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_seed1_scalar.json | 1155 ++++++++++++++++++++ 1 file changed, 1155 insertions(+) create mode 100644 artifacts/manifold_gate/vg512_region_closed_seed1_scalar.json (limited to 'artifacts/manifold_gate/vg512_region_closed_seed1_scalar.json') diff --git a/artifacts/manifold_gate/vg512_region_closed_seed1_scalar.json b/artifacts/manifold_gate/vg512_region_closed_seed1_scalar.json new file mode 100644 index 0000000..ace10c5 --- /dev/null +++ b/artifacts/manifold_gate/vg512_region_closed_seed1_scalar.json @@ -0,0 +1,1155 @@ +{ + "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": 1, + "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": 1, + "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_seed1_scalar.json", + "trajectory_output": null + }, + "channel_names": [ + "mean" + ], + "gate_a_global_ranking": { + "true": { + "mse": 1.6515487328675447, + "mse_channels": [ + 1.6515487328675447 + ], + "mse_channel_mean": 1.6515487328675447, + "conditional_kl": 5.9846260976727566, + "m30_total": 4.5000226852696406 + }, + "similarity_shift": { + "mse": 1.8542038709707924, + "mse_channels": [ + 1.8542038709707924 + ], + "mse_channel_mean": 1.8542038709707924, + "conditional_kl": 6.164616929558872, + "m30_total": 4.94133112785336 + }, + "random": { + "mse": { + "mean": 1.9998544454574585, + "std": 0.05984543636441231, + "min": 1.8279703855514526, + "true_z": 5.820087909698486, + "true_rank_among_random": 0, + "count": 1000 + }, + "mse_channel_mean": { + "mean": 1.9998544454574585, + "std": 0.05984543636441231, + "min": 1.8279703855514526, + "true_z": 5.820087909698486, + "true_rank_among_random": 0, + "count": 1000 + }, + "conditional_kl": { + "mean": 6.160747528076172, + "std": 0.012881768867373466, + "min": 6.126026153564453, + "true_z": 13.672131538391113, + "true_rank_among_random": 0, + "count": 1000 + }, + "m30_total": { + "mean": 5.231857776641846, + "std": 0.12098576128482819, + "min": 4.88247013092041, + "true_z": 6.048933982849121, + "true_rank_among_random": 0, + "count": 1000 + } + } + }, + "gate_b_transpositions": { + "pairs": 130816, + "improving_pairs": 28278, + "improving_fraction": 0.2161662181996086, + "delta_mean": 0.0019567591873738766, + "delta_min": -0.06470630204987719, + "identity_is_local_min_mse": false, + "worst_improving_swaps": [ + { + "pair": [ + 205, + 379 + ], + "delta_mse": -0.06470630204987719, + "text_cosine": 0.9405932934989035, + "visual_cosine": 0.014687241765777596 + }, + { + "pair": [ + 66, + 205 + ], + "delta_mse": -0.06380311620035345, + "text_cosine": 0.9205223555377072, + "visual_cosine": -0.007733132388556489 + }, + { + "pair": [ + 154, + 379 + ], + "delta_mse": -0.06182349409992374, + "text_cosine": 0.9414433619435776, + "visual_cosine": 0.12552948276053955 + }, + { + "pair": [ + 46, + 379 + ], + "delta_mse": -0.06177283147805319, + "text_cosine": 0.9427488462148406, + "visual_cosine": 0.06489281786813333 + }, + { + "pair": [ + 336, + 379 + ], + "delta_mse": -0.06167120887413124, + "text_cosine": 0.943367530156104, + "visual_cosine": 0.07655640412195952 + }, + { + "pair": [ + 183, + 