in preparation for this talk I actually
spent a lot of time reflecting on Joe

Mann's wonderful talk last week as
you'll see we actually have a lot of

complementary research interests and
something that's sort of driven home for

me last night is I think some of the
types of memories that are really

salient to me are when you're doing a
final look over of your slides and your

cat hops up on to your keyboard proceeds
to delete most of those slides and then

knocks the laptop off onto the floor but
it's okay the laptop still works I

undeleted the slides things are all good
she's she's sweet she means well but you

know I hope that I can take that
experience and maybe think about how to

avoid that problem in the future so
maybe I will not work on the counter by

her food bowl that might be a smart
choice and actually we humans are pretty

good at juggling the sensory information
and our goals and knowledge such as what

your cat does when you prop up next to
her food bowl in order to traverse our

world and flexibly interact with our
environment and our memories clay maybe

they really play a key part in this
because memories they don't only just

define who we are as individuals so for
example I am an academic and I'm excited

and proud to be an academic here at
Georgia Tech but we also use our

memories in order to guide decision
making and to plan for the future and so

you know when our memories fail well we
might retread past mistakes like well

truthfully I probably will have this
problem again with my cat we might also

find ourselves hopefully I mean
hopelessly lost when we're trying to

find her way over from Emory to campus
if you decided to come and see this talk

locally and we might find ourselves
failing to plan an important event and

all of these are actually really big
problems they can get us in a lot of hot

water and so because of that a major
focus of our lab is to try and

understand the complex Meccan
that support long-term memory and enable

us to avoid those types of errors and
one way that we go about studying this

in our lab is by studying spatial
navigation because that is a really

naturalistic scenario where in our daily
lives we constantly drawing on our

memories in order to make decisions
about where to go next and we might even

think about stops that we might make
along the way like oh I need to pick up

groceries before I go home and so we use
memories to potentially decide between

different routes and maybe sometimes we
would deliberately take a more

circuitous route if we have some kind of
sub goal we want to achieve and we'd so

we developed these types of tasks and
here's an example of what they can look

like they can be quite immersive these
virtual navigation tasks that we give to

people and we combine these with fMRI
and this allows us to test predictions

about how activity in different brain
areas relates to different aspects of

maybe a complex multi-step behavior that
we might actually engage in often in our

daily lives and take for granted so I've
talked about some of this stuff a bit

before at Georgia Tech I'm gonna try and
go through the background for the

research that we do fairly quickly with
the goal of telling you about some of

the cool things I've done in the past
but also highlighting some of the neat

research we're doing now in the lab so
I'll give some brief background and then

hopefully jump right in and and and get
through some interesting work where

we've tried to identify neural circuitry
that enables us to learn and remember

different experiences that maybe take
place in the same location in an

environment and understand the circuitry
that allows us to remember those

experiences without some sort of
catastrophic interference between those

different memories and I'll talk a
little bit more about what that means if

that's unfamiliar terminology
we'll also highlight some data from our

lab and my postdoc where we're looking
at the mechanism

that allow us to use long term memory in
order to guide planning and sort of goal

directed behavior basically how do we
think about the future what does that

even mean how do we generate a
simulation in our minds and then I'm

going to end the talk with some newer
stuff so if you've seen me talk before

this will be more interesting for some
of you I'll show some data highlighting

the impacts of stress on the neural
mechanisms of memory and really what the

ramifications are of those impacts for
the types of behaviors that we engage in

and then hopefully I'll have a good bit
of time both at the end but also along

the way to highlight some of the current
and future research directions that we

have going on in the lab okay all right
so as a brief background especially

those of you who already study memory or
you're intimately aware of this but the

the the neural circuitry that supports
long-term memory is incredibly complex

and honestly dr. Mann's talk was a
perfect segue into this he spent a lot

of time talking about the sub circuitry
of the hippocampus and it's mnemonic

functions this is also a major focus for
us and a lot of the data I'll talk about

