learning in simple circuits

modulation

^temporary tuning

• temporarily adjust gain up or down

• comparatively fast, and reversible

• depends on current context

• if input removes, circuit returns to baseline

learning

^lasting re-tuning (e.g. plasticity)

• circuit baseline setting shift

• persists after the trigger ends

• often builds with experience

• changes future responses to the same situation

non-associative learning

change in response to a single stimulus

• habituation - decreased response to a repeated and harmless stimulus

• sensitisation - increase response to a strong or meaningful stimulus

associative learning

learning that two events are related

• operant conditioning - learning that action changes outcome (reward / punishment)

• classical conditioning - learning that one stimulus predicts another (natural cue leads to response

typically needs a reinforcement signals

learning at the circuit level

what changes → circuit input-output, gain, timing, probability, pattern

where it changes → synapses, excitability, gating / teaching signals

how it persists → fast changes use existing proteins and last secs to mins, slow changes (consolidation) need protein synthesis and last hrs to days

reflex learning

reflexes are not fixed

even the simplest motif has adjustability

leech shortening

• non-associative learning

• repeated experience can change threshold and strength of response

habituation → repeated touch leads to decreased withdrawal response, stimulus considered harmless

sensitisation → strong stimulus leads to increased withdrawal, stimulus considered dangerous warning

mammalian spinal H-reflex

• associative learning

• operant conditioning of a reflex pathway

  • reward gates spinal plasticity

motor neuron / interneuron excitability contributes to reflex gain shifts

presynaptic inhibition via changes in effective Ca2+ entry, vesicle release probability

learning in rhythmic CPG circuits

circuits must → remain rhythmic and coordinated but adjust timing, phase, or amplitude

variables include freq set-point, phase relationships, burst duration / amplitude, probability of expressing one pattern vs another

CPG learning often recruits the same cellular mechanisms used in shorter term modulation, but engages additional processes that stabilise these changes over time

learning in complex circuits

many synapses active at once

learning must be selective and goal-relevant - need to update connections that matter for value, success, punishment, and task demands, not juts those that happen to be active

local coincidence → Hebbian learning

global teaching signals → neuromodulators e.g. dopamine

drosophila olfactory learning

aversive olfactory conditioning → selective synaptic change, increased avoidance via reduced approach

appetitive olfactory conditioning → selective synaptic change, reduced avoidance via increased approach

reinforcement learning in mammals

teaching signals gate plasticity in many systems

hippocampus tackles the same general problems (storing and retrieving experience) but for rich contexts and episodes rather than a single cue

plasticity across lifetime

is always present, but its dominant mechanisms and goals shift with age

early development → build and refine circuits, spontaneous activity and sensory experience

adolescence to adulthood → skill learning, experience- dependent refinement

adulthood → adapt and stabilise function, more limited

ageing brain → reduced flexibility, compensatory recruitment of additional networks during tasks