379 + ], + "delta_mse": -0.06145587892424647, + "text_cosine": 0.9566444897871585, + "visual_cosine": 0.13577289226153977 + }, + { + "pair": [ + 46, + 66 + ], + "delta_mse": -0.05867227346487836, + "text_cosine": 0.9239657279178864, + "visual_cosine": 0.0324886980169832 + }, + { + "pair": [ + 280, + 379 + ], + "delta_mse": -0.058060181075123354, + "text_cosine": 0.9505445577071259, + "visual_cosine": 0.21448360314604037 + }, + { + "pair": [ + 241, + 379 + ], + "delta_mse": -0.05803008919328146, + "text_cosine": 0.9452909962180366, + "visual_cosine": 0.12875603740925373 + }, + { + "pair": [ + 223, + 379 + ], + "delta_mse": -0.05783303858366183, + "text_cosine": 0.9389676382085085, + "visual_cosine": 0.10803450424988954 + }, + { + "pair": [ + 221, + 379 + ], + "delta_mse": -0.056933224419541933, + "text_cosine": 0.9488295099546681, + "visual_cosine": 0.11118152110541739 + }, + { + "pair": [ + 379, + 425 + ], + "delta_mse": -0.05656871671106512, + "text_cosine": 0.9554980502110559, + "visual_cosine": 0.21514276432413382 + }, + { + "pair": [ + 212, + 379 + ], + "delta_mse": -0.05527689001850749, + "text_cosine": 0.956155841947316, + "visual_cosine": 0.21599569958492532 + }, + { + "pair": [ + 115, + 379 + ], + "delta_mse": -0.05487197342093929, + "text_cosine": 0.9475599001382317, + "visual_cosine": 0.15574988298308967 + }, + { + "pair": [ + 28, + 379 + ], + "delta_mse": -0.05465015440522952, + "text_cosine": 0.9413998408735998, + "visual_cosine": 0.10501731344941531 + }, + { + "pair": [ + 66, + 154 + ], + "delta_mse": -0.054176873190848136, + "text_cosine": 0.9321338695145743, + "visual_cosine": 0.022143230671108707 + }, + { + "pair": [ + 66, + 223 + ], + "delta_mse": -0.05380270281669261, + "text_cosine": 0.9240761281581971, + "visual_cosine": 0.05586222586326198 + }, + { + "pair": [ + 66, + 241 + ], + "delta_mse": -0.05350183739580523, + "text_cosine": 0.920607925704046, + "visual_cosine": 0.03703040569739222 + }, + { + "pair": [ + 153, + 379 + ], + "delta_mse": -0.05252812104716774, + "text_cosine": 0.9470905062686079, + "visual_cosine": 0.21381881477198808 + }, + { + "pair": [ + 162, + 379 + ], + "delta_mse": -0.05188687825680716, + "text_cosine": 0.9454407512436758, + "visual_cosine": 0.09433780902667131 + } + ] + }, + "derangement_curve": [ + { + "k": 2, + "delta_mean": 0.0019321494037285447, + "delta_std": 0.004689590539783239, + "improving_fraction": 0.195 + }, + { + "k": 4, + "delta_mean": 0.004396525677293539, + "delta_std": 0.008218545466661453, + "improving_fraction": 0.18 + }, + { + "k": 8, + "delta_mean": 0.007437502034008503, + "delta_std": 0.010923884809017181, + "improving_fraction": 0.16 + }, + { + "k": 16, + "delta_mean": 0.015836434438824654, + "delta_std": 0.01538482028990984, + "improving_fraction": 0.125 + }, + { + "k": 32, + "delta_mean": 0.03078111633658409, + "delta_std": 0.023188168182969093, + "improving_fraction": 0.045 + }, + { + "k": 64, + "delta_mean": 0.061035942286252975, + "delta_std": 0.03195176273584366, + "improving_fraction": 0.01 + }, + { + "k": 128, + "delta_mean": 0.1171056255698204, + "delta_std": 0.04438166692852974, + "improving_fraction": 