issues
, videos zoomed in on each planarian are
extracted from a given video and are binarized (black/white).
In training our
video classification model,
we used
one frame per second
of
each
one
-
planarian
video
, and
all frames were input for analyzing the tails’
reactions
.
Experimental
Procedure steps:
1.
C
onditioning: repeated 180 conditioning trials
(trial sequence shown
below)
on a batch of 20 planarians/
day
across 4 days
:
.
CC-BY-NC 4.0 International license
available under a
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
The copyright holder for this preprint
this version posted November 1, 2022.
;
https://doi.org/10.1101/2022.10.30.514395
doi:
bioRxiv preprint
The following was done a total of 3 times
for
each of the 4 days at 8
am, 4 pm, and 12
am:
30 trials with 1
-
minute breaks in between, a
10
-
minute break
,
and another 30 trials with 1
-
minute breaks in
between
.
We added the 1
-
minute breaks because
,
w
ithout the
m
, the planarians
became fatigued from the repeated electric shocks
.
2.
E
xpose
d
the 20
planarians
under UV only
and
confirm
ed
they have
beg
u
n showing contraction
response
to weak UV alone
(the CR)
: this
indicate
s
Pavlovian conditioning
is
established
.
3.
Manually d
issect
ed
the
conditioned
whole
-
body animal
s with
a
scalpel,
as close to the middle of the planaria
n
as possible
.
The tail
halves
were
given 1
full
day to regenerate motor functions
.
4.
Regenerating tails
were
exposed to CS (weak UV) 3 times/day on
the
2nd and 3rd
day
s
after dissection,
with 3
-
minute breaks between
each exposure.
Their responses
w
ere
recorded
.
Here, UV is turned on for 3 seconds, but recording time should last
for an additional 7 seconds (
a
total
of
10 seconds
of footage
), and a
dim UV
-
free light should be on for the camera to
capture the tail
halves’ movements after UV exposure
.
5.
DNN Training:
Recorded >100 control responses of
a different group
of
20
random
ly chosen
whole
-
body
planarians to CS and US each
.
When recording responses to the US/electric shock,
we
use
d
a lamp
with a UV filter on or a UV
-
free lamp to light up the container, just
enough for the planarians to be visible
.
6.
Recorded responses were fed to the DNN
for training, such that it can
categorize a given
response
as
being
more
akin to
the UV response
o
r the ES response
.
7.
Recordings of tail halves’ reactions from step 4
were
given to DNN for
categorization
Data Availability
Raw video data and DNN categorization results:
https://drive.google.com/drive/folders/1Y47WyJePs8N9wF3ZW73_PGn03s
J5CLck?usp=share_link
.
CC-BY-NC 4.0 International license
available under a
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
The copyright holder for this preprint
this version posted November 1, 2022.
;
https://doi.org/10.1101/2022.10.30.514395
doi:
bioRxiv preprint
References
1.
Sánchez Alvarado A.
Planarian regeneration: its end is its beginning.
Cell
. 2006 Jan
27;124(2):241
-
5. doi: 10.1016/j.cell.2006.01.012. PMID: 16439195.
2.
Mc
Connell
JV, J
acobson
AL, K
imble
DP.
The effects of regeneration upon retention
of a conditioned response in the planarian. J Comp Physiol Psychol. 1959 Feb;52(1):1
-
5. doi: 10.1037/h0048028. PMID: 13641455.
https://psycnet.apa.org/doi
/10.1037/h0048028
3.
Shettigar
N, Joshi A, Dalmeida R, Gopalkrishna R, Chakravarthy A, Patnaik S,
Mathew M, Palakodeti D, Gulyani
A. Hierarchies in light sensing and dynamic
interactions between ocular and extraocular sensory networks in a flatworm.
Sci Adv
.
2017 Jul 28;3(7):e1603025. doi: 10.1126/sciadv.1603025. PMID: 28782018; PMCID:
PMC5533540.
4. Malinowski PT, Cochet
-
Escartin
O, Kaj KJ, Ronan E, Groisman A, Diamond PH,
Collins ES. Mechanics dictate where and how freshwater planarians fission.
Proc Natl
Acad Sci
U S A. 2017 Oct 10;114(41):10888
-
10893. doi: 10.1073/pnas.1700762114.
