reverse mode automatic differentiation w Julii

struct ComputableNode{OT, AT} <: Node
 op::OT
 args::AT
end
mutable struct CachedNode{NT, OUT} <: Node
 node::NT
 out::OUT
end
function register(op, args...)
 node = ComputableNode(op, args)
 out = forward(node)
 CachedNode(node, out)
end

function register(op, args...)
 node = ComputableNode(op, args)
 out = forward(node)
 CachedNode(node, out)
end
import Base: +, -, *, /
 +(x::Node, y::Node) = register(+, x, y)
 -(x::Node, y::Node) = register(-, x, y)
 *(x::Node, y::Node) = register(*, x, y)
 /(x::Node, y::Node) = register(/, x, y)
import Base: sin
sin(x::Node) = register(sin, x)

forward(cached::CachedNode) = cached.out = forward(cached.node)
forward(node::ComputableNode) = forward(node.op, map(forward, node.args)...)
forward(op::Operator, args...) = op.f(args...; kwargs...)
forward(var::Variable) = var.value

f(x) = sin(x*x)
x = Variable(5.0, 0.0)
y = f(x)  <-error

Próbuję zrozumieć ten algorytm, ale jest tu chyba jakiś błąd i nie potrafię go rozwiązać.

MethodError: no method matching forward(::typeof(*), ::Float64, ::Float64)
Closest candidates are:
  forward(!Matched::Operator, ::Any...) at In[82]:3

Stacktrace:
 [1] forward(::ComputableNode{typeof(*),Tuple{Variable{Float64},Variable{Float64}}}) at .\In[82]:2
 [2] register(::Function, ::Variable{Float64}, ::Vararg{Variable{Float64},N} where N) at .\In[81]:3
 [3] *(::Variable{Float64}, ::Variable{Float64}) at .\In[81]:9
 [4] f(::Variable{Float64}) at .\In[192]:1
 [5] top-level scope at In[192]:3

LINK: http://blog.rogerluo.me/2018/10/23/write-an-ad-in-one-day/