[Verse 1]
In ninety-six Monderer and Shapley wrote the page
Potential games where players seek their stage
Every Nash equilibrium has a special trait
Global optimization that we can calculate
Cortical columns learn through feedback loops
Game theory models how the neural groups
Align their gradients without central command
Distributed units working hand in hand
[Chorus]
Potential functions guide the way
Players converge where gradients stay
Feedback alignment breaks the chain
Backward passes need not strain
Column dynamics, game theory's dance
Differential mechanics advance
Learning rules that nature knows
How intelligence truly grows
[Verse 2]
Balduzzi came in twenty-eighteen strong
N-player games where differentials belong
Mechanics of the gradient descent flow
Saddle points and eigenvalues show
When cortical networks face the learning task
Multiple objectives behind the mask
Each column plays against the rest
Cooperative competition at its best
[Chorus]
Potential functions guide the way
Players converge where gradients stay
Feedback alignment breaks the chain
Backward passes need not strain
Column dynamics, game theory's dance
Differential mechanics advance
Learning rules that nature knows
How intelligence truly grows
[Bridge]
Lillicrap showed us feedback doesn't need
Precise transpose weights to succeed
Random connections still align the flow
Biological plausibility starts to show
Cortical columns as computational games
Each one learning different aims
Yet somehow coordination emerges clean
From this distributed learning machine
[Verse 3]
Potential games have special symmetry
What helps one player helps the harmony
No conflicting gradients tear apart
The system's unified learning heart
Cortical columns mirror this design
Independent yet their goals align
Nature's algorithm runs so deep
Game theoretic promises to keep
[Verse 4]
Continuous-time dynamics show the path
Where neural computation does its math
Local updates serve the global good
Self-organizing as evolution would
Each minicolumn seeks its payoff high
While system-wide improvements multiply
No central planner guides the way
Just local rules in neural play
[Chorus]
Potential functions guide the way
Players converge where gradients stay
Feedback alignment breaks the chain
Backward passes need not strain
Column dynamics, game theory's dance
Differential mechanics advance
Learning rules that nature knows
How intelligence truly grows
[Outro]
From Monderer's potential to feedback's grace
Mathematics finds biology's trace
Columns computing in concert sublime
Game theory's rhythm keeps perfect time