0.0 + }, + { + "k": 256, + "delta_mean": 0.2154708355665207, + "delta_std": 0.05587908998131752, + "improving_fraction": 0.0 + }, + { + "k": 512, + "delta_mean": 0.34968018531799316, + "delta_std": 0.05444056913256645, + "improving_fraction": 0.0 + } + ], + "gate_c_descent_from_true": { + "start_accuracy": 1.0, + "final_accuracy": 0.033203125, + "final_objective": 2.44409510285621, + "final_energy": { + "mse": 0.6381732719644498, + "mse_channels": [ + 0.6381732719644498 + ], + "mse_channel_mean": 0.6381732719644498, + "conditional_kl": 5.838742794636552, + "m30_total": 2.44409510285621 + }, + "accepted_moves": 748, + "moved_fraction": 0.966796875, + "mse_local_min_certificate": false, + "trajectory": [ + { + "step": 0, + "objective": 4.369927945846397, + "accuracy": 0.99609375 + }, + { + "step": 50, + "objective": 3.1699645346871628, + "accuracy": 0.802734375 + }, + { + "step": 100, + "objective": 2.8872163145203498, + "accuracy": 0.611328125 + }, + { + "step": 150, + "objective": 2.762076906976623, + "accuracy": 0.4609375 + }, + { + "step": 200, + "objective": 2.6859562633549854, + "accuracy": 0.376953125 + }, + { + "step": 250, + "objective": 2.6236145622459714, + "accuracy": 0.294921875 + }, + { + "step": 300, + "objective": 2.57972754515737, + "accuracy": 0.236328125 + }, + { + "step": 350, + "objective": 2.5434692994152774, + "accuracy": 0.166015625 + }, + { + "step": 400, + "objective": 2.515499834659864, + "accuracy": 0.119140625 + }, + { + "step": 450, + "objective": 2.495801289346254, + "accuracy": 0.09375 + }, + { + "step": 500, + "objective": 2.4814460738576205, + "accuracy": 0.064453125 + }, + { + "step": 550, + "objective": 2.4718114748304165, + "accuracy": 0.0546875 + }, + { + "step": 600, + "objective": 2.462072377322613, + "accuracy": 0.048828125 + }, + { + "step": 650, + "objective": 2.4532856167585813, + "accuracy": 0.04296875 + }, + { + "step": 700, + "objective": 2.4473830738839606, + "accuracy": 0.029296875 + } + ], + "final_permutation": [ + 249, + 46, + 233, + 209, + 300, + 424, + 313, + 429, + 111, + 318, + 22, + 282, + 55, + 42, + 14, + 27, + 207, + 17, + 422, + 202, + 262, + 303, + 304, + 23, + 29, + 231, + 169, + 306, + 288, + 478, + 71, + 32, + 480, + 52, + 39, + 443, + 228, + 454, + 263, + 31, + 333, + 321, + 150, + 40, + 36, + 16, + 66, + 128, + 200, + 219, + 492, + 240, + 102, + 247, + 138, + 235, + 432, + 351, + 368, + 287, + 175, + 159, + 434, + 179, + 385, + 412, + 314, + 309, + 450, + 310, + 355, + 156, + 185, + 163, + 440, + 264, + 485, + 272, + 302, + 193, + 80, + 506, + 442, + 142, + 469, + 245, + 56, + 199, + 45, + 124, + 204, + 292, + 261, + 494, + 381, + 76, + 96, + 181, + 389, + 250, + 462, + 226, + 97, + 477, + 356, + 151, + 137, + 107, + 448, + 153, + 374, + 186, + 253, + 7, + 365, + 15, + 339, + 132, + 161, + 57, + 289, + 87, + 122, + 444, + 234, + 414, + 446, + 182, + 218, + 93, + 455, + 417, + 420, + 296, + 315, + 347, + 180, + 394, + 140, + 12, + 205, + 468, + 305, + 246, + 223, + 255, + 244, + 489, + 1, + 453, + 217, + 401, + 211, + 327, + 379, + 409, + 491, + 129, + 106, + 268, + 61, + 256, + 236, + 362, + 345, + 403, + 358, + 