HOW WE LEARN
• The brain draws its knowledge from its environment.
• Nerve cells posses a remarkable ability to constantly adjust their synapses to signal they perceive.
• Our 23 pair of chromosomes contain 3 billion pairs of the letters A,C,G,T -the molecules adenine,cytosine,guanine and thymine. Information is contained in bits and DNA contains 6 billion bits.
• There are 86 bn neurons and a thousand trillion connections
• The capacity of brain is on the order of 100 terabytes
• Animal that possessed even a rudimentary capacity to learn had a better chance of surviving than those with fixed behaviors & they were more likely pass their genome to next generation.
• Most of what we know know about the world was not given to us by our genes: we had to learn it from our environment or from those around us.
• From making fire & designing tools to agi,exploration etc is the story of humanity of constant self exploration.
• The brain has an extra ordinary ability to formulate hypothesis & select those that fit with our environment.
• The active verification of knowledge, rejection of simple heresay & the personal construct of meanings are essential filter to protect us.
• Mindset plays an important role in learning. Having a deeply entrenched view that anyone can progress is, itself , a source of progress.
• retrieval practice- it is one of the most effective educational strategies. The mere act of putting your memory to test makes it stronger.
• self awareness or meta memory is useful b/c it allows learner to focus harder on difficult items.
• to get information into long term memory ,it is essential to study the material ,then test yourself.Self testing is one of best strategy.
• space out learning- the rule of thumb is to review the info. at intervals of app. 20% of desired memory durations-for instance , rehearse after 2 mnths if you want memory to last about 10 mnths.
• to keep information in memory as long as possible,it is best to gradually increase the time intervals. Start with rehearsels everyday,then review info. after weeks, a month,then a year. This strategy guarantees optimal memory at all points.
• 4 pillars of learning-
1 Focussed attention- it allocates resource to info. it considers most essential. Attention makes relevant info. sensitive & it increases influence on brain.
attention system
• alerting-when to attend
• orienting-what to attend to
• exec. attention-how to process attended info.
• selective attention-select the relevant ,ignore irrelevant.
multitasking:-
• our brain can process only one piece of info. at a time .
• if 2 simple task given simultaneously . 1st one might get an attention & 2nd one might get delayed or forgotten altogether.
How can we multitask?
• Automization frees the conscious workspace by routinizing an activity, we can execute it unconsciously, without tying up brains central resources.
2 active engagement-learning requires an active generation of hypothesis with motivation and curiosity.
• motivation is an essential : we learn well only if we have clear goal and we fully commit to reaching it.
• to better digest new concepts, rephrase them into words or thoughts of your own.
• curiosity- fundamental driver of an organism: it pushes us to act, it encourages us to explore
• the degree of curiosity corelates with activity of nucleus accumbens and ventral tegmental areas, 2 essential regions of dopamine brain circuit.
• the more curious you are about something, the more likely you are to remember it.
• laughing or positivity seems to increase curiosity and enhance subsequent memory.
• curiosity occurs whenever our brain detect a gap b/w what we already know & what we would like to know- a potential learning area
• to maximise what we learn,we have to constantly enrich our env, with new objects that are stimulating enough to not be discouraging.
• as learning progresses ,the expected learning gain shrinks: the better we master a field , the more we reach the limits of what it can offer & less interested we are in it.
• to maintain curiosity, one must always provide brain with stimulants that match their intelligence.
• one can restore desire to learn by offering them stimulating problems carefully tailored to their current levels.
• the neuroscience of motivation is clear: desire to do action x must be associated with an expected reward,be it material or cognitive.
3 error feedback- whenever we are surprised because world violates our expectation , error signal spreads throughout our brain.They correct our mental models,eliminate inappropriate hypothesis ad stabilise the most accurate ones.
• Every error offers an opportunity to learn.
• the quality and accuracy of feedback we receive determines how quickly we learn.
• surprise is one of the fundamental drivers of the learning.No learning is possible without an error signal. organisms can only learn when events violate their expectations. No surprise, no learning.
