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_visible_relations_seed2_scalar.json | 1145 ++++++++++++++++++++ 1 file changed, 1145 insertions(+) create mode 100644 artifacts/manifold_gate/vg512_visible_relations_seed2_scalar.json (limited to 'artifacts/manifold_gate/vg512_visible_relations_seed2_scalar.json') diff --git a/artifacts/manifold_gate/vg512_visible_relations_seed2_scalar.json b/artifacts/manifold_gate/vg512_visible_relations_seed2_scalar.json new file mode 100644 index 0000000..fb4aed3 --- /dev/null +++ b/artifacts/manifold_gate/vg512_visible_relations_seed2_scalar.json @@ -0,0 +1,1145 @@ +{ + "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": "visible_relations", + "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_tiers/text_features_visible_relations.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_visible_relations_seed2_scalar.json", + "trajectory_output": null + }, + "channel_names": [ + "mean" + ], + "gate_a_global_ranking": { + "true": { + "mse": 1.4155266393943742, + "mse_channels": [ + 1.4155266393943742 + ], + "mse_channel_mean": 1.4155266393943742, + "conditional_kl": 6.283051742321277, + "m30_total": 4.087663627253004 + }, + "similarity_shift": { + "mse": 1.7178209001574276, + "mse_channels": [ + 1.7178209001574276 + ], + "mse_channel_mean": 1.7178209001574276, + "conditional_kl": 6.482761757946626, + "m30_total": 4.732194151904181 + }, + "random": { + "mse": { + "mean": 1.999996304512024, + "std": 0.05232740193605423, + "min": 1.811036467552185, + "true_z": 11.169476509094238, + "true_rank_among_random": 0, + "count": 1000 + }, + "mse_channel_mean": { + "mean": 1.999996304512024, + "std": 0.05232740193605423, + "min": 1.811036467552185, + "true_z": 11.169476509094238, + "true_rank_among_random": 0, + "count": 1000 + }, + "conditional_kl": { + "mean": 6.48287296295166, + "std": 0.008014952763915062, + "min": 6.461653232574463, + "true_z": 24.931026458740234, + "true_rank_among_random": 0, + "count": 1000 + }, + "m30_total": { + "mean": 5.296567440032959, + "std": 0.10527576506137848, + "min": 4.920307159423828, + "true_z": 11.483210563659668, + "true_rank_among_random": 0, + "count": 1000 + } + } + }, + "gate_b_transpositions": { + "pairs": 130816, + "improving_pairs": 20486, + "improving_fraction": 0.15660163894324852, + "delta_mean": 0.0033729334376237236, + "delta_min": -0.03626393045316718, + "identity_is_local_min_mse": false, + "worst_improving_swaps": [ + { + "pair": [ + 255, + 434 + ], + "delta_mse": -0.03626393045316718, + "text_cosine": 0.9682047164868949, + "visual_cosine": 0.23098643997959836 + }, + { + "pair": [ + 337, + 434 + ], + "delta_mse": -0.03355996724213786, + "text_cosine": 0.9787182729960455, + "visual_cosine": 0.09216829164950303 + }, + { + "pair": [ + 136, + 255 + ], + "delta_mse": -0.03186520450336888, + "text_cosine": 0.9810521165181114, + "visual_cosine": 0.09676062127482941 + }, + { + "pair": [ + 136, + 337 + ], + "delta_mse": -0.03185896759204771, + "text_cosine": 0.9718823700262801, + "visual_cosine": 0.12016714134023178 + }, + { + "pair": [ + 24, + 49 + ], + "delta_mse": -0.031670533423651864, + "text_cosine": 