461, + 90, + 501, + 364, + 68, + 298, + 439, + 173, + 479, + 416, + 308, + 411, + 241, + 118, + 484, + 230, + 297, + 280, + 431, + 94, + 72, + 387, + 239, + 312, + 481, + 131, + 43, + 285, + 380, + 286, + 162, + 294, + 112, + 201, + 421, + 172, + 123, + 210, + 507, + 177, + 99, + 117, + 357, + 397, + 38, + 319, + 109, + 147, + 101, + 238, + 447, + 213, + 405, + 437, + 360, + 376, + 342, + 63, + 470, + 145, + 467, + 74, + 168, + 3, + 203, + 65, + 125, + 354, + 49, + 126, + 348, + 438, + 279, + 410, + 330, + 277, + 299, + 81, + 324, + 48, + 373, + 445, + 472, + 75, + 499, + 252, + 197, + 340, + 451, + 383, + 113, + 260, + 58, + 41, + 413, + 307, + 10, + 464, + 503, + 473, + 334, + 367, + 143, + 407, + 353, + 212, + 498, + 167, + 149, + 488, + 301, + 82, + 316, + 457, + 281, + 344, + 248, + 121, + 115, + 196, + 133, + 190, + 270, + 343, + 8, + 13, + 157, + 105, + 336, + 98, + 426, + 393, + 9, + 4, + 291, + 456, + 490, + 259, + 108, + 26, + 258, + 496, + 283, + 390, + 482, + 332, + 400, + 44, + 341, + 363, + 37, + 338, + 399, + 487, + 146, + 328, + 350, + 227, + 370, + 67, + 320, + 329, + 85, + 396, + 378, + 293, + 418, + 214, + 419, + 152, + 459, + 222, + 78, + 408, + 195, + 136, + 91, + 120, + 266, + 295, + 119, + 510, + 290, + 187, + 369, + 349, + 30, + 194, + 51, + 35, + 184, + 104, + 77, + 441, + 322, + 386, + 34, + 504, + 164, + 254, + 189, + 242, + 19, + 158, + 251, + 372, + 183, + 69, + 28, + 359, + 377, + 174, + 460, + 155, + 430, + 493, + 188, + 323, + 271, + 18, + 335, + 388, + 404, + 366, + 391, + 103, + 130, + 84, + 395, + 371, + 415, + 92, + 11, + 83, + 317, + 47, + 269, + 220, + 392, + 33, + 208, + 62, + 0, + 425, + 509, + 154, + 406, + 433, + 178, + 73, + 192, + 232, + 139, + 224, + 284, + 216, + 474, + 398, + 170, + 311, + 5, + 428, + 267, + 89, + 6, + 352, + 86, + 276, + 325, + 100, + 225, + 511, + 449, + 54, + 141, + 274, + 427, + 166, + 257, + 176, + 135, + 435, + 497, + 375, + 24, + 331, + 114, + 95, + 361, + 64, + 50, + 458, + 278, + 229, + 79, + 215, + 2, + 237, + 144, + 382, + 466, + 486, + 110, + 206, + 116, + 505, + 53, + 337, + 70, + 127, + 384, + 21, + 465, + 452, + 59, + 160, + 60, + 423, + 273, + 148, + 20, + 475, + 346, + 198, + 165, + 402, + 326, + 243, + 191, + 476, + 221, + 436, + 171, + 265, + 502, + 134, + 500, + 483, + 463, + 25, + 495, + 508, + 88, + 471, + 275 + ] + }, + "gate_c_descent_from_random": [ + { + "start_accuracy": 0.001953125, + "final_accuracy": 0.001953125, + "final_objective": 2.3479748516707404, + "final_energy": { + "mse": 0.5904302149564162, + "mse_channels": [ + 0.5904302149564162 + ], + "mse_channel_mean": 0.5904302149564162, + "conditional_kl": 5.835572108789539, + "m30_total": 2.3479748516707404 + }, + "accepted_moves": 810, + "moved_fraction": 0.9921875, + "mse_local_min_certificate": false, + "trajectory": [ + { + "step": 0, + "objective": 5.14488708567028, + "accuracy": 0.001953125 + }, + { + "step": 50, + "objective": 3.553393027797632, + "accuracy": 0.001953125 + }, + { + "step": 100, + "objective": 3.116621588543678, + "accuracy": 0.001953125 + }, + { + "step": 150, + "objective": 2.880577415129924, + "accuracy": 