diff options
| author | Yuren Hao <blackhao0426@gmail.com> | 2026-08-01 14:10:03 -0500 |
|---|---|---|
| committer | Yuren Hao <blackhao0426@gmail.com> | 2026-08-01 14:10:03 -0500 |
| commit | a62cf4d2a99b4a7985c61b2a7feb92a82a8218b7 (patch) | |
| tree | ee2248078db7edf3812a07f195afa3d9bd6f10c6 /artifacts/manifold_gate/vg512_visible_relations_seed0_scalar.json | |
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 <noreply@anthropic.com>
Diffstat (limited to 'artifacts/manifold_gate/vg512_visible_relations_seed0_scalar.json')
| -rw-r--r-- | artifacts/manifold_gate/vg512_visible_relations_seed0_scalar.json | 1145 |
1 files changed, 1145 insertions, 0 deletions
diff --git a/artifacts/manifold_gate/vg512_visible_relations_seed0_scalar.json b/artifacts/manifold_gate/vg512_visible_relations_seed0_scalar.json new file mode 100644 index 0000000..3b62d37 --- /dev/null +++ b/artifacts/manifold_gate/vg512_visible_relations_seed0_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": 0, + "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": 0, + "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_seed0_scalar.json", + "trajectory_output": null + }, + "channel_names": [ + "mean" + ], + "gate_a_global_ranking": { + "true": { + "mse": 1.4717835012445672, + "mse_channels": [ + 1.4717835012445672 + ], + "mse_channel_mean": 1.4717835012445672, + "conditional_kl": 5.968007727757813, + "m30_total": 4.137168548040697 + }, + "similarity_shift": { + "mse": 1.7293597443292565, + "mse_channels": [ + 1.7293597443292565 + ], + "mse_channel_mean": 1.7293597443292565, + "conditional_kl": 6.148225399493324, + "m30_total": 4.688364568557178 + }, + "random": { + "mse": { + "mean": 2.0038411617279053, + "std": 0.05808873474597931, + "min": 1.7806980609893799, + "true_z": 9.159394264221191, + "true_rank_among_random": 0, + "count": 1000 + }, + "mse_channel_mean": { + "mean": 2.0038411617279053, + "std": 0.05808873474597931, + "min": 1.7806980609893799, + "true_z": 9.159394264221191, + "true_rank_among_random": 0, + "count": 1000 + }, + "conditional_kl": { + "mean": 6.156534194946289, + "std": 0.00829392485320568, + "min": 6.131365776062012, + "true_z": 22.730690002441406, + "true_rank_among_random": 0, + "count": 1000 + }, + "m30_total": { + "mean": 5.238989353179932, + "std": 0.11693011969327927, + "min": 4.790948390960693, + "true_z": 9.422901153564453, + "true_rank_among_random": 0, + "count": 1000 + } + } + }, + "gate_b_transpositions": { + "pairs": 130816, + "improving_pairs": 23412, + "improving_fraction": 0.17896893346379647, + "delta_mean": 0.002980705542127867, + "delta_min": -0.05056307950251094, + "identity_is_local_min_mse": false, + "worst_improving_swaps": [ + { + "pair": [ + 150, + 433 + ], + "delta_mse": -0.05056307950251094, + "text_cosine": 0.9454795294170473, + "visual_cosine": 0.015450217703071843 + }, + { + "pair": [ + 119, + 433 + ], + "delta_mse": -0.05054901771345786, + "text_cosine": 0.9528133656622267, + "visual_cosine": 0.06594341264786825 + }, + { + "pair": [ + 102, + 433 + ], + "delta_mse": -0.05053417681193542, + "text_cosine": 0.9642451117370466, + "visual_cosine": 0.06057722018289307 + }, + { + "pair": [ + 332, + 433 + ], + "delta_mse": -0.04561659689666343, + "text_cosine": 0.9704587447488836, + "visual_cosine": 0.254645657215378 + }, + { + "pair": [ + 163, + 433 + ], + "delta_mse": -0.04471826068142126, + "text_cosine": 0.9661196382265661, + "visual_cosine": 0.08575847621183837 + }, + { + "pair": [ + 245, + 433 + ], + "delta_mse": -0.04402158801547627, + "text_cosine": 0.963429125301189, + "visual_cosine": 0.32392893843045517 + }, + { + "pair": [ + 50, + 433 + ], + "delta_mse": -0.043589376231549244, + "text_cosine": 0.9688035100438744, + "visual_cosine": 0.21323794017292516 + }, + { + "pair": [ + 392, + 433 + ], + "delta_mse": -0.04191438919387446, + "text_cosine": 0.9603847980862464, + "visual_cosine": 0.18498394025619516 + }, + { + "pair": [ + 225, + 433 + ], + "delta_mse": -0.040448057381542744, + "text_cosine": 0.9603364203411391, + "visual_cosine": 0.14080634948066656 + }, + { + "pair": [ + 187, + 433 + ], + "delta_mse": -0.04036040340949781, + "text_cosine": 0.9565432095417611, + "visual_cosine": 0.16897480318538813 + }, + { + "pair": [ + 412, + 433 + ], + "delta_mse": -0.03945616752740333, + "text_cosine": 0.9533033730298265, + "visual_cosine": 0.3143936606273374 + }, + { + "pair": [ + 212, + 433 + ], + "delta_mse": -0.039123560794400596, + "text_cosine": 0.9608494764367717, + "visual_cosine": 0.2842317288233234 + }, + { + "pair": [ + 433, + 498 + ], + "delta_mse": -0.0378671378247089, + "text_cosine": 0.9687232009357419, + "visual_cosine": 0.13798397785505803 + }, + { + "pair": [ + 433, + 470 + ], + "delta_mse": -0.03750848088934985, + "text_cosine": 0.9597514644692255, + "visual_cosine": 0.3018466738128574 + }, + { + "pair": [ + 180, + 433 + ], + "delta_mse": -0.03716789084723679, + "text_cosine": 0.9476805683213432, + "visual_cosine": 0.16397094048305214 + }, + { + "pair": [ + 31, + 433 + ], + "delta_mse": -0.036332523870666836, + "text_cosine": 0.957539011939088, + "visual_cosine": 0.1068629875582747 + }, + { + "pair": [ + 140, + 433 + ], + "delta_mse": -0.035355310900221947, + "text_cosine": 0.9596050421692573, + "visual_cosine": 0.2023845699084794 + }, + { + "pair": [ + 8, + 433 + ], + "delta_mse": -0.0351586264958246, + "text_cosine": 0.9654667157558119, + "visual_cosine": 0.38412422696105014 + }, + { + "pair": [ + 410, + 433 + ], + "delta_mse": -0.034387601126457445, + "text_cosine": 0.9612785394198398, + "visual_cosine": 0.09670278236856066 + }, + { + "pair": [ + 376, + 433 + ], + "delta_mse": -0.03414024524304242, + "text_cosine": 0.9688445602822803, + "visual_cosine": 0.2420523404604336 + } + ] + }, + "derangement_curve": [ + { + "k": 2, + "delta_mean": 0.0025993201415985823, + "delta_std": 0.0046549285762012005, + "improving_fraction": 0.145 + }, + { + "k": 4, + "delta_mean": 0.00649539427831769, + "delta_std": 0.008177580311894417, + "improving_fraction": 0.16 + }, + { + "k": 8, + "delta_mean": 0.011319151148200035, + "delta_std": 0.011565388180315495, + "improving_fraction": 0.11 + }, + { + "k": 16, + "delta_mean": 0.02397003583610058, + "delta_std": 0.015329316258430481, + "improving_fraction": 0.02 + }, + { + "k": 32, + "delta_mean": 