0.9770026280954156, + "visual_cosine": 0.20870996876302955 + }, + { + "pair": [ + 385, + 434 + ], + "delta_mse": -0.031136255974333133, + "text_cosine": 0.9825694344173559, + "visual_cosine": 0.15032951289485877 + }, + { + "pair": [ + 153, + 259 + ], + "delta_mse": -0.03113005510234278, + "text_cosine": 0.9864639199413597, + "visual_cosine": 0.3147666924737696 + }, + { + "pair": [ + 259, + 434 + ], + "delta_mse": -0.03091067660158774, + "text_cosine": 0.9765995726747245, + "visual_cosine": 0.0657274536235636 + }, + { + "pair": [ + 49, + 255 + ], + "delta_mse": -0.0304886762393401, + "text_cosine": 0.971545143801546, + "visual_cosine": 0.0859699333139035 + }, + { + "pair": [ + 74, + 434 + ], + "delta_mse": -0.03029578239495467, + "text_cosine": 0.9789578631785946, + "visual_cosine": 0.1847280711072476 + }, + { + "pair": [ + 153, + 255 + ], + "delta_mse": -0.029289924037326963, + "text_cosine": 0.9629528287834702, + "visual_cosine": 0.10592284061575978 + }, + { + "pair": [ + 24, + 434 + ], + "delta_mse": -0.02925469431987961, + "text_cosine": 0.9643931233591037, + "visual_cosine": 0.03597514188254883 + }, + { + "pair": [ + 49, + 337 + ], + "delta_mse": -0.028488895350703248, + "text_cosine": 0.9693913746394484, + "visual_cosine": 0.25117291884033527 + }, + { + "pair": [ + 136, + 385 + ], + "delta_mse": -0.028453931293649412, + "text_cosine": 0.9691895189562915, + "visual_cosine": 0.18626422720007163 + }, + { + "pair": [ + 434, + 503 + ], + "delta_mse": -0.02800244496275287, + "text_cosine": 0.9711292990774341, + "visual_cosine": 0.15946363248958148 + }, + { + "pair": [ + 202, + 434 + ], + "delta_mse": -0.027976409193869147, + "text_cosine": 0.9764741122827478, + "visual_cosine": 0.19168784806462472 + }, + { + "pair": [ + 295, + 434 + ], + "delta_mse": -0.027734832984260115, + "text_cosine": 0.9763151264633181, + "visual_cosine": 0.12446259682681478 + }, + { + "pair": [ + 434, + 501 + ], + "delta_mse": -0.027487210615992606, + "text_cosine": 0.9667190377640694, + "visual_cosine": 0.18326231978182317 + }, + { + "pair": [ + 434, + 450 + ], + "delta_mse": -0.02745031647190968, + "text_cosine": 0.9853516113751454, + "visual_cosine": 0.2692349991889342 + }, + { + "pair": [ + 24, + 153 + ], + "delta_mse": -0.02740684506199136, + "text_cosine": 0.9635098984508902, + "visual_cosine": -0.014965486188357201 + } + ] + }, + "derangement_curve": [ + { + "k": 2, + "delta_mean": 0.0036709720734506845, + "delta_std": 0.00494008045643568, + "improving_fraction": 0.12 + }, + { + "k": 4, + "delta_mean": 0.006497486960142851, + "delta_std": 0.007768876850605011, + "improving_fraction": 0.12 + }, + { + "k": 8, + "delta_mean": 0.01287675742059946, + "delta_std": 0.010565615259110928, + "improving_fraction": 0.1 + }, + { + "k": 16, + "delta_mean": 0.025680432096123695, + "delta_std": 0.015553798526525497, + "improving_fraction": 0.025 + }, + { + "k": 32, + "delta_mean": 0.055412475019693375, + "delta_std": 0.02287478931248188, + "improving_fraction": 0.005 + }, + { + "k": 64, + "delta_mean": 0.10384117066860199, + "delta_std": 0.027940962463617325, + "improving_fraction": 