today are in the human hippocampus maybe
what I'll just say here is I want to

highlight that the functions of the
hippocampus are intimately related to an

in part derived from their interactions
with a broader cortical circuitry and

that includes the medial temporal lobe
and parietal cortex and the prefrontal

cortex and I'll highlight some really
cool data today showing interactions

between the hippocampus and the
prefrontal cortex and in the context of

translating memory into goal directed
action and planning okay so one

anchoring point for studying the
relationship between long-term memory

and goal directed behavior is really the
brain's ability to sequence information

so when we retrieve some sort of
sequential

either it's an episodic memory as I'm
sure you often engage in or having

conversations with friends and family
may be my recounting of the cat last

night is a perfect example but also
retrieving sequential memory can apply

to retrieving a navigational route when
we retrieve these sorts of memories the

populations of neurons that carry
information about or represent the

features of those memories can fire in
sequential order and that sequential

firing actually allows us to either
remember or even generate predictions

about what will happen next for a given
action that we take and this is a major

focus for the types of navigation tasks
that we designed so an example here

that's salient to me is when I need to
learn the route from my office over in

Coon to navigate down here to EB B and
give this talk my memory system forms

sequential associations between the
landmarks the objects the scenes the

types of information that fall along
that route and I can later retrieve that

information either to communicate what
that route entails to someone else if

I'm giving them directions or I can
retrieve that information sort of online

in order to get here for this talk and
we've known for a long time now

that the hippocampus is really important
for retrieving this type of sequential

information so an example from the late
great Howard I can bombs work here and

he's demonstrated his lab demonstrated
that lesions to the hippocampus can

really have profound impacts on the
ability of a rodent to retrieve a

sequence of odors particularly when
there's interference from another

representation and there's that term
again and I'll talk more about that in

just a second in parallel work in humans
we tried to mimic these types of studies

and and linked those mechanism
to human memory function and so using

fMRI we've shown that similar non
spatial tasks the hippocampus is

important for retrieving sequences of
stimuli and it's actually more important

for retrieving those sequences than
retrieving associated stimuli that don't

have any sort of temporal structure to
them and that's actually really

interesting finding to my mind so what
do I mean about interference though why

is this important to this story well in
real life

many of our memories actually share a
lot of features with each other right so

if you're trying to remember Joe Mann's
talk from last week it took split took

place in the same room here you're
reading lunch you might even been

sitting in the same chair how do you
discriminate that experience from this

experience well this happens a lot in
the context of spatial navigation as

well so many of the routes that we
follow actually cross paths with other

routes so I've got my learned route here
to EB B from my office but I also

probably more often than I should take a
little hike out from my office and go

down to the Student Center and get
myself some Blue donkey coffee and what

I've done here is highlighted the
overlap between those routes in yellow

and this is really a major problem for
our memory system because when I get to

this sort of location the visual cues in
front of me are actually associated with

two really strong sets of associations
and I can't just sort of blindly follow

along whatever feels familiar to me
because actually there are two different

things that feel very familiar to me so
my ability to navigate here today and

not wander off to Blue donkey like a
daft fellow presumably depends on the

hippocampus some of our prior work
suggests that's the case and this is

putative Lee because the hippocampus
helps us to form discriminable memory

traces even when those memory traces
have shared information between them

okay so we can actually retrieve
different sequences through a same set

of information say these these common
spatial locations okay so why is the

hippocampus theorized to be important
for this well one idea is that some of

the sub circuitry in the hippocampus is
believed to support a process known as

pattern separation and what this means
is that neural inputs to the hippocampus

then might overlap a lot so say we have
some cortical perceptual input to the

hippocampus
well between those two different routes

those inputs overlap a lot because there
is a lot of shared perceptual

information but some of the sub
circuitry in the hippocampus appears to

be specialized in a way where it can
actually recode this input into a more

authority set of representations and so
this has been a focus of some of my

earlier work it's not an easy thing to
study in humans although as I'll tell

you in a moment we have some new tricks
up our sleeve where we're going to try

and get after this more specifically one
of the ways that we tried to test

whether that theory is correct in humans
is is by developing these virtual

navigation tasks where we present
interference between spatial memories

and we test the degree to which
hippocampal function predicts your

ability to learn different associations
for the same location and an environment

so what we did here was we created a
bunch of virtual navigational routes

this is an example of one and they have
these rich but but funny cues in them

this is a television and that's actually
me on the beach I'm famous so anyway we

create a bunch of virtual roots here and
participants learn to navigate these

routes from a first-person perspective
using a button box and they navigate

these routes over and over and over
again and they're made up of hallways in

intersections between hallways and at
each intersect

they they gradually learn to make the
correct turn for the roof based on

feedback we do all of this outside the
scanner and we do this to create really

strong memory traces say the memory
trace for the route from my office to

here and then the next day we bring them
in to the mr environment and we give

them an fMRI task and in that task we
have them continue to navigate these

highly familiar routes but we also ask
them to learn some new routes and what's

critical is that some of these new
routes actually traverse locations that

are familiar from those previously
learned paths so here's my blue donkey

or while okay Mike there's my my e bebe
route and now they're learning the blue

donkey route and this forces them to
confront that sort of naturalistic

scenario that I set up for you and so by
comparing the relationship with learning