0.001953125 + }, + { + "step": 200, + "objective": 2.7160473643810423, + "accuracy": 0.001953125 + }, + { + "step": 250, + "objective": 2.60750162297464, + "accuracy": 0.001953125 + }, + { + "step": 300, + "objective": 2.53645562583565, + "accuracy": 0.001953125 + }, + { + "step": 350, + "objective": 2.486366099883642, + "accuracy": 0.001953125 + }, + { + "step": 400, + "objective": 2.449537778892605, + "accuracy": 0.001953125 + }, + { + "step": 450, + "objective": 2.4226684250942085, + "accuracy": 0.0 + }, + { + "step": 500, + "objective": 2.401752395801379, + "accuracy": 0.0 + }, + { + "step": 550, + "objective": 2.384395702091102, + "accuracy": 0.0 + }, + { + "step": 600, + "objective": 2.3724916253011115, + "accuracy": 0.0 + }, + { + "step": 650, + "objective": 2.362153197846242, + "accuracy": 0.001953125 + }, + { + "step": 700, + "objective": 2.3554742037721557, + "accuracy": 0.001953125 + }, + { + "step": 750, + "objective": 2.351592774443901, + "accuracy": 0.001953125 + }, + { + "step": 800, + "objective": 2.348086888531469, + "accuracy": 0.001953125 + } + ] + }, + { + "start_accuracy": 0.0, + "final_accuracy": 0.005859375, + "final_objective": 2.3746819125340863, + "final_energy": { + "mse": 0.6044047390975571, + "mse_channels": [ + 0.6044047390975571 + ], + "mse_channel_mean": 0.6044047390975571, + "conditional_kl": 5.829362171694861, + "m30_total": 2.3746819125340863 + }, + "accepted_moves": 759, + "moved_fraction": 0.990234375, + "mse_local_min_certificate": false, + "trajectory": [ + { + "step": 0, + "objective": 5.051127873390432, + "accuracy": 0.0 + }, + { + "step": 50, + "objective": 3.539315315998911, + "accuracy": 0.0 + }, + { + "step": 100, + "objective": 3.1149268659600717, + "accuracy": 0.001953125 + }, + { + "step": 150, + "objective": 2.8914776731537, + "accuracy": 0.001953125 + }, + { + "step": 200, + "objective": 2.738158488888165, + "accuracy": 0.001953125 + }, + { + "step": 250, + "objective": 2.642374740652749, + "accuracy": 0.005859375 + }, + { + "step": 300, + "objective": 2.5722662348244705, + "accuracy": 0.005859375 + }, + { + "step": 350, + "objective": 2.5139342365666906, + "accuracy": 0.00390625 + }, + { + "step": 400, + "objective": 2.47262065414534, + "accuracy": 0.00390625 + }, + { + "step": 450, + "objective": 2.4430651620532586, + "accuracy": 0.00390625 + }, + { + "step": 500, + "objective": 2.4208576671129585, + "accuracy": 0.00390625 + }, + { + "step": 550, + "objective": 2.4044564189635125, + "accuracy": 0.00390625 + }, + { + "step": 600, + "objective": 2.392696456280015, + "accuracy": 0.00390625 + }, + { + "step": 650, + "objective": 2.3842900798271276, + "accuracy": 0.005859375 + }, + { + "step": 700, + "objective": 2.3790345365721377, + "accuracy": 0.005859375 + }, + { + "step": 750, + "objective": 2.3750385127977625, + "accuracy": 0.005859375 + } + ] + } + ], + "verdict": { + "true_m30_total": 4.5000226852696406, + "identity_is_local_min_mse": false, + "descent_from_true_stays": 0.033203125, + "on_manifold_counterfeit_found": true, + "best_random_descent_m30_total": 2.3479748516707404 + } +} \ No newline at end of file -- cgit v1.2.3