Epub 2017 Sep 25. PMID: 28973880; PMCID: PMC5642
676.
5
.
Rescorla
RA.
Behavioral studies of Pavlovian conditioning.
Annu Rev Neurosci
.
1988;11:329
-
52. doi: 10.1146/annurev.ne.11.030188.001553. PMID: 3284445.
6.
Shomrat
, Tal and Michael Levin. “An automated training paradigm reveals long
-
term
memory in planarians and its persistence through head regeneration.”
Journal of
Experimental Biology
216
(2013): 3799
-
3810.
7.
Samuel, K., Suviseshamuthu
, E.S. and Fichera, M.E. (2021) Addiction
-
Related
Memory Transfer and Retention in Planaria.
BioRxiv
, doi:
https://doi.org/10.1101/2021.09.12.459965.
8.
Hicks
,
C.,
Sorocco
, D. and Levin, M. (2006) Automated analysis of behavior: A
computer
-
controlled system fo
r drug screening and the investigation of learning.
Journal
of Neurobiology
, 66
-
9, 977
-
990.
https://doi.org/10.1002/
neu.20290
.
9.
Blackiston
, D., Shomrat, T., Cindy L. Nicolas, C.L., Christopher Granata, C. and
Levin, M. (2010) A Second
-
Generation Device for Auto
mated Training and Quantitative
Behavior Analyses of Molecularly
-
Tractable Model Organisms.
PlosOne
, 6,(1):
https://doi.org/10.1371/journal.pone.0014370.
10. Chen
, W., Aranda, L., & Luco, J. (1970). Learning and long
-
and short
-
term memory
in cockroaches.
Animal Behaviour
,
18
, 725
-
732. https://doi.org/10.1016/0003
-
3472(70)90018
-
7
.
CC-BY-NC 4.0 International license
available under a
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
The copyright holder for this preprint
this version posted November 1, 2022.
;
https://doi.org/10.1101/2022.10.30.514395
doi:
bioRxiv preprint
11.
MacMillan, D. L. (1973). A classical
conditioning paradigm for the study of learning
in a ganglion of the cockroach (Periplaneta Americana).
Animal Behaviour
,
21
(3), 492
-
500. https://doi.org/10.1016/S0003
-
3472(73)80009
-
0
12. BLACKISTON DJ, SHOMRAT T, LEVIN M (2015). The stability of memories
during
brain remodeling: A perspective.
Commun Integr Biol
8
:
e1073424.
https://doi.org/10.1101/2021.09.12.459965
13.
Simonyan, K., & Zisserman, A. (2014). Two
-
stream convolutional networks for
action recognition in videos.
Advances in neural information processing systems
,
27
.
14.
Jing, L., Parag, T., Wu, Z., Tian, Y., & Wang, H. (2021). Videossl: Semi
-
supervised
learning for video classification. In
Proceedings of the IEEE/CVF Winter Conference on
Applications of Computer Vision
(pp. 1110
-
1119).
15
.
Simonyan, K., & Zisserman, A.
(2014). Very deep convolutional networks for large
-
scale image recognition.
arXiv preprint arXiv:1409.1556
.
16.
He, K., Zhang, X., Ren, S., & Sun, J. (2016). Deep residual learning for image
recognition. In
Proceedings of the IEEE conference on computer vision and pattern
recognition
(pp. 770
-
778).
17.
Russakovsky
, O., Deng, J., Su, H., Krause, J., Satheesh, S., Ma
, S., ... & Fei
-
Fei, L.
(2015). Imagenet large scale visual recognition challenge.
International journal of
computer vision
,
115
(3), 211
-
252.
Acknowledgments
This project was s
upported by
the
Masason
Foundation (Japan) and Chen
Institute for
Neuroscience at Caltech
.
Author Contributions
K.S. and S.S.
contributed to conceiving and designing
the experiment
.
E.S.
and K.S.
created the setup and ran the experiment
. R.K. analyzed the data
by
training and running
the DNN
, and T.A. supervised this data analysis.
K.S., E.S., S.S., and R.K.
contributed to the preparation of
the manuscript.
The author
s
declare no competing interests.
.
CC-BY-NC 4.0 International license
available under a
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
The copyright holder for this preprint
this version posted November 1, 2022.
;
https://doi.org/10.1101/2022.10.30.514395
doi:
bioRxiv preprint