in the hippocampus for this overlapping
context with non overlapping scenarios

where there is no ambiguity there is no
interference between memories we can try

and identify whether it is in fact the
case that hippocampal activity seems to

support seem to play a special role in
learning these different associations

for the same location any questions
about that task okay

so what's interesting is if you look at
participants performance in this task

people are smart and they're actually
pretty good at learning to navigate

these overlapping locations and these
non overlapping locations so this is an

example of the group learning curve for
one of these overlapping decision points

here and what you'll see is that people
learn to navigate an overlapping

decision point with similar efficiency
to learning to navigate a new distinct

non-overlapping route but even though
we're sort of behaviorally equally good

at doing this if we do a targeted
analysis of HIPAA

sample activity what we find is that
actually not only does hippocampal

activity predict the learning rate
within each individual for these

overlapping roots but it actually is a
significantly stronger relationship with

overlapping root learning rates the non
overlapping root learning rates and that

suggests to us that something about the
hippocampal computations is more closely

coupled to our ability to learn this
type of root in green than this type of

root and orange even though we're able
to do both problems with similar

efficiency and it's worth noting these
findings actually converge with a non

spatial version of this task something
it was sort of a human analog to the

rodent oder tasks that I showed you from
the icon bomb group they found the same

result so this is kind of a nice
convergence there across modalities so

this isn't just a story about space this
was kind of this was a really exciting

finding for me at the time and it was an
important component of my actually my

graduate work time flies but there are
actually two explanations probably more

but they're two primary explanations you
might come up with for this one is a

pattern separation story and that is
that given a mismatch between our

existing knowledge of what to do at one
of those intersections and our current

new goal pattern separation could help
us form a new discriminable memory trace

for that same location such that it
directly supports our ability to

retrieve one or the other memory on the
fly as needed but that coupling between

hippocampal activity and overlapping
root learning could also be attributed

to maybe some sort of integration story
so maybe we learn non-overlapping routes

a certain way there pretty easy
maybe we kind of jump right in to try

and habitual eyes those sorts of routes
but when we're learning to routes

through the same environment maybe the
fact that we recognize it's the same

environment
encourages us to bind those roots

together and build a cognitive map of
that space and so maybe that's why

hippocampal activity actually predicts
the learning rate for overlapping more

than non overlapping roots and with that
older work we couldn't really directly

address that but it's something we're
going after now in our lab so we've

developed a new version of this paradigm
where we can take advantage of some more

