Files
IronpenGPU/src/learn.jl
2023-09-23 09:57:40 +07:00

572 lines
22 KiB
Julia

module learn
export learn!, compute_paramsChange!
using Statistics, Random, LinearAlgebra, JSON3, Flux, CUDA, Dates
using GeneralUtils
using ..type, ..snnUtil
#------------------------------------------------------------------------------------------------100
function compute_paramsChange!(kfn::kfn_1, modelError::CuArray, outputError::CuArray, label)
lifComputeParamsChange!(kfn.timeStep,
kfn.lif_phi,
kfn.lif_epsilonRec,
kfn.lif_eta,
kfn.lif_eRec,
kfn.lif_wRec,
kfn.lif_exInType,
kfn.lif_wRecChange,
kfn.on_wOut,
kfn.lif_firingCounter,
kfn.lif_firingTargetFrequency,
kfn.lif_arrayProjection4d,
kfn.lif_error,
modelError,
outputError,
kfn.inputSize,
kfn.bk,
label,
)
alifComputeParamsChange!(kfn.timeStep,
kfn.alif_phi,
kfn.alif_epsilonRec,
kfn.alif_eta,
kfn.alif_eRec,
kfn.alif_wRec,
kfn.alif_exInType,
kfn.alif_wRecChange,
kfn.on_wOut,
kfn.alif_firingCounter,
kfn.alif_firingTargetFrequency,
kfn.alif_arrayProjection4d,
kfn.alif_error,
modelError,
outputError,
kfn.inputSize,
kfn.bk,
label,
kfn.alif_epsilonRecA,
kfn.alif_beta,
)
onComputeParamsChange!(kfn.on_phi,
kfn.on_epsilonRec,
kfn.on_eta,
kfn.on_eRec,
kfn.on_wOutChange,
kfn.on_arrayProjection4d,
kfn.on_error,
kfn.on_synapticActivityCounter,
outputError,
)
# error("DEBUG -> kfn compute_paramsChange! $(Dates.now())")
end
function lifComputeParamsChange!( timeStep::CuArray,
phi::CuArray,
epsilonRec::CuArray,
eta::CuArray,
eRec::CuArray,
wRec::CuArray,
exInType::CuArray,
wRecChange::CuArray,
wOut::CuArray,
firingCounter::CuArray,
firingTargetFrequency::CuArray,
arrayProjection4d::CuArray,
nError::CuArray,
modelError::CuArray,
outputError::CuArray,
inputSize::CuArray,
bk::CuArray,
label,
)
eRec .= phi .* epsilonRec
# 2D wRec matrix contain input, lif, alif neurons. I need only lif neurons
startIndex = prod(inputSize) +1
stopIndex = startIndex + size(wRec, 3) -1
startCol = CartesianIndices(wRec)[startIndex][2]
stopCol = CartesianIndices(wRec)[stopIndex][2]
# some RSNN neuron that has direct connection to output neuron need to get Bjk
# from output neuron that represent correct answer, the rest of RSNN get random Bjk
onW = @view(wOut[:, startCol:stopCol, sum(label+1), 1]) # label+1 because julia is 1-based index
_bk = @view(bk[:, startCol:stopCol, 1])
mask = iszero.(onW)
bk_ = mask .* _bk
bkComposed = onW .+ bk_
nError = bkComposed .* modelError
nError = reshape(nError, (1,1,:,1))
# compute wRecChange of all neurons wrt to iᵗʰ output neuron
wRecChange .+= (eta .* nError .* eRec)
# frequency regulator
targetFiringCount = firingTargetFrequency .* timeStep
freqError = (firingCounter .- targetFiringCount) ./ timeStep
freqWRecChange = -1 .* freqError .* eta .* eRec
wRecChange .+= freqWRecChange
# reset epsilonRec
epsilonRec .= 0
end
function alifComputeParamsChange!( timeStep::CuArray,
phi::CuArray,
epsilonRec::CuArray,
eta::CuArray,
eRec::CuArray,
wRec::CuArray,
exInType::CuArray,
wRecChange::CuArray,
wOut::CuArray,
firingCounter::CuArray,
firingTargetFrequency::CuArray,
arrayProjection4d::CuArray,
nError::CuArray,
modelError::CuArray,
outputError::CuArray,
inputSize::CuArray,
bk::CuArray,
label,
epsilonRecA::CuArray,
beta::CuArray,
)
eRec .= phi .* (epsilonRec .- (beta .* epsilonRecA)) # use eq. 25
# 2D wRec matrix contain input, lif, alif neurons. I need only lif neurons
startIndex = prod(inputSize) +1
stopIndex = startIndex + size(wRec, 3) -1
startCol = CartesianIndices(wRec)[startIndex][2]
stopCol = CartesianIndices(wRec)[stopIndex][2]