0.045578815042972565, + "delta_std": 0.02171756885945797, + "improving_fraction": 0.0 + }, + { + "k": 64, + "delta_mean": 0.0897127240896225, + "delta_std": 0.028539210557937622, + "improving_fraction": 0.0 + }, + { + "k": 128, + "delta_mean": 0.1738407015800476, + "delta_std": 0.03862808644771576, + "improving_fraction": 0.0 + }, + { + "k": 256, + "delta_mean": 0.32519611716270447, + "delta_std": 0.05331876128911972, + "improving_fraction": 0.0 + }, + { + "k": 512, + "delta_mean": 0.5326806902885437, + "delta_std": 0.05615127086639404, + "improving_fraction": 0.0 + } + ], + "gate_c_descent_from_true": { + "start_accuracy": 1.0, + "final_accuracy": 0.037109375, + "final_objective": 2.2841391441805214, + "final_energy": { + "mse": 0.5519732376658675, + "mse_channels": [ + 0.5519732376658675 + ], + "mse_channel_mean": 0.5519732376658675, + "conditional_kl": 5.900963344243933, + "m30_total": 2.2841391441805214 + }, + "accepted_moves": 673, + "moved_fraction": 0.962890625, + "mse_local_min_certificate": false, + "trajectory": [ + { + "step": 0, + "objective": 4.036089902485733, + "accuracy": 0.99609375 + }, + { + "step": 50, + "objective": 3.0040949706432825, + "accuracy": 0.80078125 + }, + { + "step": 100, + "objective": 2.7449894619442476, + "accuracy": 0.611328125 + }, + { + "step": 150, + "objective": 2.5969882990102438, + "accuracy": 0.45703125 + }, + { + "step": 200, + "objective": 2.5035933177908194, + "accuracy": 0.341796875 + }, + { + "step": 250, + "objective": 2.4428887614170582, + "accuracy": 0.27734375 + }, + { + "step": 300, + "objective": 2.4021188415619057, + "accuracy": 0.21875 + }, + { + "step": 350, + "objective": 2.3705488268511172, + "accuracy": 0.169921875 + }, + { + "step": 400, + "objective": 2.345823514121186, + "accuracy": 0.138671875 + }, + { + "step": 450, + "objective": 2.326289215336553, + "accuracy": 0.11328125 + }, + { + "step": 500, + "objective": 2.3111323904000294, + "accuracy": 0.08203125 + }, + { + "step": 550, + "objective": 2.2993051420283113, + "accuracy": 0.064453125 + }, + { + "step": 600, + "objective": 2.291767646258156, + "accuracy": 0.046875 + }, + { + "step": 650, + "objective": 2.2867593184699704, + "accuracy": 0.041015625 + } + ], + "final_permutation": [ + 37, + 387, + 278, + 200, + 232, + 179, + 425, + 7, + 157, + 274, + 83, + 209, + 447, + 362, + 18, + 128, + 418, + 131, + 415, + 156, + 52, + 80, + 71, + 509, + 11, + 63, + 350, + 420, + 323, + 451, + 30, + 433, + 269, + 346, + 188, + 267, + 494, + 444, + 134, + 108, + 40, + 508, + 193, + 390, + 44, + 320, + 271, + 176, + 328, + 58, + 480, + 221, + 207, + 261, + 474, + 332, + 5, + 330, + 219, + 338, + 138, + 284, + 381, + 498, + 382, + 427, + 307, + 231, + 23, + 137, + 234, + 8, + 456, + 265, + 436, + 500, + 310, + 303, + 315, + 298, + 345, + 393, + 492, + 15, + 70, + 252, + 172, + 247, + 254, + 289, + 211, + 260, + 123, + 184, + 46, + 402, + 119, + 168, + 454, + 225, + 410, + 503, + 53, + 292, + 49, + 348, + 48, + 107, + 178, + 14, + 136, + 275, + 302, + 371, + 92, + 380, + 311, + 117, + 112, + 133, + 1, + 372, + 60, + 146, + 431, + 141, + 326, + 462, + 214, + 76, + 78, + 227, + 162, + 201, + 411, + 159, + 455, + 