recent analytic techniques to try and
analyze the representational content

that's coded within patterns of fMRI
activity and I'll talk about that a

little more later in the talk but the
general idea is we can try and decode

what sort of mnemonic information is
active at a given moment from the

pattern of fMRI activity and what's
really cool for this type of research

question is we can track whether the
similarity between representations for

different memories changes over the
course of learning so we could test now

whether these representations grow
together and a sort of integration story

as we learn the environments better
better or whether they start to pull

apart and become more distinct memories
we can also look for where that happens

in the brain it doesn't have to be a
purely hippocampal story so I'm very

excited about that work that we have
ongoing and I also want to highlight

that not all of the choices in these
environments are created equal if you

read papers on spatial navigation a lot
of root navigation tasks are discussed

as if a root is a unitary construct but
truthfully something very different

could happen when two routes converge
with one another and when they diverge

from one another and that's because when
two routes converge the participant

actually winds up heading in the same
direction and if the participant has a

good spatial map of the environment they
might become aware of that when two

routes diverge they're actually headed
in a different direction so if you're

learning these routes in a more
spatially oriented way this might be an

easy decision for you because you're
headed in the same direction regardless

so what route you're following this is a
difficult decision because you're

actually choosing between two options
but if you're learning the routes in a

more response based way like you want to
learn hmm I turn left at this landmark

well this is a situation where you make
a different response at the same

location and this is a situation where
you make a different response in the

same location so this could be slightly
different things depending on who you

are as an individual and it turns out
that is the case people differ in the

degree to which they can navigate these
two locations with the same efficiency

and that's very interesting to us
actually the reaction time curves are

even more revealing than the accuracy
curves that I have here but so we have

some ongoing research where we're trying
to pin down why are people different at

this and what's different about
someone's mnemonic structure if they are

treating these as very different choice
points in the environment versus

treating them as quite similar okay so
that's some of our old work and I think

it's a nice segue into some of the
things we're doing now in the lab but

it's also important to note that that
research is really dealing with ongoing

behavior so how do we learn to make
different responses to the same position

in an environment and avoid getting
confused and making the wrong choice

that sort of thing but we don't just use
our memory to react to our environment

and it's not just a retrospective
phenomenon so as I mentioned at the

beginning we also use our memories to
plan and to think about future choices

that we want to make so this is a
particularly important function that

might also be attributed to the
hippocampus and it's important to think

about this because when you consider
diseases like Alzheimer's that are

affecting the long-term memory system
they might not just impact our ability

to discriminate different memories from
one another but they could have a really

profound impact on our ability to
fashion

we plan for the future and decide on
optimal choices that aren't immediately

in front of us it's a really
catastrophic consideration so this is a

type of big-picture problem that we're
worried about and it's another direction

that we're taking our research so we're
trying to understand what sort of

mechanisms enable us to mentally
simulate future experiences and how

those mechanisms might relate to our
ability to retrieve different goal

relevant representations on the fly and
so I'll get to give you a little bit of

background on how the hippocampus could
support prospective thinking and then

I'll tell you about two studies that are
really closely related that address this

question so think about what it means to
plan a navigational route I tell you go

to EB B this morning there's going to be
a really hopefully interesting talk you

need to think about the long term goal
where it is in the environment but you

also need to think about the route
you're going to take to get there what

sort of critical turns you're going to
face along the way what sort of

landmarks should you be looking for and
so a major focus of recent research in

this area has actually been inspired by
again really exciting work that's been

done in rodents it's been found that
hippocampal neurons that code for

locations place cells and they are here
illustrated by these different colored

dots not only can they fire depending on
what sort of goal the rodent has but

they I can actually fire sequentially
ahead of the animal from its current

position towards its long-term goals so
we see the sort of prospective replay

from the memory system in the
hippocampus that's governed by current

goals in the environment and so when the
rodent in this task when the rodents

goals change so too does the scope of
the forward sweep of hippocampal

activity now in parallel work in humans
prior work that's tried to look at

prospective
thought has highlighted that this same

network that I laid out for you in the
beginning hippocampus connected to MTL

and parietal cortex and prefrontal
cortex that same network is

characteristically active in prospective
planning tasks and the overall picture

here that we're working from is the idea
that this sort of prospective replay in

the hippocampus could drive
reinstatement out in the cortex of the

representations for the objects and
scenes and things that are relevant

along the route that we think we want to
take okay so we can actually visualize

through this type of reinstatement what
we think we're going to encounter along

the route that we want to take and so
this is a potential mechanism whereby

the hippocampus could help us to
simulate future experiences and we

wanted to test whether this was
plausible whether this seemed to be the

case in humans and this is not an easy
problem to solve either fMRI in general

has a lot of limitations you have to
think outside the box a bit when you're

trying to test some of these mechanistic
questions in humans our general approach

