HOPFIELD NETWORK
Draw patterns, train the network, and watch it remember
Tools
Draw
Erase
Select
Clear
Noise
Store
Recall
Step
Mode
Draw
Erase
Select
Clear
Invert
Upload Image
Presets
Network
Store Pattern
Noise:
30%
Add Noise
Speed:
10
Recall
Step
Noise
30%
Speed
10
Upload
Pattern State
Weight Map
Per Neuron
Full Matrix
Inhibit
0
Excite
Click a cell on the grid to select a neuron
W
ij
=
1
/
N
Σ
p
p
i
· p
j
Cyan
= neurons that
agree
across patterns (both ON or both OFF → excitatory).
Red
= neurons that
disagree
(one ON, one OFF → inhibitory).
Neuron Update (Inference Step)
No neuron selected
Click any cell on the grid to inspect how that neuron decides its next state during recall.
The network computes: h
i
= Σ
j
W
ij
· s
j
→ s
i
= sign(h
i
)
Stored Patterns
No patterns stored yet
Capacity
0 / 55
Energy Landscape
Energy Landscape
▼
Basins form at stored patterns
●
current state
Energy Trace
Energy Trace
▼
Status
Idle
Steps
0
Energy
--