Convolutional bars experiment

Actual CPU training outputs, 18 October 2026. No pretrained weights. Each model trained for 1,000 updates on 128 procedural 16×16 grayscale bars. Evaluation uses 64 fixed noise vectors; the first four rows request vertical bars, the last four horizontal bars. Before/after use the same noise and labels.

This is not evidence of recognizable cat/dog generation. A simple orientation heuristic scores 64/64 classical and 61/64 hybrid outputs after training. The images still show limited location diversity and imperfect shapes. The score does not establish full distribution learning or a quantum advantage.

classical

before

after

hybrid

before

after

Reproduce

python benchmarks/image_learning.py

Recorded settings and results

{
  "scope": "Single-seed procedural bars; not pet-image validation or a quantum-advantage benchmark",
  "torch_version": "2.14.1+cpu",
  "device": "cpu",
  "threads": 1,
  "steps": 1000,
  "batch_size": 16,
  "samples": 128,
  "image_size": 16,
  "channels": 1,
  "latent_dim": 8,
  "hidden": 32,
  "learning_rate": 0.0002,
  "seed": 7,
  "data_seed": 29,
  "evaluation_seed": 101,
  "evaluation_samples": 64,
  "orientation_definition": "Vertical if variance of column means exceeds variance of row means, otherwise horizontal; ties count incorrect",
  "limitation": "Orientation alone does not establish realistic bars, location diversity, generalization or animal semantics; previews show limited location diversity",
  "models": [
    {
      "kind": "classical",
      "before": {
        "orientation_accuracy": 0.46875,
        "mean": 0.5038063526153564,
        "pixel_std": 0.005083031952381134,
        "orientation_contrast": 2.695317334655556e-06
      },
      "snapshots": [
        {
          "step": 200,
          "orientation_accuracy": 0.5,
          "mean": 0.13609468936920166,
          "pixel_std": 0.2498246729373932,
          "orientation_contrast": 0.0221474040299654
        },
        {
          "step": 400,
          "orientation_accuracy": 0.546875,
          "mean": 0.11157720535993576,
          "pixel_std": 0.2751159369945526,
          "orientation_contrast": 0.014023473486304283
        },
        {
          "step": 600,
          "orientation_accuracy": 0.609375,
          "mean": 0.12446727603673935,
          "pixel_std": 0.3008456528186798,
          "orientation_contrast": 0.015934886410832405
        },
        {
          "step": 800,
          "orientation_accuracy": 0.96875,
          "mean": 0.11770956963300705,
          "pixel_std": 0.29196691513061523,
          "orientation_contrast": 0.02967148832976818
        },
        {
          "step": 1000,
          "orientation_accuracy": 1.0,
          "mean": 0.13062536716461182,
          "pixel_std": 0.32457178831100464,
          "orientation_contrast": 0.051596660166978836
        }
      ],
      "after": {
        "orientation_accuracy": 1.0,
        "mean": 0.13062536716461182,
        "pixel_std": 0.32457178831100464,
        "orientation_contrast": 0.051596660166978836
      }
    },
    {
      "kind": "hybrid",
      "before": {
        "orientation_accuracy": 0.1875,
        "mean": 0.514197587966919,
        "pixel_std": 0.003048446262255311,
        "orientation_contrast": 5.581701998380595e-07
      },
      "snapshots": [
        {
          "step": 200,
          "orientation_accuracy": 0.5,
          "mean": 0.11994701623916626,
          "pixel_std": 0.1818249672651291,
          "orientation_contrast": 0.014576568268239498
        },
        {
          "step": 400,
          "orientation_accuracy": 0.640625,
          "mean": 0.11676280945539474,
          "pixel_std": 0.2691108286380768,
          "orientation_contrast": 0.0292088333517313
        },
        {
          "step": 600,
          "orientation_accuracy": 0.609375,
          "mean": 0.08491843193769455,
          "pixel_std": 0.25586211681365967,
          "orientation_contrast": 0.03203911334276199
        },
        {
          "step": 800,
          "orientation_accuracy": 0.828125,
          "mean": 0.1821681261062622,
          "pixel_std": 0.3639906346797943,
          "orientation_contrast": 0.0364963635802269
        },
        {
          "step": 1000,
          "orientation_accuracy": 0.953125,
          "mean": 0.10341374576091766,
          "pixel_std": 0.2859596908092499,
          "orientation_contrast": 0.01759444922208786
        }
      ],
      "after": {
        "orientation_accuracy": 0.953125,
        "mean": 0.10341374576091766,
        "pixel_std": 0.2859596908092499,
        "orientation_contrast": 0.01759444922208786
      }
    }
  ]
}