was to use a region of interest analysis
where we extract patterns of activity

from different key areas of the brain
areas in this network and we get these

patterns of activity using whole brain
high-resolution fMRI and this is

actually a really recent development in
imaging technology it's something we can

now do at cabie and that's very exciting
and so now we can test fine-grained

representational information across the
brain at the same time so using this we

developed a virtual navigation task to
mirror that used in the rodent study and

I'll briefly orient you to what this
task looked like and what the demands

were for participants so this is an
overhead view here what we had were five

hidden goal locations and they're
demarcated by these blue ellipses and

outside that the scanner participants
learned to navigate back and forth back

and forth the
Queen these gold locations until they

were really sick of them and then we put
them in the scanner and we did an fMRI

task and what they saw when they were
performing this task was a really sparse

environment they really only had these
distal queues available to them so these

mountains and clouds and that really
forced participants to rely on precise

spatial memory in order to localize the
correct goal location on a given trial

those goal locations were actually
associated during planning with a pair

of fractal stimuli and there's an
example of this here those served as

sort of landmarks that we could give to
a participant later during the fMRI task

and we could use those to cue a goal
location for them so we could say hey

please go to where you remember seeing
these fractals they would then be hidden

from view in the actual navigation task
but the participant would use their

spatial memory to get to this position
and so participants learned to find

these locations really well and in the
fMRI tasks we would give them one of

these cues and they'd have to navigate
from their current position to their

current goal location it's a little
different from the rodent task in the

sense that they're not generating their
own goal or giving them a goal so what

the way this worked was pretty
straightforward basically each trial

participant would start facing the
ground and while they're looking at the

ground we'd show them one of those two
fractals for a location I would say

please think about going to this
destination we then take the fractal

away and we'd hold them staring at the
ground for what we call the planning

period and the reason it's a planning
period in this task is their task was

actually to take the shortest path
around the circle from their current

position to the current goal location
it's a fairly low-level planning task

but actually they're constantly
zigzagging back and forth around the

circle and so they do always have to
take a moment to reorient to where am i

right now which way am I facing where's
my current goal so they use this period

to think about how to get there then
they make their selection for what route

they want to take and they actually
navigate the environment without those

fractals visible
to where they think the current goal

location is okay so I mentioned these
sort of newer fMRI analysis tricks to

you already we're going to apply them
here we can use machine learning type

pattern analyses to attempt to decode
neural signals that are nested within

patterns of activity in say the
hippocampus and see if there is a code

nested in those patterns that's
different for different goal locations

in the environment and the really
interesting thing for our research goal

is not only is there a code for
different goal locations in the

environment in the hippocampus but is
that code reinstated prospectively

during the planning period so before
they've actually got to that goal and

they're thinking about how to get there
instead and what we find is that in fact

there is what's shown here is a three
dimensional transformation of the

pattern classifiers confuse ability
matrix which is a mouthful but what this

yellow bar here shows is that actually
during planning when a participant is

say here in the environment there's
significant detectable evidence for the

future goal location that the
participant is planning to go on to

navigate to and so this complements
nicely some of the work that we see in

the rodents one of the things that we
also observed is that this was not

specific to the hippocampus but it also
encompassed some subset of the memory

system outside the hippocampus and so we
see these prospective representations

rising in the hippocampus but they're
also actually correlated on a trial by

trial level with the representation of
future locations out in the cortex we've

got parapa campbell parry Rhino retro
spleen neocortex out in the brain and so

this to us is consistent with this
memory replay type framing of

prospective planning so we have some
prospective signals in the hippocampus

and we can't infer causality from this
but perhaps they are driving

replay of the complementary
representations out in the cortex for

this future event now I don't know if
any of you are looking at this with two

critical and I but we did show them a
fractal cue during the planning period

for their current goal and here we're
decoding their current goal so you might

be wondering well how do you know you're
not just decoding the fractal and that's

a very fair question and in fact it's
true that the fractal is a critical part

of the mnemonic representation of the
goal location but we had to actually

exciting data points from the study that
kind of suggests that that this

representational information extends
beyond just decoding the fractal so

first at a more basic level we can look
at how the the machine learning

algorithm is thinking about the task and
one of the things that's really

important is that the pattern classifier
can generalize across multiple cues for

the same location and so at the very
least the representations that we're

decoding are maybe a conjunction of the
fractals we hope they're even more than

that there's something about the place
itself but what was really important for

us is that if you look at the pattern
information that's instantiated during

planning in this task you can then go to
the pattern information when they

actually arrive at the goal and what we
find is that the neural code for goal

location 5 that we observed during
planning is actually extremely similar

to the neural code that's actually
elicited when you go on to get to go

location 5 and so this really supported
our view that this is some sort of

prospective replay of that future
position and I mean we would note of

course that the fractals are not a part
of the perceptual information when they

actually get to the goal in the fMRI
task now

if if this is about mental simulation we
should see some additional information

here in the brain so I already mentioned
your when you're planning your route to

EB B you're not just bringing the goal
to mind be also thinking about the route

you're going to take to get there and so
what we should see if the hippocampus is