• being immersed in fear free,stimulating env. can re-open synaptic plasticity.
4 daily rehearsel and nightly consolidation- our brain compiles what it has acquired and transfer it to long term memory, thus freeing neural resources for further learning.Repetition plays an important role in consolidation process.
• learning is more efficient when done at regular intervals, rather cramming an entire session in 1 day, we are better off spreading out the learning. The reason is simple: every night our brain consolidates what it has learned during the day.
• while we sleep,our brain remains active, it runs a specific algorithm that replays the important events it recorded during previous days & gradually transfer them into more efficient compartment of memory
• durng sleep,the neural circuits that we used during preceding day gets reactivated.
• experiments indicate that information you learned is better consolidated the next morning if you had slept while being exposed to another smell.
• If we smell an aroma while we take on new knowledge & then sleep next to source to that same odor, we will find it easier to recall info. at later date.
• practice during the day and sleep during night to reactivate and consolidate what we acquired.
• By constantly attending to probabilities and uncertainties, brain optimises its ability to learn.
• To learn is to progressively form an internal model of the outside world.
• At its core ,intelligence is a process that converts unstructured information to useful and actionable knowledge.
• Through learning ,raw data that strikes our senses turned into refined idea,abstract enough to be reused in a new context.
• Behind the scenes,our sensory areas ceaselessly compute with probabilities & only most likely model makes it to our consciousness.
• It is the brains projection that ultimately give meaning to flow of data that reaches us from our senses.In the absense of an internal model,raw sensory inputs would remain meaningless.
• Learning allows our brain to grasp a fragment of reality that it had previosly missed & to use it to build a new model of the world. It can be part of external reality as when we learn from history,science etc but also brain learns to map the reality internals to our bodies, as we learn to play musical instrument etc.
• 7 definition of what leaning means
1. learning is adjusting the parameters of mental models-it boils down to searching among internal models that best correspond to the external model
2 learning is exploiting a combinatorial explosion
• combinatorial explosion-the exponential increase that occurs when you combine even a small no of possibilities.
• In our brain ,there are 86 bn neurons, each with about 10k synaptic connections whose strength can vary.The space of mental representation that opens up is practically infinite.
• Human bran breaks down the problems of learning by creating a hierarchical,multilevel model
3 Learning is minimising errors
• Brain observe the errors & use them to adjust internal state in directions that they feel is best able to reduce error
4 learning is exploring space of possibilities- random exploration,random curosity & noisy firing plays an essential role in learning .
• characteristic of human species is relentless search for abstract rules, high level conclusion that are extracted from a specific situation & subsequently tested on a new observation
5 learning is optimising a reward function- The network can correct itself by calculating the diff. b/w its response and correct answer. This procedure is known as “supervised learning”.
• reinforcement learning
• act & self evaluate
• actor-critic combination is one of the most effective strategies of contemp. AI, when backed by hierarchical neural network , it works wonders.
• bootstrapping-a neural network can become a world champion simply by playing againsti itself.
• adversarial learning- consists of training 2 opponent system-one that learns to become an expert & other whose sole goal is to make 1st one fail.
• metacognition-the ability to know oneself,to self evaluate,to mentally stimulate what would happen if we acted this way or that way-plays a fundamental role in human learning.
• meta rules allows to search for a possible meaning among object around them, rather tha treating word as verb or adjective.
6 learning is restricting a search space- the more parameter the internal model has, the more difficult it is to find the best way to adjust it.
• curse of dimensionality- having too many free parameter can be curse while system learns ,it is unable to generalise a new situation.The ability to generalise is key to learning.
• one of the most effective interventions ,which can both accelerate learning and improve generalisation is to simplify the models.
• what is learned in one place can be reused everywhere else.
• the system has to tune only a single filter that it applies everywhere ,rather than a plethora of diff. connections for each locations.
7 learning is projecting a priori hypothesis
• exploitation of innate knowledge- convolutional neural network learns better & faster than other type of neural network b/c they donot learn everything.