0.0 + }, + { + "k": 128, + "delta_mean": 0.1984347105026245, + "delta_std": 0.040574394166469574, + "improving_fraction": 0.0 + }, + { + "k": 256, + "delta_mean": 0.3611311614513397, + "delta_std": 0.053706053644418716, + "improving_fraction": 0.0 + }, + { + "k": 512, + "delta_mean": 0.5867682099342346, + "delta_std": 0.05643010139465332, + "improving_fraction": 0.0 + } + ], + "gate_c_descent_from_true": { + "start_accuracy": 1.0, + "final_accuracy": 0.041015625, + "final_objective": 2.334395868764019, + "final_energy": { + "mse": 0.5461154577027073, + "mse_channels": [ + 0.5461154577027073 + ], + "mse_channel_mean": 0.5461154577027073, + "conditional_kl": 6.2108247667930225, + "m30_total": 2.334395868764019 + }, + "accepted_moves": 664, + "moved_fraction": 0.958984375, + "mse_local_min_certificate": false, + "trajectory": [ + { + "step": 0, + "objective": 4.015279390559284, + "accuracy": 0.99609375 + }, + { + "step": 50, + "objective": 3.0377498605475575, + "accuracy": 0.80078125 + }, + { + "step": 100, + "objective": 2.7726795950050644, + "accuracy": 0.6171875 + }, + { + "step": 150, + "objective": 2.63222634722375, + "accuracy": 0.47265625 + }, + { + "step": 200, + "objective": 2.5441657954792802, + "accuracy": 0.359375 + }, + { + "step": 250, + "objective": 2.483417156614065, + "accuracy": 0.263671875 + }, + { + "step": 300, + "objective": 2.442043096723371, + "accuracy": 0.193359375 + }, + { + "step": 350, + "objective": 2.4096594441873975, + "accuracy": 0.1484375 + }, + { + "step": 400, + "objective": 2.3868558201705925, + "accuracy": 0.119140625 + }, + { + "step": 450, + "objective": 2.3683927748278224, + "accuracy": 0.087890625 + }, + { + "step": 500, + "objective": 2.3555280046546736, + "accuracy": 0.0703125 + }, + { + "step": 550, + "objective": 2.3456709249215595, + "accuracy": 0.0625 + }, + { + "step": 600, + "objective": 2.339726516769034, + "accuracy": 0.052734375 + }, + { + "step": 650, + "objective": 2.335199328246582, + "accuracy": 0.041015625 + } + ], + "final_permutation": [ + 87, + 142, + 203, + 462, + 195, + 324, + 259, + 227, + 241, + 10, + 330, + 135, + 365, + 62, + 222, + 181, + 55, + 187, + 39, + 306, + 410, + 457, + 486, + 253, + 49, + 488, + 147, + 92, + 173, + 268, + 379, + 245, + 423, + 414, + 443, + 418, + 333, + 155, + 407, + 219, + 40, + 179, + 356, + 296, + 159, + 210, + 163, + 50, + 487, + 8, + 79, + 352, + 247, + 128, + 120, + 66, + 340, + 133, + 67, + 395, + 28, + 275, + 25, + 63, + 478, + 169, + 323, + 511, + 188, + 345, + 70, + 150, + 58, + 84, + 207, + 255, + 14, + 114, + 90, + 338, + 201, + 59, + 107, + 286, + 118, + 381, + 292, + 494, + 37, + 303, + 184, + 157, + 223, + 186, + 124, + 22, + 257, + 432, + 298, + 469, + 477, + 2, + 194, + 400, + 250, + 405, + 310, + 396, + 499, + 109, + 350, + 111, + 29, + 34, + 314, + 229, + 215, + 271, + 454, + 57, + 211, + 265, + 3, + 123, + 373, + 125, + 101, + 6, + 261, + 217, + 331, + 318, + 290, + 23, + 75, + 74, + 284, + 232, + 358, + 412, + 193, + 447, + 456, + 137, + 209, + 161, + 89, + 11, + 243, + 32, + 83, + 437, + 258, + 165, + 427, + 404, + 61, + 30, + 421, + 216, + 47, + 475, + 