providing a mechanism for mental
simulation is some sort of prospective

representation of locations that are
falling along the route to the long-term

goal and we call these sub goals and
what we indeed observed you can kind of

see this here although there are
actually various ways of looking at this

this signal what we found though is that
actually there was greater neural

evidence for the representation of the
location that falls along the route that

they chose then representation for
locations that fall along the

alternative route that they could have
taken but they chose not to take and

they were not supposed to take it
because it was the longer one and so

this was very exciting because it really
suggests that in humans we can at least

with the right paradigm major
prospective route specific replay of a

route that a participant wants to take
okay so I showed you that nice figure at

the beginning of the hippocampus and
it's anatomical connections out to the

cortex and I mentioned already that
memory replay is probably a whole brain

phenomenon at least on some level one of
the things I want to highlight here as

the transition into the final study I
want to talk about is that on a trial by

trial level the degree to which future
goal information was brought to mind in

the hippocampus was predicted by the
degree of activity in the medial and

lateral frontal polar cortex and this is
putative planning machinery in the human

brain and this was particularly
interesting to us because there's been a

lot of theorizing and speculation about
this part of the brain

okay so if you look at the prior
literature on prospective planning and

and route planning there's sort of a
characteristic network of activity that

includes some of the areas I've talked
about we've got retro spleen neocortex

etc we very reliably find activity in
the anterior medial prefrontal cortex

and in the anterior lateral prefrontal
cortex and in the past including some of

my own work that route disambiguation
type paradigm that I showed you before

we've seen that these anterior
prefrontal regions are associated with

rerouting yourself in an environment to
a future goal usually in response to

some sort of changing task rule or
encountering some kind of barrier and

saying boy now I have to actually think
about this how am I gonna get to the

goal because there's a police barrier
here and so it's exciting to see that

these are the areas that pop out and
correlate on a trial wise level with

prospective memory information in the
hippocampus and from that prior

literature it's it's been proposed that
those areas may be are helping to

integrate the output from the memory
system into some kind of strategic plan

but one of the problems with that
circular track design it's probably

already in some of your minds is boy it
is boring it is the most simplistic

paradigm you could give someone they
just drive back and forth around the

circle for three hours so that's no fun
in an effort to more directly link these

computations in these areas to mental
simulation and maybe planning something

new that you've never quite done before
I developed a more complex paradigm and

I'm actually going to talk about this in
the context of a stress manipulation

because it's very interesting to think
about how stress affects our ongoing

cognition and our behavioral flexibility
and when I present you this study it's

sort of a two-part story then it's going
to tell you a little bit about how

stress affects memory
but in so doing it's also going to tell

you a little bit more about how
prospective thought might occur in the

brain in relation to these two areas
that I've highlighted

okay so we've recently shown in a non
spatial study that psychological stress

when you when you manipulate this in
individuals can actually have a really

powerful memory impact on the memory
system so it impairs our ability to

engage in recollection and even when we
do engage in recollection under

psychological stress the details that we
recollect are less accurate and less

extensive and we've attributed this to
glucocorticoid action that has a

powerful impact on the hippocampus but
there's also there's slower

catecholamine effects from the stressor
as well but if stress impairs

hippocampal function if it disrupts
prefrontal computations then it should

have a really powerful impact on
navigational planning because we can't

recollect in an efficient way all of the
details that we need to in order to plan

a new route through the environment and
so I developed a task to really try and

tax this this is an overhead view of one
of twelve virtual towns that we built

this is an example of what they look
like to participants basically what we

did was we had participants learn these
circuitous routes throughout the town

outside the scanner and they navigate
these routes over and over and over

again kind of like the circular track
study they get really really good at

navigating these familiar routes the
environments are set up in a way where

structurally there are good shortcuts
between a given location and long-term

goal and so if we give them what we call
the prospects in tasks for this study we

