From a62cf4d2a99b4a7985c61b2a7feb92a82a8218b7 Mon Sep 17 00:00:00 2001 From: Yuren Hao Date: Sat, 1 Aug 2026 14:10:03 -0500 Subject: World Alignment: unpaired cross-modal correspondence by relational identifiability Method: scene states are sets of part states; relation fields are built within each modality and are invariant to how each side labels its own features; the cross-modal bridge is a coupling searched under an energy that is a closed-form functional of one matrix; solving is spectral initialisation followed by exact local refinement. Evidence: in a procedurally generated closed world, blind recovery of a hidden image-caption correspondence reaches 95.3% at 256 scenes against 0.39% chance, and the recovered pairs transfer to 200 held-out scenes at 93.0% exact retrieval with random-pair and shuffled-image controls at or near chance. Cross-modal value correspondence is derived from disjoint corpora rather than declared. On Visual Genome the field correlation reaches 0.656 against the 0.9 that polynomial recovery needs, with the deficit attributed away from segmentation and discretisation. Protocol: no image-text pair enters any objective, optimiser, initialisation, or model selection; hidden pairs score orderings only. Co-Authored-By: Claude --- .../vg512_region_closed_seed2_scalar.json | 1160 ++++++++++++++++++++ 1 file changed, 1160 insertions(+) create mode 100644 artifacts/manifold_gate/vg512_region_closed_seed2_scalar.json (limited to 'artifacts/manifold_gate/vg512_region_closed_seed2_scalar.json') diff --git a/artifacts/manifold_gate/vg512_region_closed_seed2_scalar.json b/artifacts/manifold_gate/vg512_region_closed_seed2_scalar.json new file mode 100644 index 0000000..dd38278 --- /dev/null +++ b/artifacts/manifold_gate/vg512_region_closed_seed2_scalar.json @@ -0,0 +1,1160 @@ +{ + "protocol": "Assignments permute real frozen text states; hidden pairs are used only to place the true assignment in the ranking. The energy terms are the cross-modal relation MSE and conditional KL; the language-only terms of the falsified free-particle energy are permutation-invariant on this space.", + "meta": { + "dataset": "visual_genome_5k", + "tier": "region_closed", + "samples": 512, + "subset_seed": 2, + "bundle_channels": false + }, + "args": { + "dataset": "vg", + "text_orbits": "artifacts/text_orbits_qwen0p5b.pt", + "text_mode": "orbit_mean", + "vg_vision": "artifacts/vg_5k/vision_features.pt", + "vg_text": "artifacts/vg_5k/text_features.pt", + "vg_ground_truth": "artifacts/vg_5k/ground_truth.private.jsonl", + "vg_bundle_channels": false, + "split": "test", + "samples": 512, + "subset_seed": 2, + "random_perms": 1000, + "derangement_samples": 200, + "descent_restarts": 2, + "descent_max_steps": 200000, + "descent_objective": "m30_total", + "descent_verify_top": 64, + "device": "cpu", + "seed": 20260729, + "output": "artifacts/manifold_gate/vg512_region_closed_seed2_scalar.json", + "trajectory_output": null + }, + "channel_names": [ + "mean" + ], + "gate_a_global_ranking": { + "true": { + "mse": 1.5497406845217934, + "mse_channels": [ + 1.5497406845217934 + ], + "mse_channel_mean": 1.5497406845217934, + "conditional_kl": 6.289259164714802, + "m30_total": 4.357333201986547 + }, + "similarity_shift": { + "mse": 1.8006471777633417, + "mse_channels": [ + 1.8006471777633417 + ], + "mse_channel_mean": 1.8006471777633417, + "conditional_kl": 6.507446142390554, + "m30_total": 4.902783584004794 + }, + "random": { + "mse": { + "mean": 1.9997849464416504, + "std": 0.05623829737305641, + "min": 