How Your Brain Chooses What to Remember

throughout our lives our brains face a constant challenge deciding which memories to keep and which to discard think about it you probably remember what you had for breakfast today but that memory will likely fade within months yet if you win a lottery today that memory would stay with you for the rest of your life this remarkable ability to distinguish between everyday moments and life-changing events reveals one of brain's most intriguing features a system for memory selection but how exactly does the brain make these crucial decisions and how does this selection take place Recent research has

en failed An Elegant solution how a pattern of neural activity called a sharp wve Ripple acts as your brain's internal bookmarking system today we are exploring an Innovative study published in science earlier this year that shows how these events help tag important memories during the day and ensure they get properly stored while you sleep if you're interested stay [Music] tuned to solve this puzzle of memory selection we need to look deep inside the brain at a seahorse shaped structure called the hippocampus it plays a crucial role in episodic memory our ability to remember personal experiences

as sequences of events in temporal order like remembering what happened this morning or on your first day at a new job what makes the hippocampus particularly intriguing is how it switches between two distinct modes during waking hours it diligently records experiences tracking where things happen and in what order it builds what's called a cognitive map a rich internal model of the environment and the events within it however during sleep when sensory inputs Qui down the hippocampus enters an offline mode where it replays specific experiences from the day this replay is accompanied by a distinctive pattern

of brain activity called a sharp wve Ripple here is a way to think about it imagine a region of the hippocampus that spent the day building an updating its cognitive map during sleep at the onset of the replay event a massive wave of synchronized input arrives from Upstream hipocampal area triggering a so-called sharp wave this wave sweeps through the neural network like a tide and awakens countless previously quiet neurons but the brain is prepared for this surge it has safety mechanisms built into the circuit networks of inhibitory cells rapidly spring into action when activity Rises

too high suppressing other neurons this creates a rapid back and forth between excitation and inhibition that appears as a highfrequency ripple riding on top of the sharp wave this choreography of neural activity serves a crucial purpose it creates a powerful selection mechanism for memories when the wave of excitation sweeps through the network it primes many neurons to fire yet they can't all fire at once the inhibitory interneurons act as Gatekeepers limiting overall activity and creating narrow Windows of opportunity brief moments when only selected groups of neurons can be active such interaction between excitation and inhibition

sets up a form of competition in the neural network throughout the day our experiences have left their Mark by strengthening certain connections between neurons while leaving others unchanged or even weakening them some patterns of activity have become robust Pathways While others remain weak Trails when the wave of excitation arrives these different patterns each encoding a different memory compete for expression in this neural contest the strongest patterns typically win and remarkably these winning patterns that get to be replayed tend to represent the day's most significant events apart for competition such neural replays have another important feature

they occur at a high speed like watching your days experiences on Fast Forward at the neural level what originally took seconds during Behavior gets compressed into around 100 milliseconds during sleep this temporal compression turns out to be absolutely crucial for memory storage you see during sleep the neocortex the outer layer of the brain enters a special state where it becomes receptive to signals from the hippocampus the compressed timing of these replays ensures that neural activity arrives in precise temporal Windows perfectly timed to strengthen connections between neurons in the neocortex through this process patterns encoding important

events get repeatedly reactivated compressed and transferred into the CeX for permanent storage a process known as consolidation this raises an intriguing puzzle throughout the day countless experiences update the cognitive map but only some of them need to be preserved for later consolidation what does the hippocampus do with such patterns that somehow need to be preserved until sleep when the neocortex isn't yet ready to receive them this is where today's paper makes a crucial Discovery scientists had previously observed sharp wave Ripples and Associated replays during waking States particularly during brief periods of rest or immobility but

these awake ripples were puzzling why would the brain replay memories when the near cortex isn't ready to store them and also the frequency with which such awake replays occur is much lower than in sleep which is probably insufficient as proper consolidation requires more repetitions it turns out that these awake replays serve as memory bookmarks when they occur right after important events they tag specific experiences for priority consolidation during sleep let's dive deeper into the paper to explore how exactly it was discovered but before we do that I'd like to mention that the authors of the

paper also built an awesome interactive website and a collab notebook where you can play with the data and explore the plots on your own all right so to investigate the memory taging mechanism researchers needed a way to watch memories being formed tagged and Consolidated in real time they used An Elegant experiment with a simly simple task that would engage episodic memory their setup involved mice running through a figure eight maze with two identical arms each containing a potential reward site the challenge for the mice was to learn a specific strategy alternate between the arms to

receive rewards if they found a reward on the left arm in one trial the next reward would be on the right by recording from hundreds of neurons in the hippocampus as mice ran through the maze over multiple days researchers could watch the learning process unfold each day mice would run trials sleep and return the next day to try again gradually getting better at the task but this race A new challenge how do you make sense of the activity patterns from hundreds of neurons when you recording from 400 neurons simultaneously each moment creates a complex Symphony

of neural firing that is impossible to interpret by ey a popular approach in Neuroscience to make sense of such complex neural patterns is to look at the population activity and describe the collective behavior of cells in the network think about it this way at any given moment each neuron's activity level can be represented by a number how rapidly it is firing if you had just three numbers you could easily plot them in three-dimensional space with each number representing a position along X Y or Z axis but when instead of just three dimensions you have 400