plot them down here and we say hey go to
location B they actually have two

solutions they can follow a highly
familiar route or they can generate a

shortcut that they've never
before to get there and that's a I think

an important difference from the
circular track study this is a bigger

decision making problem for them and it
also presents an opportunity to simulate

a truly novel experience now to do this
study we had a stress group and we had a

control group the stress manipulation
was threat of electric shock sounds a

lot worse than it is if any of you want
to try it out I can shock you it's not a

big deal
but it is kind of creepy to think about

and so it does elicit a stress response
and we use the same manipulation that we

use for that recollection study that I
alluded to during training the

participants were unaware of what group
they'd be in so everybody learned the

environment sort of free of any abnormal
stress other than you know the prospect

that they might become a member of the
stress group the next day we put them in

the scanner actually after two days of
training we put them in the scanner and

we give them that prospect asked we plop
them down at a location like this and

first they're spun around slowly so that
they can see where they are in the

environment go okay I recognized this
location we then give them a taxi-driver

assignment so they're supposed to go
pick up George Clooney in this example

but it could be any number of famous
folks they're then held in place for

what we call a perspex in period just
like in the circular track study and

then they're cut loose and they navigate
using whatever strategy they want to get

to the long-term goal and then we say
congratulations you got to George

Clooney and in the stress group they're
under threat of electric shock while

doing this and so they're kind of upset
about the possibility that they'll get

shocked and what we can do is we can
compare their behavior on this task and

we can also look at some of the neural
signals in these key areas like

hippocampus and frontal polar cortex and
the first thing I'll show you is that

the stress manipulation works so this is
a chart

showing the baseline corrected salivary
cortisol measure from our two groups of

participants can no sorry my label is
gone so blue is the control group and

orange is the the stress group and so
what we did was on their second day of

training before they knew what group
they'd be and we collected a saliva

sample we get a baseline for this
certain time of day for these

individuals and then we bring them back
at roughly the same time the next day

for the prospection tasks in the scanner
and so these time points are during the

prospection task and so you can see that
we've significantly elevated the

cortisol in these individuals under our
threat of shock now what was really cool

for us to observe is that it has a
pretty profound impact on behavior so

there are variety of behaviors of
participant could engage in to solve

this task but we'll focus it on the two
most easily interpretable ones here so

they could take the novel shortcut or
they could take the highly familiar

route and what you'll see is that under
psychological stress there's a big shift

here where stressed individuals take
fewer shortcuts and they take more

familiar routes so already behaviorally
we're seeing that people are behaving

less efficiently and less flexibly on
this task and you might attribute some

of this to a failure to plan out what
sort of route they're going to take and

you might attribute that to impaired
spatial memory retrieval and that is in

fact what we find so in the stress group
there was dramatically reduced

hippocampal and frontal polar cortex
activity during the planning period of

this task and if you look at the
magnitude of activity in those regions

of interest it actually predicts across
participants and on a trial by trial

level how efficiently they're going to
go on subsequently to navigate the

environment so across participants
people who navigate more efficiently

recruit frontal polar cortex more during
planning than people who primarily in

the stress group
to the stressor and they go on to behave

less efficiently you can also look at
the trial by trial level and say wow we

can actually predict whether you're
going to take a really short efficient

route on this trial
or a long secure Road circuitous route

based on how strongly you're activating
those two areas that I've been

highlighting for you we can also turn to
this machine learning pattern decoding

type approach and we can see that not
only does the degree to which you take a

shortcut not only is that predicted by
replay evidence for these future routes

that you go on to take but activity in
these areas predicts the degree to which

you see replay for the future route
you're going to take and one of the one

of the things you can see here oranges
again the stress group of blue is the

control group and there is a slight
bifurcation there as might be expected

because the control group is taking more
shortcuts and they are also activating

this area more than the stress group the
last thing I want to highlight from this