# some RSNN neuron that has direct connection to output neuron need to get Bjk
# from output neuron that represent correct answer, the rest of RSNN get random Bjk
onW = @view(wOut[:, startCol:stopCol, sum(label+1), 1]) # label+1 because julia is 1-based index
_bk = @view(bk[:, startCol:stopCol, 1])
mask = iszero.(onW)
bk_ = mask .* _bk
bkComposed = onW .+ bk_
nError = bkComposed .* modelError
nError = reshape(nError, (1,1,:,1))
wRecChange .+= (eta .* nError .* eRec)
# frequency regulator
targetFiringCount = firingTargetFrequency .* timeStep
freqError = (firingCounter .- targetFiringCount) ./ timeStep
freqWRecChange = -1 .* freqError .* eta .* eRec
wRecChange .+= freqWRecChange
# wRecChange .+= 0.01 .* ((firingTargetFrequency - (firingCounter./timeStep)) ./ timeStep) .*
# eta .* eRec
# reset epsilonRec
epsilonRec .= 0
epsilonRecA .= 0
# error("DEBUG -> alifComputeParamsChange! $(Dates.now())")
end
function onComputeParamsChange!(phi::CuArray,
epsilonRec::CuArray,
eta::CuArray,
eRec::CuArray,
wOutChange::CuArray,
arrayProjection4d::CuArray,
nError::CuArray,
synapticActivityCounter,
outputError::CuArray # outputError is output neuron's error
)
eRec .= phi .* epsilonRec
nError .= reshape(outputError, (1, 1, :, size(outputError, 2))) .* arrayProjection4d
wOutChange .+= (eta .* nError .* eRec)
# reset epsilonRec
epsilonRec .= 0
# error("DEBUG -> onComputeParamsChange! $(Dates.now())")
end
function lifComputeParamsChange!( phi::AbstractArray,
epsilonRec::AbstractArray,
eta::AbstractArray,
wRec::AbstractArray,
wRecChange::AbstractArray,
wOut::AbstractArray,
modelError::AbstractArray)
d1, d2, d3, d4 = size(epsilonRec)
# Bₖⱼ in paper, sum() to get each neuron's total wOut weight
wOutSum = reshape(sum(wOut, dims=3), (d1, :, d4))
for j in 1:d4, i in 1:d3 # compute along neurons axis of every batch
# how much error of this neuron 1-spike causing each output neuron's error
view(wRecChange, :, :, i, j) .+= (-1 * view(eta, :, :, i, j)[1]) .*
# eRec
(
(view(phi, :, :, i, j)[1] .* view(epsilonRec, :, :, i, j)) .*
# nError a.k.a. learning signal
(
view(modelError, :, j)[1] * # dopamine concept, this neuron receive summed error signal
# RSNN neuron's total wOut weight (neuron synaptic subscription .* wOutSum)
view(wOutSum, :, :, j)[i]
)
)
end
end
function alifComputeParamsChange!( phi::AbstractArray,
epsilonRec::AbstractArray,
epsilonRecA::AbstractArray,
eta::AbstractArray,
wRec::AbstractArray,
wRecChange::AbstractArray,
beta::AbstractArray,
wOut::AbstractArray,
modelError::AbstractArray)
d1, d2, d3, d4 = size(epsilonRec)
# Bₖⱼ in paper, sum() to get each neuron's total wOut weight
wOutSum = reshape(sum(wOut, dims=3), (d1, :, d4))
for j in 1:d4, i in 1:d3 # compute along neurons axis of every batch
# how much error of this neuron 1-spike causing each output neuron's error
view(wRecChange, :, :, i, j) .+= (-1 * view(eta, :, :, i, j)[1]) .*
# eRec
(
# eRec_v
(view(phi, :, :, i, j)[1] .* view(epsilonRec, :, :, i, j)) .+
# eRec_a
((view(phi, :, :, i, j)[1] * view(beta, :, :, i, j)[1]) .*
view(epsilonRecA, :, :, i, j))
) .*
# nError a.k.a. learning signal
(
view(modelError, :, j)[1] *
# RSNN neuron's total wOut weight (neuron synaptic subscription .* wOutSum)
view(wOutSum, :, :, j)[i]
# sum(GeneralUtils.isNotEqual.(view(wRec, :, :, i, j), 0) .*
# view(wOutSum, :, :, j))
)
end
end
function onComputeParamsChange!(phi::AbstractArray,
epsilonRec::AbstractArray,
eta::AbstractArray,
wOutChange::AbstractArray,
outputError::AbstractArray)
d1, d2, d3, d4 = size(epsilonRec)
