The benchmark that scores the correspondence

urdu acid jazz, afro-cuban jazz, hyper-afrikaner folk · 4:28

Listen on 93

Lyrics

[Verse 1]
In the lab where neurons fire and dance
Scientists built a scoring chance
Brain-Score measures what computers see
Against the circuits, you and me
Layer by layer, V1 through four
IT cortex keeping neural score
Schrimpf and DiCarlo found the way
To test if silicon thinks like clay

[Chorus]
Brain-Score, Brain-Score, counting every match
Model to biology, see what neurons catch
V1, V2, V4, IT in the stack
Brain-Score tells us which ones bring the magic back
Correspondence measured, number on the board
Higher scores mean artificial minds accord

[Verse 2]
Feed the same image to both sides
Primate vision and computer guides
Record the spikes from living cells
Compare with what the model tells
Hierarchical features climbing high
Convolution patterns multiply
Each synthetic layer must align
With nature's own visual design

[Chorus]
Brain-Score, Brain-Score, counting every match
Model to biology, see what neurons catch
V1, V2, V4, IT in the stack
Brain-Score tells us which ones bring the magic back
Correspondence measured, number on the board
Higher scores mean artificial minds accord

[Bridge]
Normalization keeps the signal clean
Recurrent loops enhance the scene
The models scoring at the peak
Mirror biology's technique
Load-bearing motifs rise above
The architectures neurons love

[Verse 3]
When the benchmark gives its final grade
Evidence of progress can be weighed
Artificial stacks that score the best
Have passed biology's toughest test
Module ten, method ten precise
Formal proof rolled like loaded dice
The cleanest measure we possess
Of visual processing success

[Verse 4]
Neural noise and trial-to-trial change
Brain-Score handles variance range
Bootstrapping confidence intervals tight
Statistical testing gets it right
Cross-validation splits the data clean
Best performing models rise supreme
Temporal dynamics, static views
Every correlation gives us clues

[Chorus]
Brain-Score, Brain-Score, counting every match
Model to biology, see what neurons catch
V1, V2, V4, IT in the stack
Brain-Score tells us which ones bring the magic back
Correspondence measured, number on the board
Higher scores mean artificial minds accord

[Chorus]
Brain-Score, Brain-Score, counting every match
Model to biology, see what neurons catch
V1, V2, V4, IT in the stack
Brain-Score tells us which ones bring the magic back
Correspondence measured, number on the board
Higher scores mean artificial minds accord

[Outro]
From pixels raw to recognition
Brain-Score judges the rendition
Benchmark singing nature's tune
Artificial dawn meets neural noon

← Where artificial and biological vision diverge (and why it matters) | The practical takeaway for building biologically-grounded vision →