174, + 266, + 339, + 139, + 313, + 383, + 115, + 55, + 417, + 322, + 461, + 32, + 259, + 121, + 334, + 106, + 212, + 99, + 103, + 438, + 489, + 0, + 327, + 458, + 161, + 349, + 67, + 319, + 111, + 177, + 440, + 171, + 33, + 77, + 182, + 360, + 386, + 392, + 142, + 19, + 169, + 329, + 90, + 460, + 89, + 305, + 38, + 93, + 452, + 122, + 373, + 394, + 88, + 487, + 25, + 308, + 268, + 194, + 129, + 132, + 406, + 510, + 86, + 95, + 358, + 486, + 28, + 496, + 65, + 409, + 446, + 218, + 283, + 469, + 180, + 75, + 466, + 51, + 317, + 321, + 287, + 256, + 12, + 370, + 43, + 257, + 239, + 314, + 238, + 281, + 437, + 164, + 388, + 365, + 471, + 423, + 477, + 270, + 3, + 312, + 105, + 378, + 243, + 355, + 158, + 96, + 192, + 143, + 223, + 235, + 73, + 430, + 288, + 342, + 419, + 41, + 22, + 343, + 155, + 499, + 220, + 187, + 258, + 363, + 69, + 185, + 130, + 244, + 216, + 391, + 102, + 203, + 150, + 21, + 264, + 470, + 2, + 351, + 464, + 376, + 429, + 435, + 175, + 17, + 357, + 50, + 421, + 97, + 408, + 35, + 9, + 341, + 10, + 389, + 291, + 56, + 479, + 309, + 295, + 81, + 505, + 79, + 208, + 304, + 98, + 400, + 293, + 273, + 213, + 91, + 276, + 186, + 467, + 463, + 149, + 445, + 154, + 369, + 6, + 422, + 395, + 318, + 403, + 506, + 279, + 377, + 127, + 296, + 290, + 147, + 181, + 251, + 413, + 228, + 148, + 24, + 253, + 374, + 263, + 453, + 385, + 493, + 233, + 473, + 4, + 240, + 344, + 475, + 84, + 352, + 347, + 237, + 224, + 331, + 476, + 439, + 222, + 204, + 282, + 490, + 324, + 94, + 66, + 414, + 280, + 379, + 262, + 459, + 356, + 62, + 250, + 173, + 47, + 191, + 118, + 249, + 135, + 457, + 424, + 285, + 366, + 472, + 325, + 337, + 54, + 299, + 167, + 384, + 104, + 368, + 300, + 202, + 34, + 36, + 246, + 511, + 294, + 197, + 16, + 397, + 333, + 478, + 116, + 286, + 126, + 189, + 501, + 85, + 248, + 375, + 407, + 483, + 206, + 57, + 405, + 100, + 64, + 45, + 151, + 359, + 145, + 109, + 255, + 404, + 242, + 399, + 491, + 74, + 432, + 495, + 163, + 82, + 449, + 217, + 210, + 443, + 412, + 434, + 468, + 497, + 401, + 198, + 215, + 297, + 364, + 125, + 68, + 124, + 488, + 354, + 13, + 448, + 229, + 426, + 465, + 31, + 335, + 190, + 416, + 507, + 442, + 26, + 502, + 428, + 152, + 441, + 241, + 316, + 29, + 114, + 101, + 195, + 87, + 361, + 120, + 166, + 165, + 336, + 450, + 110, + 140, + 59, + 272, + 113, + 481, + 306, + 199, + 484, + 61, + 72, + 153, + 170, + 277, + 230, + 398, + 482, + 27, + 205, + 485, + 196, + 340, + 353, + 367, + 226, + 20, + 301, + 42, + 504, + 39, + 396, + 183, + 236, + 144, + 160, + 245 + ] + }, + "gate_c_descent_from_random": [ + { + "start_accuracy": 0.001953125, + "final_accuracy": 0.001953125, + "final_objective": 2.2986974676937493, + "final_energy": { + "mse": 0.5572359088601652, + "mse_channels": [ + 0.5572359088601652 + ], + "mse_channel_mean": 0.5572359088601652, + "conditional_kl": 5.921128249867095, + "m30_total": 2.2986974676937493 + }, + "accepted_moves": 786, + "moved_fraction": 0.994140625, + "mse_local_min_certificate": false, + "trajectory": [ + { + "step": 0, + "objective": 5.197856043548812, + "accuracy": 0.001953125 + }, + { + "step": 50, + "objective": 