for j in 1:d4, i in 1:d3 # compute along neurons axis of every batch
# how much error of this neuron 1-spike causing each output neuron's error
view(wOutChange, :, :, i, j) .+= (-1 * view(eta, :, :, i, j)[1]) .*
# eRec
(
(view(phi, :, :, i, j)[1] .* view(epsilonRec, :, :, i, j)) .*
# nError a.k.a. learning signal, output neuron receives error of its own answer - correct answer.
view(outputError, :, j)[i]
)
end
end
function learn!(kfn::kfn_1, progress, device=cpu)
if sum(kfn.timeStep) == 800
println("zitCumulative ", sum(kfn.zitCumulative[:,:,784:size(kfn.zitCumulative, 3)], dims=3))
println("on_synapticActivityCounter 0 ", kfn.on_synapticActivityCounter[:,:,1])
println("on_synapticActivityCounter 5 ", kfn.on_synapticActivityCounter[:,:,6])
println("wOut 0 ", sum(kfn.on_wOut[:,:,1,1], dims=3))
println("wOut 5 ", sum(kfn.on_wOut[:,:,1,1], dims=3))
end
#WORKING compare output neuron 0 synapse activity when input are label 0 and 5, (!isequal).(wOut)
# lif learn
kfn.lif_wRec, kfn.lif_neuronInactivityCounter, kfn.lif_synapticActivityCounter, kfn.lif_synapseReconnectDelay =
lifLearn(kfn.lif_wRec,
kfn.lif_wRecChange,
kfn.lif_exInType,
kfn.lif_arrayProjection4d,
kfn.lif_neuronInactivityCounter,
kfn.lif_synapseReconnectDelay,
kfn.lif_synapseConnectionNumber,
kfn.lif_synapticActivityCounter,
kfn.lif_eta,
kfn.lif_vt,
kfn.zitCumulative,
progress,
device)
# alif learn
kfn.alif_wRec, kfn.alif_neuronInactivityCounter, kfn.alif_synapticActivityCounter, kfn.alif_synapseReconnectDelay =
alifLearn(kfn.alif_wRec,
kfn.alif_wRecChange,
kfn.alif_exInType,
kfn.alif_arrayProjection4d,
kfn.alif_neuronInactivityCounter,
kfn.alif_synapseReconnectDelay,
kfn.alif_synapseConnectionNumber,
kfn.alif_synapticActivityCounter,
kfn.alif_eta,
kfn.alif_vt,
kfn.zitCumulative,
progress,
device)
# on learn
onLearn!(kfn.on_wOut,
kfn.on_wOutChange,
kfn.on_eta,
kfn.on_arrayProjection4d,
progress,)
# wrap up learning session
if kfn.learningStage == [3]
kfn.learningStage = [0]
end
# error("DEBUG -> kfn learn! $(Dates.now())")
end
function lifLearn(wRec,
wRecChange,
exInType,
arrayProjection4d,
neuronInactivityCounter,
synapseReconnectDelay,
synapseConnectionNumber,
synapticActivityCounter,
eta,
vt,
zitCumulative,
progress,
device)
# transfer data to cpu
arrayProjection4d_cpu = arrayProjection4d |> cpu
wRec_cpu = wRec |> cpu
wRecChange_cpu = wRecChange |> cpu
eta_cpu = eta |> cpu
exInType_cpu = exInType |> cpu
neuronInactivityCounter_cpu = neuronInactivityCounter |> cpu
synapseReconnectDelay_cpu = synapseReconnectDelay |> cpu
synapticActivityCounter_cpu = synapticActivityCounter |> cpu
zitCumulative_cpu = zitCumulative |> cpu
# neuroplasticity, work on CPU side
wRec_cpu, neuronInactivityCounter_cpu, synapticActivityCounter_cpu, synapseReconnectDelay_cpu =
neuroplasticity(synapseConnectionNumber,
zitCumulative_cpu,
wRec_cpu,
exInType_cpu,
wRecChange_cpu,
vt,
eta_cpu,
neuronInactivityCounter_cpu,
synapseReconnectDelay_cpu,
synapticActivityCounter_cpu,
progress,)
# transfer data backto gpu
wRec = wRec_cpu |> device
neuronInactivityCounter = neuronInactivityCounter_cpu |> device
synapticActivityCounter = synapticActivityCounter_cpu |> device
synapseReconnectDelay = synapseReconnectDelay_cpu |> device
# error("DEBUG -> lifLearn! $(Dates.now())")
return wRec, neuronInactivityCounter, synapticActivityCounter, synapseReconnectDelay
end
function alifLearn(wRec,
wRecChange,
exInType,
arrayProjection4d,
neuronInactivityCounter,
synapseReconnectDelay,
synapseConnectionNumber,
synapticActivityCounter,
eta,
vt,
zitCumulative,
progress,
device)
# transfer data to cpu
arrayProjection4d_cpu = arrayProjection4d |> cpu
wRec_cpu = wRec |> cpu
wRecChange_cpu = wRecChange |> cpu
eta_cpu = eta |> cpu
exInType_cpu = exInType |> cpu