• what I learn in 1 place can be generalised anywhere else.
• rather than learning from scratch ,it is more effective to rely on prior assumption that clearly describe the basic laws of domain that must be explored & integerate this into very architecture of system.
• to learn is to eliminate.
why our brain learn better than current machines?
1. unlike a computer, we posses the ability to question our beliefs & refocus our attention on those aspects of image that dont fit with our 1st impression
2. human learning is not just setting a pattern recognition filter, but also forming an abstract model of the world.
3. social learning- we learn a lot from our fellow human through language.
• what human do exceedingly well is to integerate new information with an existing network of knowledge.
4 the ability to discover a general laws that lie behind specific cases-
• the ability to make infinite use of finite means characterises human thought.
• humans can integerate piecewise knowledge into single formula & integerated pattern of activity happens in “Brocas area”.
How scientific reasoning work?
1 researchers apply a simple logic-they state several theories, unravel the web of ensuing predictions and eliminate the theories whose predictions are invalidated by experiments and observations
2 In each of our mind,ignorance is gradually erased as our brain successfully formulates increasingly accurate theories of the outside world through observations.
3 learning is reasoning like detective-going to hidden cause of phenomenon, in order to deduce the most plausible models that govern them.
3 bayesian approach -
• To learn is to be able to draw as many inference as possible from each observations
• tracing every observation to its most plausible cause allows us to return to foundation of logic.
• the bayesian theory allows us to travel from observation to causes.The theory explains how to update our belief after each observations.
• the brain should calculate an error signal : the diff. b/w what the model predicted and what has been observed. The bayesian algorithm indicates how to use this error signal to modify internal models of the world.
• when you eliminate the impossible,whatever remains ,however improbable, must be the truth.
4 our adult judgement combines 2 levels of insights-innate knowledge of species and our personal experiences.
5 learning is most effective when we have vast spaces of hypothesis, a set of mental model with myriad settings to choose from & sophisticated algorithm that adjust those settings a/c to data received.
self organisation
1 self organisation is ubiquitous in the developing brain.
2 grid cells are ‘GPS of brain’- it forms a network of triangle which grouped together to form a network.
3 hexagon being frequently produced from giraffe skin to beehives
• hexagon is common b/c it is a shape that fills a plane with equal size units and leaves no wasted space.It is most mechanically stable
• brain contains a neuronal map with hexagonal symmetry.
4 foundation of our core brain circuits arise through self organisations by bootstrapping themselves from database generated inside the system.
5 the knowledge of mathematical thing is almost innate in us. For people who are illiterate also know how to count and reckon.
neuroplasticity
• in addition to reinforcement of pre established pathways,new pathways are created by the progressive growth of terminal dendritic and axonal process.
• neurons, synapses and microcircuits that they form are material hardware of brain plasticity, they adjust each time we learn.
• as we learn, all elements can change: the presence ,number and strength of synapses and neurons.
• our brain contains about 100 trillions of synapses & synapses are genuine nanoprocessors of the brain.
• hubbs rule-when 2 neurons are activated at same time or in short successions,their connections strengthens
• emotional circuits of our brain considered most significant.
• when we learn, the explosion of new synapses forces the neurons to grow additional branches.
• to make & break a few million synapses/second , it requires a balanced diet,oxygenation & physical exercises.
• vitamin B1 which contains thiamine is imp. for brains health.
• the acquisition of novel skill doesnt require a radical rewriting of cortical circuits but a repurposing of existing org.
memory
1 episodic memory- records the episodes of our daily lives
• neurons in hippocampus memorize the context of each event
• they encode where,when,how and with whom things happened
• hippocampus is involved in all kinds of rapid learning.
2 semantic memory-memory donot stay in hippocampus forever.
At night,brain plays them back and moves them to a new location within cortex.There,they are transformed into permanant knowledge.
3 procedural memory- when we repeat the same activity over and over again .Neurons gets strengthened and information flows better in future.