3.6246038212854557, + "accuracy": 0.001953125 + }, + { + "step": 100, + "objective": 3.164414241397255, + "accuracy": 0.001953125 + }, + { + "step": 150, + "objective": 2.939607618548007, + "accuracy": 0.001953125 + }, + { + "step": 200, + "objective": 2.7447945813014134, + "accuracy": 0.0 + }, + { + "step": 250, + "objective": 2.628339595128332, + "accuracy": 0.0 + }, + { + "step": 300, + "objective": 2.547931396314548, + "accuracy": 0.0 + }, + { + "step": 350, + "objective": 2.484709691369049, + "accuracy": 0.0 + }, + { + "step": 400, + "objective": 2.433397659353605, + "accuracy": 0.0 + }, + { + "step": 450, + "objective": 2.394079749167995, + "accuracy": 0.0 + }, + { + "step": 500, + "objective": 2.363902156389477, + "accuracy": 0.0 + }, + { + "step": 550, + "objective": 2.3413969947514417, + "accuracy": 0.001953125 + }, + { + "step": 600, + "objective": 2.3259410995794925, + "accuracy": 0.001953125 + }, + { + "step": 650, + "objective": 2.3160112487522206, + "accuracy": 0.001953125 + }, + { + "step": 700, + "objective": 2.308551453476427, + "accuracy": 0.0 + }, + { + "step": 750, + "objective": 2.3029278444732855, + "accuracy": 0.001953125 + } + ] + }, + { + "start_accuracy": 0.0, + "final_accuracy": 0.009765625, + "final_objective": 2.265278257348334, + "final_energy": { + "mse": 0.541804771765918, + "mse_channels": [ + 0.541804771765918 + ], + "mse_channel_mean": 0.541804771765918, + "conditional_kl": 5.908343569082491, + "m30_total": 2.265278257348334 + }, + "accepted_moves": 779, + "moved_fraction": 0.994140625, + "mse_local_min_certificate": false, + "trajectory": [ + { + "step": 0, + "objective": 4.971222264891033, + "accuracy": 0.0 + }, + { + "step": 50, + "objective": 3.5705983585083563, + "accuracy": 0.0 + }, + { + "step": 100, + "objective": 3.126768088621347, + "accuracy": 0.0 + }, + { + "step": 150, + "objective": 2.8637858869772934, + "accuracy": 0.0 + }, + { + "step": 200, + "objective": 2.680101845021605, + "accuracy": 0.0 + }, + { + "step": 250, + "objective": 2.564101870929306, + "accuracy": 0.0 + }, + { + "step": 300, + "objective": 2.484654824680079, + "accuracy": 0.0 + }, + { + "step": 350, + "objective": 2.426162852488167, + "accuracy": 0.00390625 + }, + { + "step": 400, + "objective": 2.3796121472377463, + "accuracy": 0.00390625 + }, + { + "step": 450, + "objective": 2.3465863436089927, + "accuracy": 0.00390625 + }, + { + "step": 500, + "objective": 2.3204555586291113, + "accuracy": 0.005859375 + }, + { + "step": 550, + "objective": 2.301818044364172, + "accuracy": 0.005859375 + }, + { + "step": 600, + "objective": 2.289287216093343, + "accuracy": 0.0078125 + }, + { + "step": 650, + "objective": 2.2794507296197324, + "accuracy": 0.0078125 + }, + { + "step": 700, + "objective": 2.2726524575546176, + "accuracy": 0.009765625 + }, + { + "step": 750, + "objective": 2.267518096700906, + "accuracy": 0.009765625 + } + ] + } + ], + "verdict": { + "true_m30_total": 4.137168548040697, + "identity_is_local_min_mse": false, + "descent_from_true_stays": 0.037109375, + "on_manifold_counterfeit_found": true, + "best_random_descent_m30_total": 2.265278257348334 + } +}
\ No newline at end of file |