neuronInactivityCounter_cpu = neuronInactivityCounter |> cpu
synapseReconnectDelay_cpu = synapseReconnectDelay |> cpu
synapticActivityCounter_cpu = synapticActivityCounter |> cpu
zitCumulative_cpu = zitCumulative |> cpu
# neuroplasticity, work on CPU side
wRec_cpu, neuronInactivityCounter_cpu, synapticActivityCounter_cpu, synapseReconnectDelay_cpu =
neuroplasticity(synapseConnectionNumber,
zitCumulative_cpu,
wRec_cpu,
exInType_cpu,
wRecChange_cpu,
vt,
eta_cpu,
neuronInactivityCounter_cpu,
synapseReconnectDelay_cpu,
synapticActivityCounter_cpu,
progress,)
# transfer data backto gpu
wRec = wRec_cpu |> device
neuronInactivityCounter = neuronInactivityCounter_cpu |> device
synapticActivityCounter = synapticActivityCounter_cpu |> device
synapseReconnectDelay = synapseReconnectDelay_cpu |> device
# error("DEBUG -> alifLearn! $(Dates.now())")
return wRec, neuronInactivityCounter, synapticActivityCounter, synapseReconnectDelay
end
# function onLearn!(wOut,
# wOutChange,
# arrayProjection4d)
# # merge learning weight with average learning weight
# wOut .+= (sum(wOutChange, dims=4) ./ (size(wOut, 4))) .* arrayProjection4d
# # adaptive wOut to help convergence using c_decay
# wOut .-= 0.001 .* wOut
# end
function onLearn!(wOut,
wOutChange,
eta,
arrayProjection4d,
progress,)
if progress != 0
# merge learning weight with average learning weight
wOut .+= (sum(wOutChange, dims=4) ./ (size(wOut, 4))) .* arrayProjection4d
# adaptive wOut to help convergence using c_decay
wOut .-= 0.1 .* eta .* wOut # wOut .-= 0.001 .* wOut
else
#TESTING skip
wOutChange .= 0
end
end
function neuroplasticity(synapseConnectionNumber,
zitCumulative, # (row, col)
wRec, # (row, col, n)
exInType,
wRecChange,
vt,
eta,
neuronInactivityCounter,
synapseReconnectDelay,
synapticActivityCounter,
progress,) # (row, col, n)
if progress == 2 # no need to learn for current neural pathway
# skip neuroplasticity
#TODO I may need to do something with neuronInactivityCounter and other variables
wRecChange .= 0
# error("DEBUG -> neuroplasticity")
elseif progress != 0 # progress increase
# ready to reconnect synapse must not have wRecChange
mask = (!isequal).(wRec, 0)
wRecChange .*= mask
# merge learning weight, all resulting negative wRec will get pruned
mergeLearnWeight!(wRec, exInType, wRecChange, synapticActivityCounter, synapseReconnectDelay)
# adjust wRec based on repeatition (90% +w, 10% -w)
growRepeatedPath!(wRec, synapticActivityCounter, eta)
# -w all non-fire connection except mature connection
weakenNotMatureSynapse!(wRec, synapticActivityCounter, eta)
# prune weak synapse
pruneSynapse!(wRec, synapticActivityCounter, synapseReconnectDelay)
# rewire synapse connection
rewireSynapse!(wRec, neuronInactivityCounter, synapticActivityCounter,
synapseReconnectDelay, synapseConnectionNumber, zitCumulative)
# error("DEBUG -> neuroplasticity 1")
elseif progress == 0 # no progress, no weight update, only rewire
# #TESTING -w all non-fire connection except mature connection
# weakenNotMatureSynapse!(wRec, synapticActivityCounter, eta)
# prune weak synapse
pruneSynapse!(wRec, synapticActivityCounter, synapseReconnectDelay)
# rewire synapse connection
rewireSynapse!(wRec, neuronInactivityCounter, synapticActivityCounter,
synapseReconnectDelay, synapseConnectionNumber, zitCumulative)
# error("DEBUG -> neuroplasticity")
else
error("undefined condition line $(@__LINE__)")
end
# error("DEBUG -> neuroplasticity $(Dates.now())")
return wRec, neuronInactivityCounter,
synapticActivityCounter, synapseReconnectDelay
end
# learningLiquidity(x) = -0.0001x + 1 # -10000 to +10000; f(x) = -5e-05x+0.5
function learningLiquidity(x)
if x > 10000
y = 0.0
elseif x < -10000
y = 1.0
else
y = -5e-05x+0.5 # range -10000 to +10000
end
return y
end
end # module