113, + 94, + 491, + 380, + 304, + 347, + 342, + 348, + 139, + 174, + 9, + 99, + 335, + 192, + 13, + 501, + 378, + 425, + 33, + 283, + 131, + 141, + 213, + 489, + 18, + 431, + 317, + 445, + 367, + 102, + 282, + 212, + 305, + 91, + 52, + 16, + 360, + 41, + 235, + 152, + 362, + 337, + 164, + 51, + 291, + 483, + 171, + 497, + 391, + 248, + 112, + 146, + 392, + 149, + 97, + 451, + 218, + 450, + 220, + 382, + 479, + 389, + 503, + 353, + 461, + 326, + 384, + 273, + 100, + 199, + 398, + 439, + 233, + 200, + 397, + 332, + 313, + 480, + 240, + 420, + 312, + 244, + 238, + 321, + 205, + 444, + 54, + 506, + 12, + 371, + 119, + 438, + 121, + 166, + 490, + 202, + 168, + 153, + 260, + 249, + 225, + 42, + 264, + 38, + 471, + 182, + 433, + 467, + 376, + 361, + 302, + 78, + 308, + 473, + 237, + 189, + 278, + 399, + 300, + 208, + 224, + 1, + 485, + 45, + 221, + 81, + 419, + 251, + 349, + 160, + 484, + 76, + 466, + 144, + 136, + 190, + 453, + 15, + 505, + 401, + 468, + 311, + 460, + 493, + 498, + 307, + 126, + 246, + 336, + 185, + 44, + 214, + 228, + 143, + 455, + 266, + 346, + 5, + 320, + 463, + 198, + 322, + 344, + 394, + 436, + 226, + 351, + 328, + 204, + 138, + 510, + 500, + 288, + 82, + 105, + 256, + 385, + 482, + 73, + 440, + 43, + 68, + 459, + 496, + 285, + 24, + 60, + 267, + 430, + 383, + 197, + 108, + 95, + 280, + 104, + 377, + 495, + 64, + 48, + 297, + 117, + 363, + 178, + 110, + 357, + 472, + 53, + 309, + 370, + 36, + 20, + 424, + 374, + 509, + 167, + 366, + 434, + 130, + 387, + 507, + 464, + 422, + 88, + 270, + 274, + 408, + 411, + 470, + 390, + 316, + 27, + 269, + 325, + 339, + 122, + 183, + 72, + 279, + 103, + 85, + 277, + 96, + 329, + 191, + 481, + 343, + 175, + 239, + 69, + 17, + 148, + 196, + 127, + 415, + 236, + 65, + 206, + 334, + 359, + 287, + 21, + 293, + 504, + 134, + 242, + 368, + 386, + 115, + 476, + 35, + 86, + 234, + 176, + 272, + 413, + 151, + 132, + 262, + 492, + 301, + 263, + 319, + 129, + 0, + 276, + 429, + 446, + 449, + 230, + 428, + 46, + 402, + 7, + 116, + 315, + 80, + 299, + 327, + 355, + 502, + 341, + 474, + 364, + 388, + 106, + 441, + 140, + 417, + 403, + 458, + 172, + 254, + 448, + 56, + 154, + 289, + 416, + 170, + 180, + 295, + 252, + 77, + 393, + 93, + 145, + 156, + 4, + 158, + 452, + 508, + 294, + 426, + 465, + 281, + 372, + 406, + 354, + 162, + 375, + 98, + 177, + 409, + 26, + 31, + 435, + 442, + 231, + 369, + 19, + 71 + ] + }, + "gate_c_descent_from_random": [ + { + "start_accuracy": 0.001953125, + "final_accuracy": 0.005859375, + "final_objective": 2.3229403254004986, + "final_energy": { + "mse": 0.5388995375823609, + "mse_channels": [ + 0.5388995375823609 + ], + "mse_channel_mean": 0.5388995375823609, + "conditional_kl": 6.2257062511788845, + "m30_total": 2.3229403254004986 + }, + "accepted_moves": 780, + "moved_fraction": 0.9921875, + "mse_local_min_certificate": false, + "trajectory": [ + { + "step": 0, + "objective": 4.902198605273588, + "accuracy": 0.001953125 + }, + { + "step": 50, + "objective": 3.611665206207447, + "accuracy": 0.00390625 + }, + { + "step": 100, + "objective": 3.220172324456457, + "accuracy": 0.00390625 + }, + { + "step": 