HYBRID CENTER-SURROUND CONTRAST GRID RESULTS
======================================================================

EXPERIMENTAL DESIGN:
- CRF diameter: 5 pixels (center)
- eCRF diameter: 37 pixels (surround)
- Spatial frequency: 40.0 cycles/image (fixed)
- Center contrasts tested: [0.25, 0.5, 0.75, 1.0]
- Surround contrasts tested: [0.25, 0.5, 0.75, 1.0]
- Total conditions: 4 × 4 = 16
- Decoder contrast: 1.0 (100%, trained ONCE)

PROTOCOL:
1. Train decoder ONCE on 100% contrast CRF-only gratings
2. For each (center_contrast, surround_contrast) pair:
   a. Generate hybrid stimuli (center at center_contrast, surround at surround_contrast)
   b. Run multi-orientation optimization (6 orientations: [0, 30, 60, 90, 120, 150])
   c. Rotate and average to get orientation-invariant structure
   d. Save visualizations with both mean and max channel aggregation

KEY FINDINGS:

MEAN AGGREGATION:

Excitatory Adjustment Strength:
[[6.51269220e-05 6.47864144e-05 1.35433263e-04 1.18366806e-04]
 [6.46188273e-05 8.12446888e-05 1.40131044e-04 8.27938638e-05]
 [6.66776614e-05 6.10479474e-05 7.42294214e-05 6.74536132e-05]
 [8.36108957e-05 6.26657420e-05 7.75634471e-05 7.79443799e-05]]

Inhibitory Adjustment Strength:
[[0.00101608 0.00183674 0.00272228 0.00249173]
 [0.00125585 0.00191737 0.0027818  0.00257297]
 [0.00040849 0.00117691 0.00216338 0.00260384]
 [0.00025831 0.00127918 0.001968   0.00246693]]

E-I Balance:
[[-0.0010268  -0.00182645 -0.00260715 -0.00238942]
 [-0.00127491 -0.00188746 -0.0026606  -0.00253219]
 [-0.00040018 -0.0011797  -0.00218447 -0.00259862]
 [-0.00028818 -0.00127756 -0.00197166 -0.00247464]]

MAX AGGREGATION:

Excitatory Adjustment Strength:
[[0.00075596 0.00076598 0.00090926 0.00089322]
 [0.00071581 0.00082101 0.00098863 0.00089942]
 [0.00075497 0.00077526 0.00084908 0.00080702]
 [0.00077159 0.00075251 0.00085732 0.00090013]]

Inhibitory Adjustment Strength:
[[0.00183065 0.00268016 0.00368713 0.00340051]
 [0.00211727 0.00284197 0.00375919 0.00364316]
 [0.00122037 0.00217648 0.00332258 0.00372048]
 [0.00115652 0.00228468 0.00301593 0.00357704]]

E-I Balance:
[[-0.0010747  -0.00191418 -0.00277787 -0.00250729]
 [-0.00140146 -0.00202096 -0.00277056 -0.00274374]
 [-0.0004654  -0.00140123 -0.00247349 -0.00291346]
 [-0.00038493 -0.00153217 -0.0021586  -0.0026769 ]]

BOTH AGGREGATIONS:

Final Loss (lower = better rescue):
[[3.93181288e+01 3.29466195e+06 1.71347029e+00 3.59198616e+00]
 [1.04406864e+01 3.08611869e+06 2.84487378e+01 5.39136055e+01]
 [1.83782914e+04 1.38843494e+04 1.25141391e+01 1.35317895e+00]
 [7.81374393e+07 1.91957822e+00 1.55377399e+01 3.91021994e+00]]

INTERPRETATION:
This experiment uses hybrid stimuli where the center (CRF) and surround (eCRF)
have different contrasts. The decoder is trained ONCE on 100% contrast CRF-only
gratings, ensuring consistent baseline. This design reveals:

1. How center-surround contrast interactions modulate recurrent compensation
2. Whether the recurrent mechanism adapts to contrast mismatches
3. Spatial structure of E/I adjustments across different contrast ratios

Patterns to look for:
- Diagonal (matched contrast): How does overall contrast affect recurrence?
- Above diagonal (surround > center): Does surround dominate?
- Below diagonal (center > surround): Does center drive compensation?
- E/I balance shifts: How does contrast ratio affect excitation vs inhibition?

Visualization Notes:
- Mean aggregation: Averages across all channels (more holistic view)
- Max aggregation: Takes maximum across channels (highlights strongest units)
- Expected pattern: Excitatory adjustments higher near center, inhibitory higher in periphery