1.8321346044540405, + "true_z": 8.00245189666748, + "true_rank_among_random": 0, + "count": 1000 + }, + "mse_channel_mean": { + "mean": 1.9997849464416504, + "std": 0.05623829737305641, + "min": 1.8321346044540405, + "true_z": 8.00245189666748, + "true_rank_among_random": 0, + "count": 1000 + }, + "conditional_kl": { + "mean": 6.495845794677734, + "std": 0.011792183853685856, + "min": 6.460770606994629, + "true_z": 17.518962860107422, + "true_rank_among_random": 0, + "count": 1000 + }, + "m30_total": { + "mean": 5.298739433288574, + "std": 0.11354231834411621, + "min": 4.958786487579346, + "true_z": 8.291236877441406, + "true_rank_among_random": 0, + "count": 1000 + } + } + }, + "gate_b_transpositions": { + "pairs": 130816, + "improving_pairs": 25166, + "improving_fraction": 0.19237707925636008, + "delta_mean": 0.0024890511109841645, + "delta_min": -0.06585485097052292, + "identity_is_local_min_mse": false, + "worst_improving_swaps": [ + { + "pair": [ + 259, + 434 + ], + "delta_mse": -0.06585485097052292, + "text_cosine": 0.9545299262745018, + "visual_cosine": 0.0657274536235636 + }, + { + "pair": [ + 122, + 434 + ], + "delta_mse": -0.06478267821122188, + "text_cosine": 0.9492532393176104, + "visual_cosine": 0.244871262159981 + }, + { + "pair": [ + 281, + 434 + ], + "delta_mse": -0.0635609983077372, + "text_cosine": 0.9483341406152255, + "visual_cosine": 0.22140571540847073 + }, + { + "pair": [ + 55, + 434 + ], + "delta_mse": -0.06209616793457745, + "text_cosine": 0.9465071763423647, + "visual_cosine": 0.0908466488471182 + }, + { + "pair": [ + 57, + 119 + ], + "delta_mse": -0.061492540375367236, + "text_cosine": 0.9123714371950096, + "visual_cosine": 0.062815058902934 + }, + { + "pair": [ + 434, + 503 + ], + "delta_mse": -0.06070523150329002, + "text_cosine": 0.9474141389533427, + "visual_cosine": 0.15946363248958148 + }, + { + "pair": [ + 24, + 434 + ], + "delta_mse": -0.05943472596230676, + "text_cosine": 0.9405513476037503, + "visual_cosine": 0.03597514188254883 + }, + { + "pair": [ + 119, + 434 + ], + "delta_mse": -0.05930419821940719, + "text_cosine": 0.9296708688195825, + "visual_cosine": -0.009839870729349807 + }, + { + "pair": [ + 255, + 434 + ], + "delta_mse": -0.05900164234442869, + "text_cosine": 0.9510532116396718, + "visual_cosine": 0.23098643997959836 + }, + { + "pair": [ + 144, + 434 + ], + "delta_mse": -0.057927929782469294, + "text_cosine": 0.9446349054252449, + "visual_cosine": 0.11686968884530634 + }, + { + "pair": [ + 209, + 434 + ], + "delta_mse": -0.057585626577308105, + "text_cosine": 0.9499004442756396, + "visual_cosine": 0.14278533731284068 + }, + { + "pair": [ + 337, + 434 + ], + "delta_mse": -0.05750346364655796, + "text_cosine": 0.9465816343599123, + "visual_cosine": 0.09216829164950303 + }, + { + "pair": [ + 295, + 434 + ], + "delta_mse": -0.05658425941401525, + "text_cosine": 0.944979460666805, + "visual_cosine": 0.12446259682681478 + }, + { + "pair": [ + 296, + 434 + ], + "delta_mse": -0.05461165478348438, + "text_cosine": 0.9330566873361857, + "visual_cosine": 0.09130649378323508 + }, + { + "pair": [ + 362, + 434 + ], + "delta_mse": -0.05452655454753379, + "text_cosine": 0.9465066969394126, + "visual_cosine": 0.05843326134511321 + }, + { + "pair": [ + 378, + 434 + ], + "delta_mse": -0.05349981938913902, + "text_cosine": 0.9349592861513891, + "visual_cosine": 0.17400766885797223 + }, + { + "pair": [ + 41, + 434 + ], + "delta_mse": -0.05308207424783738, + "text_cosine": 0.9454407512436758, + "visual_cosine": 0.09433780902667131 + }, + { + "pair": [ + 150, + 434 + ], + "delta_mse": -0.05211335967944682, + "text_cosine": 0.9339074527689419, + "visual_cosine": 0.02372383600957501 + }, + { + "pair": [ + 7, + 434 + ], + "delta_mse": -0.051900037887693984, + "text_cosine": 0.9477017612730887, + "visual_cosine": 0.17797706935205507 + }, + { + "pair": [ + 434, + 501 + ], + "delta_mse": -0.05110361325468303, + "text_cosine": 0.9405716042935885, + "visual_cosine": 0.18326231978182317 + } + ] + }, + "derangement_curve": [ + { + "k": 2, + "delta_mean": 0.0022332703229039907, + "delta_std": 0.004554439801722765, + "improving_fraction": 0.18 + }, + { + "k": 4, + "delta_mean": 0.004869141150265932, + "delta_std": 0.009268179535865784, + "improving_fraction": 0.145 + }, + { + "k": 8, + "delta_mean": 0.00990293174982071, + "delta_std": 0.015287930145859718, + "improving_fraction": 0.135 + }, + { + "k": 16, + "delta_mean": 0.01916361041367054, + "delta_std": 0.017353815957903862, + "improving_fraction": 0.065 + }, + { + "k": 32, + "delta_mean": 0.04231706261634827, + "delta_std": 0.028036441653966904, + "improving_fraction": 0.03 + }, + { + "k": 64, + "delta_mean": 0.07876143604516983, + "delta_std": 0.03174426406621933, + "improving_fraction": 0.0 + }, + { + "k": 128, + "delta_mean": 0.15009598433971405, + "delta_std": 0.04559198394417763, + "improving_fraction": 0.0 + }, + { + "k": 256, + "delta_mean": 0.27420204877853394, + "delta_std": 0.060302022844552994, + "improving_fraction": 0.0 + }, + { + "k": 512, + "delta_mean": 0.455189049243927, + "delta_std": 0.055225081741809845, + "improving_fraction": 0.0 + } + ], + "gate_c_descent_from_true": { + "start_accuracy": 1.0, + "final_accuracy": 0.029296875, + "final_objective": 2.575195924817873, + "final_energy": { + "mse": 0.6694562603367474, + "mse_channels": [ + 0.6694562603367474 + ], + "mse_channel_mean": 0.6694562603367474, + "conditional_kl": 6.181417020721889, + "m30_total": 2.575195924817873 + }, + "accepted_moves": 731, + "moved_fraction": 0.970703125, + "mse_local_min_certificate": false, + "trajectory": [ + { + "step": 0, + "objective": 4.224573665837802, + "accuracy": 0.99609375 + }, + { + "step": 50, + "objective": 3.308524779345391, + "accuracy": 0.80078125 + }, + { + "step": 100, + "objective": 3.040780975048601, + "accuracy": 0.615234375 + }, + { + "step": 150, + "objective": 2.913406852175558, + "accuracy": 0.4609375 + }, + { + "step": 200, + "objective": 2.833824502843334, + "accuracy": 0.3515625 + }, + { + "step": 250, + "objective": 2.7682729348053003, + "accuracy": 0.271484375 + }, + { + "step": 300, + "objective": 2.7199335217138554, + "accuracy": 0.20703125 + }, + { + "step": 350, + "objective": 2.679727083940402, + "accuracy": 0.154296875 + }, + { + "step": 400, + "objective": 2.6507798067781785, + "accuracy": 0.1015625 + }, + { + "step": 450, + "objective": 2.630545998321178, + "accuracy": 0.087890625 + }, + { + "step": 500, + "objective": 2.614648256102141, + "accuracy": 0.078125 + }, + { + "step": 550, + "objective": 2.601285969571643, + "accuracy": 0.05859375 + }, + { + "step": 600, + "objective": 2.5909114674673925, + "accuracy": 0.04296875 + }, + { + "step": 650, + "objective": 2.5826889323922813, + "accuracy": 0.0390625 + }, + { + "step": 700, + "objective": 2.5776051365858854, + "accuracy": 0.03125 + } + ], + "final_permutation": [ + 374, + 239, + 424, + 246, + 494, + 195, + 175, + 486, + 394, + 148, + 183, + 269, + 146, + 384, + 185, + 110, + 111, + 336, + 39, + 470, + 231, + 357, + 119, + 182, + 191, + 301, + 94, + 92, + 99, + 262, + 317, + 313, + 441, + 362, + 126, + 213, + 234, + 155, + 312, + 