it is impossible to visualize fortunately neurons don't act independently their activity is constrained by their connections and the inputs they receive meaning that the network can only generate certain patterns it's like a complex dance where dancers movements are coordinated while each dancer could theoretically move anywhere the choreography restricts them to specific patterns in terms of population activity this means that the brain's Dynamics is confined to a specific region in this vast 400 dimensional space called a manifold but how can we describe this manifold in such a high dimensional space while while 400 coordinates might seem

overwhelming the structured nature of neural Activity Works in our favor the patterns we observe can actually be described with far fewer numbers than 400 think of it like describing a curved line while it exists in 3D space with X Y and Z coordinates you really need just one number to specify where you are along the curve since the coordinates are related to each other by the equation the curve this Insight leads us to methods of dimensionality reduction techniques for finding simpler ways to describe complex patterns one particularly powerful method which is Central to this paper

is called umap uniform manifold approximation and projection while the mathematical details are beyond the scope of this video the intuitive idea is quite simple U map searches for patterns of brain activity that cluster together first it identifies which patterns of activity are similar to each other in the original 400 dimensional space creating kind of a graph of neural activity relationships then it finds a way to preserve this graph structure while bringing the data down to just three dimensions that we can visualize it's like carefully unfolding a complex origami structure to reveal the flat paper underneath

umap unwraps our neural manifold from 400 Dimensions to three exposing the hidden structure in the brain's activ ity patterns when applied to the hipocampal data umap reveals something remarkable a looped structure that perfectly mirrors the layout of the Figure 8 maze when we color each point based on where the animal was physically located in the Maze when that neural activity pattern occurred we see an exact correspondence what makes this particularly striking is that umab discovered the structure purely from neural activity without any information about the animal's actual position but the spal location is just one

dimension of this task what about the animals learning progress across trials when we color the same points by trial number instead of position we see another layer of structure emerge changes that reflect the learning progression this means that the brain's activity patterns are changing in a systematic way as the animal gets better and better at the task creating a clear trajectory through this neural space this gives us a powerful tool we can now map any pattern of neural activity onto this maze manifold and determine whether it resembles actual task related activity if it does we

can even try to identify which trial and position in the Maze it corresponds to as you might guess this becomes crucial for decoding what information gets replayed and consolidated during sharp wve ripples let's start with awake replays that occur when the animal pauses to consume the reward by projecting neural activity during these ripples onto the Learned maze manifold we can decode their content it's important to note that our umap structure was established purely from neural patterns during active maze running that's what shaped our manifold so when we find Ripple events that don't fall onto this

manifold it doesn't necessarily mean they are random noise these off manifold events simply contain patterns that look completely different from anything um map saw during running Behavior they might represent other memories future planning or different cognitive processes we just can't decode their content because we lack the right reference frame many of the ripples however do nicely fall onto our maze manifold and they reveal something fascinating these events correspond to temporally compressed replays of the maze trajectory that just led to the reward because we can decode the trial number we can confirm that these replays specifically

match the current trial the animal had just completed in other words awake replays capture recent memories of successful paths to the reward but what about the ripples during subsequent sleep while we can't know the animals true position or trial number number during sleep we can still decode Ripple content by mapping it onto our maze manifold remarkably the Sleep ripples that fall onto the manifold show striking similarities to the awake ripples they replay similar trials and Maze locations in contrast ripples recorded during sleep before the learning contain completely different patterns this reveals a crucial mechanism a

wake ripples serve to taag specific events for later consolidation but why not consolidate memories immediately during these awake replays the answer lies in how memory consolidation works first the cortex needs to be in a specific brain state to receive information from the hippocampus a state that occurs only during sleep second a handful of replays isn't enough patterns need to be repeated multiple times to be properly transferred to cortical circuits during waking hours the hippocampus can't dedicate itself to endless replay it needs to keep tracking ongoing experiences and maintaining the cognitive map as the animal performs

the task so here is the elegant solution the brain has evolved awake ripples identify and temporarily store important events in hipac cample circuits like bookmarking key pages in a book then during sleep when conditions are right for consolidation these bookmark pattern s get repeatedly reactivated and transferred to cortical networks for permanent storage while the exact cellular mechanisms of this bookmarking are still under investigation it is plausible that awake ripples trigger local synaptic plasticity within the hippocampus these changes might alter the Network's Dynamics in a way that makes C neural sequences more likely to reactivate during

sleep ripples like carving out preferred paths in the landscape of possible activity patterns such two-stage process ensures that important memories get selected during wakefulness and properly Consolidated during sleep when the hippocampus can fully dedicate itself to the task of memory transfer these insights from hipocampal physiology highlight The crucial role of direct experience and practice in consolidating new information this brings me to our today's sponsor brilliant.org brilliant is an Innovative online platform that helps you excel in stem topics they offer engaging interactive courses allowing you to learn by doing instead of passively absorbing information you get

to develop intuitive understanding of complex topics and sharpen problem solving skills in bite-sized lessons if you're interested in computational analysis we discussed in this video I particularly recommend brilliant's course on computer science foundations which covers the essential Concepts such as array operations and graph algorithms that are at the heart of many methods we talked about about today brilliant offers a large collection of courses in math science and related topics their curriculum is designed to support both beginners building core knowledge as well as experts looking to explore New Frontiers if you're ready to supercharge your learning

head to brilliant.org aramov to get a 30-day free trial of everything brilliant has to offer plus a 20% discount on annual subscription if you like the video subscribe to the channel if you haven't already and consider supporting me on patreon where you can vote for video topics and find extended versions of video scripts stay tuned for more neuroscience and machine learning topics coming up goodbye and thank you for the interest in the brain