is that I think introspectively we all
know that when we're stressed we don't

usually just curl up on the floor in a
ball and cry I mean we still engage in

planning we still think we try to solve
the tasks and so this isn't really a

story about people just collapsing on
this task but it's it's more one of a

shift from goal directed prospective
cognition towards maybe something that's

more of a safe familiar action that you
know is going to get the job done and so

what we can do is we can take these
pattern analyses pattern analysis

measures and we can also look at how
people think about the task so if we

look at pattern evidence for the
different locations in the environment

so we've got the sub goal that falls
along the novel shortcut we've got the

long term goal and then we've got
evidence for locations along the

familiar route what you can see is that
control participants show a similar

pattern to what I showed you from the
circular track study so when they're

planning how to get to the goal we see
really strong long-term goal

evidence a little bit of subgoal
evidence and then weakest is the

evidence for the alternative route in
the brain what I want to highlight is

just that the stressed individuals even
when they're planning a shortcut seem to

do this in a fundamentally different way
what we see is very little evidence for

the long term goal and relatively higher
evidence for the immediate alternative

choices in front of them and there are a
few different ways this analysis could

go but I actually like this finding
because it fits one of the hypotheses we

had for this study which is that because
stress makes it difficult for us to

effectively retrieve information from
our cognitive map our prospective

thought might be not only impaired but
restricted in its scope we might focus

more on immediate choices in front of us
and think less about the long term

things that are down the road whatever
we'll deal with those intersections when

they get to them and this kind of looks
like that although there are some

limitations with the measure here that
you know there's still some alternatives

open but this was a really interesting
observation okay so I'll end the talk

just by summarizing a little bit and
that is that the overarching theme of

the research that we're doing is trying
to get a better characterization of the

neural mechanisms that help us retrieve
memories in an efficient goal-directed

way and in so doing enable us to plan
and navigate our world efficiently and

so we've seen a few highlights from
these different experiments here maybe

in the interest of time give you some
time to asks questions what I'll do is

just take one or two minutes to plug
some of the other cool work we have

going in our lab the builds on this and
then then open it up so I you know I'm

really proud about some of the we built
up a lab really quickly in the first

year here and I'm really proud and
excited about some of the new work we

have going on one of the things that's
really important for understanding

flexible navigation and your ability to
traverse an environment like that those

virtual towns there's is trying to
understand how barriers and and the

structure of the space actually
influences your memory for the space and

so we have some research ongoing and in
development to look at how both

perceptual and cognitive barriers affect
the way that your brain represents space

and now we have some tools to actually
look at both those things we can get the

cognitive measures we can actually look
at what sort of information people are

retrieving and in pursuing these sort of
research questions we can also in the

process look at whether we can develop
interventions to maybe say protect you

from the effects of stress or even
improve spatial learning so just as a

cool visual here we can ask a question
like well if we give someone the ability

to on-the-fly get x-ray vision in the
environment can this reduce the

difficulty for them of integrating
individual root components into some

sort of cohesive cognitive map of the
environment and maybe not everybody's

the same at doing this so a really
interesting question for us is what are

the cognitive and neural mechanisms that
differentiate people who can take that

x-ray vision and run with it and really
build a super-awesome map of the space

in their minds versus someone who they
just struggle at learning Maps and you

can give them all the tools that you
want but they're just not going to get

any better at it with those extra tools
why are these people different it's a

good question and so these are some of
the things we're pursuing so super brief

thanks - well a lot of my past
collaborators but also my surprisingly

large and growing lab here at Georgia
Tech and I'll just stop and take any

questions

yes that's a very good question I didn't
explain much about the interface so

typically we have participants using a
button box where they've got a forward a

left and a right arrow it's usually a
continuous measure so they can kind of

drive they can start and stop as desired
just by holding down Drive and holding

down left and they'll go in an arc for
example but different people are better

at some people are better at this than
others and so that's one reason why

we've avoided things like joysticks
because that's extra that's an extra

level of sort of prior experience it's
in terms of its impact on some of these

findings and a lot of the cases we're
looking at we're treating participant

variability as a random factor but I
think when you look at those individual

differences like in the stress study it
can be a big big factor to think about

one of the things I I guess I kind of
glossed over it but they so for like the

stress experiment they train for two
days on the task before we go and get

those sort of critical measures so I
think that's gonna flatten out a lot of

individual differences with the actual
interface itself but it's it's

definitely an important concern and and
it might still have an impact on their

learning yeah yes

right so I think the question is when
you look at for example that last slide

with differences between the way people
appear to be thinking about the task how

do you differentiate this being sort of
an altered memory signal from a

different type of strategy different way
of thinking about the problem and you

know that's a that's a good question I
don't have an answer for you from this

particular paradigm one of the things we
can do is we can look at other

behavioral measures to try and get a
sense for how they're thinking about the

task and then map that back onto the
neural data so I didn't mention this but

you can look at the number of pauses
people make how often they stop and go

when they're going through the
environment and what you see is that

even though these people are taking
novel shortcuts more often they actually

also doing this more confidently more
efficiently they very rarely stop and

they seem to take sort of straight as an
arrow people in the stress group even

when they take the familiar route they
actually pause more frequently and they

take longer to get there and so it's
it's almost like they didn't pre plan

what they were going to do they're
taking it as they go and it seems like a

very different more reactive strategy to
approach in the environment

you