hopfield-net.apl

⍝ My attempt to implement a hopfield network with synchronous update rule in APL

⍝ Our network is a weight matrix created from array of patterns (vectors)
net ← {ws ← ⊃+⌿∘.×⍨¨⍵ ⋄ (1 1⍉ws) ← 0 ⋄ ws}

⍝ LHS argument is the net, RHS is the probe we want to update
update ← {(2×0∘<⍵+.×⍺)-1}

⍝ Example use

⍝ Here are two patterns we want to store in the network
patterns ← (¯1 ¯1 ¯1 1 1 1) (¯1 1 ¯1 1 ¯1 1)

⍝ We train the net with these patterns
network ← net patterns

⍝ We create a noisy probe
probe ← 1 ¯1 ¯1 1 1 1

⍝ And we use the network to retrieve a memory based on the probe
network update probe