17, + 227, + 285, + 338, + 47, + 319, + 405, + 278, + 447, + 448, + 121, + 18, + 16, + 350, + 267, + 10, + 74, + 29, + 498, + 20, + 478, + 243, + 426, + 62, + 225, + 432, + 78, + 109, + 260, + 33, + 69, + 207, + 464, + 453, + 304, + 400, + 75, + 204, + 472, + 169, + 219, + 368, + 15, + 114, + 150, + 4, + 85, + 455, + 133, + 258, + 397, + 401, + 73, + 457, + 331, + 137, + 50, + 342, + 476, + 229, + 189, + 445, + 101, + 113, + 228, + 323, + 508, + 396, + 496, + 435, + 471, + 158, + 352, + 487, + 473, + 314, + 167, + 428, + 70, + 253, + 57, + 235, + 91, + 434, + 14, + 143, + 347, + 308, + 28, + 360, + 230, + 13, + 458, + 181, + 355, + 327, + 294, + 466, + 34, + 201, + 232, + 64, + 42, + 477, + 48, + 96, + 462, + 419, + 437, + 361, + 333, + 83, + 245, + 79, + 165, + 36, + 222, + 134, + 386, + 404, + 353, + 311, + 40, + 506, + 422, + 463, + 339, + 80, + 0, + 194, + 59, + 170, + 123, + 281, + 192, + 233, + 180, + 11, + 381, + 67, + 290, + 295, + 84, + 296, + 186, + 151, + 398, + 370, + 414, + 491, + 127, + 184, + 53, + 377, + 340, + 12, + 403, + 118, + 454, + 383, + 485, + 164, + 483, + 90, + 159, + 76, + 324, + 98, + 495, + 208, + 3, + 256, + 359, + 149, + 21, + 277, + 452, + 330, + 54, + 372, + 310, + 276, + 480, + 297, + 35, + 399, + 264, + 365, + 87, + 160, + 102, + 58, + 9, + 38, + 107, + 202, + 442, + 469, + 124, + 460, + 420, + 511, + 241, + 168, + 190, + 95, + 291, + 344, + 105, + 6, + 249, + 259, + 373, + 223, + 490, + 272, + 179, + 325, + 199, + 147, + 171, + 131, + 242, + 431, + 348, + 298, + 275, + 505, + 376, + 423, + 366, + 292, + 510, + 163, + 218, + 268, + 321, + 220, + 153, + 8, + 178, + 389, + 174, + 282, + 104, + 271, + 391, + 279, + 257, + 364, + 273, + 467, + 379, + 407, + 489, + 367, + 89, + 51, + 214, + 142, + 30, + 390, + 93, + 198, + 65, + 493, + 371, + 369, + 343, + 203, + 22, + 41, + 380, + 322, + 356, + 307, + 300, + 128, + 415, + 468, + 106, + 320, + 474, + 430, + 427, + 196, + 224, + 197, + 52, + 238, + 438, + 450, + 68, + 444, + 475, + 63, + 408, + 157, + 86, + 283, + 23, + 24, + 440, + 56, + 141, + 326, + 456, + 254, + 499, + 289, + 507, + 436, + 210, + 55, + 45, + 497, + 413, + 417, + 221, + 410, + 139, + 443, + 130, + 387, + 363, + 502, + 501, + 345, + 173, + 140, + 354, + 200, + 329, + 509, + 215, + 465, + 5, + 32, + 206, + 116, + 425, + 335, + 280, + 481, + 46, + 77, + 82, + 212, + 270, + 19, + 1, + 138, + 193, + 88, + 255, + 334, + 27, + 293, + 492, + 451, + 286, + 37, + 284, + 211, + 188, + 266, + 61, + 378, + 136, + 209, + 388, + 60, + 226, + 117, + 161, + 217, + 393, + 2, + 303, + 392, + 144, + 288, + 358, + 250, + 237, + 261, + 306, + 318, + 346, + 316, + 244, + 132, + 418, + 484, + 43, + 421, + 503, + 488, + 287, + 406, + 412, + 152, + 31, + 100, + 162, + 263, + 251, + 351, + 247, + 166, + 449, + 479, + 177, + 305, + 236, + 461, + 337, + 315, + 429, + 7, + 309, + 328, + 154, + 341, + 97, + 71, + 332, + 205, + 411, + 274, + 248, + 25, + 187, + 172, + 44, + 125, + 265, + 108, + 26, + 433, + 439, + 72, + 382, + 252, + 115, + 302, + 395, + 145, + 446, + 482, + 122, + 385, + 416, + 81, + 349, + 459, + 156, + 120, + 299, + 375, + 49, + 500, + 103, + 216, + 409, + 504, + 135, + 66, + 402, + 129, + 112, + 176, + 240 + ] + }, + "gate_c_descent_from_random": [ + { + "start_accuracy": 0.001953125, + "final_accuracy": 0.0, + "final_objective": 2.5423502964013167, + "final_energy": { + "mse": 0.6531657976107715, + "mse_channels": [ + 0.6531657976107715 + ], + "mse_channel_mean": 0.6531657976107715, + "conditional_kl": 6.180093505898869, + "m30_total": 2.5423502964013167 + }, + "accepted_moves": 840, + "moved_fraction": 0.99609375, + "mse_local_min_certificate": false, + "trajectory": [ + { + "step": 0, + "objective": 5.069657387472825, + "accuracy": 0.001953125 + }, + { + "step": 50, + "objective": 3.7466026030460986, + "accuracy": 0.0 + }, + { + "step": 100, + "objective": 3.3386028087856516, + "accuracy": 0.0 + }, + { + "step": 150, + "objective": 3.1144443091661405, + "accuracy": 0.0 + }, + { + "step": 200, + "objective": 2.954518786860014, + "accuracy": 0.0 + }, + { + "step": 250, + "objective": 2.842810467783562, + "accuracy": 0.0 + }, + { + "step": 300, + "objective": 2.7677407461739643, + "accuracy": 0.0 + }, + { + "step": 350, + "objective": 2.711587753191367, + "accuracy": 0.0 + }, + { + "step": 400, + "objective": 2.668306972476886, + "accuracy": 0.001953125 + }, + { + "step": 450, + "objective": 2.633034195217526, + "accuracy": 0.001953125 + }, + { + "step": 500, + "objective": 2.6071695436130695, + "accuracy": 0.0 + }, + { + "step": 550, + "objective": 2.5882842780063644, + "accuracy": 0.0 + }, + { + "step": 600, + "objective": 2.575576077410493, + "accuracy": 0.0 + }, + { + "step": 650, + "objective": 2.5647948728678704, + "accuracy": 0.0 + }, + { + "step": 700, + "objective": 2.5569371136211876, + "accuracy": 0.0 + }, + { + "step": 750, + "objective": 2.5508814172542884, + "accuracy": 0.0 + }, + { + "step": 800, + "objective": 2.545639359132948, + "accuracy": 0.0 + } + ] + }, + { + "start_accuracy": 0.0, + "final_accuracy": 0.00390625, + "final_objective": 2.5897492567343394, + "final_energy": { + "mse": 0.6768999448958204, + "mse_channels": [ + 0.6768999448958204 + ], + "mse_channel_mean": 0.6768999448958204, + "conditional_kl": 6.179746834713493, + "m30_total": 2.5897492567343394 + }, + "accepted_moves": 808, + "moved_fraction": 0.994140625, + "mse_local_min_certificate": false, + "trajectory": [ + { + "step": 0, + "objective": 4.96258322594762, + "accuracy": 0.0 + }, + { + "step": 50, + "objective": 3.7095089944982753, + "accuracy": 0.0 + }, + { + "step": 100, + "objective": 3.3330897062212768, + "accuracy": 0.001953125 + }, + { + "step": 150, + "objective": 3.1494493479473418, + "accuracy": 0.001953125 + }, + { + "step": 200, + "objective": 3.023983655083515, + "accuracy": 0.001953125 + }, + { + "step": 250, + "objective": 2.914316036739912, + "accuracy": 0.0 + }, + { + "step": 300, + "objective": 2.8293032478004916, + "accuracy": 0.0 + }, + { + "step": 350, + "objective": 2.764819602821273, + "accuracy": 0.0 + }, + { + "step": 400, + "objective": 2.719758962329603, + "accuracy": 0.0 + }, + { + "step": 450, + "objective": 2.6850640481131327, + "accuracy": 0.001953125 + }, + { + "step": 500, + "objective": 2.65992290357804, + "accuracy": 0.001953125 + }, + { + "step": 550, + "objective": 2.6399968928068382, + "accuracy": 0.001953125 + }, + { + "step": 600, + "objective": 2.6233857191804297, + "accuracy": 0.001953125 + }, + { + "step": 650, + "objective": 2.6104496020458563, + "accuracy": 0.001953125 + }, + { + "step": 700, + "objective": 2.6003136601859076, + "accuracy": 0.001953125 + }, + { + "step": 750, + "objective": 2.5934369203239847, + "accuracy": 0.001953125 + }, + { + "step": 800, + "objective": 2.5898725083285195, + "accuracy": 0.001953125 + } + ] + } + ], + "verdict": { + "true_m30_total": 4.357333201986547, + "identity_is_local_min_mse": false, + "descent_from_true_stays": 0.029296875, + "on_manifold_counterfeit_found": true, + "best_random_descent_m30_total": 2.5423502964013167 + } +} \ No newline at end of file -- cgit v1.2.3