150, + "objective": 2.9571406644856273, + "accuracy": 0.00390625 + }, + { + "step": 200, + "objective": 2.7599260365911844, + "accuracy": 0.00390625 + }, + { + "step": 250, + "objective": 2.6261551980178712, + "accuracy": 0.00390625 + }, + { + "step": 300, + "objective": 2.536972599217939, + "accuracy": 0.001953125 + }, + { + "step": 350, + "objective": 2.4781931596557047, + "accuracy": 0.00390625 + }, + { + "step": 400, + "objective": 2.4380291906975864, + "accuracy": 0.00390625 + }, + { + "step": 450, + "objective": 2.4039255178327887, + "accuracy": 0.001953125 + }, + { + "step": 500, + "objective": 2.3812263059342933, + "accuracy": 0.001953125 + }, + { + "step": 550, + "objective": 2.3625650342101205, + "accuracy": 0.001953125 + }, + { + "step": 600, + "objective": 2.3489923265882693, + "accuracy": 0.00390625 + }, + { + "step": 650, + "objective": 2.336676872804995, + "accuracy": 0.005859375 + }, + { + "step": 700, + "objective": 2.330324894878429, + "accuracy": 0.005859375 + }, + { + "step": 750, + "objective": 2.324542627850033, + "accuracy": 0.005859375 + } + ] + }, + { + "start_accuracy": 0.0, + "final_accuracy": 0.001953125, + "final_objective": 2.3452468629207983, + "final_energy": { + "mse": 0.5496196952707577, + "mse_channels": [ + 0.5496196952707577 + ], + "mse_channel_mean": 0.5496196952707577, + "conditional_kl": 6.230037361896413, + "m30_total": 2.3452468629207983 + }, + "accepted_moves": 770, + "moved_fraction": 0.99609375, + "mse_local_min_certificate": false, + "trajectory": [ + { + "step": 0, + "objective": 5.155891494550196, + "accuracy": 0.0 + }, + { + "step": 50, + "objective": 3.748140068633493, + "accuracy": 0.001953125 + }, + { + "step": 100, + "objective": 3.270736721624397, + "accuracy": 0.0078125 + }, + { + "step": 150, + "objective": 2.9709530427040196, + "accuracy": 0.0078125 + }, + { + "step": 200, + "objective": 2.7713090756413754, + "accuracy": 0.0078125 + }, + { + "step": 250, + "objective": 2.652638247026789, + "accuracy": 0.0078125 + }, + { + "step": 300, + "objective": 2.5699323330316863, + "accuracy": 0.005859375 + }, + { + "step": 350, + "objective": 2.51172832596485, + "accuracy": 0.005859375 + }, + { + "step": 400, + "objective": 2.466045444700269, + "accuracy": 0.00390625 + }, + { + "step": 450, + "objective": 2.4319681005907015, + "accuracy": 0.001953125 + }, + { + "step": 500, + "objective": 2.407473563312645, + "accuracy": 0.001953125 + }, + { + "step": 550, + "objective": 2.385633340775565, + "accuracy": 0.0 + }, + { + "step": 600, + "objective": 2.370593973275401, + "accuracy": 0.0 + }, + { + "step": 650, + "objective": 2.3610556914299674, + "accuracy": 0.001953125 + }, + { + "step": 700, + "objective": 2.3541518872739857, + "accuracy": 0.001953125 + }, + { + "step": 750, + "objective": 2.347135607547836, + "accuracy": 0.001953125 + } + ] + } + ], + "verdict": { + "true_m30_total": 4.087663627253004, + "identity_is_local_min_mse": false, + "descent_from_true_stays": 0.041015625, + "on_manifold_counterfeit_found": true, + "best_random_descent_m30_total": 2.3229403254004986 + } +} \ No newline at end of file -- cgit v1.2.3