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 --- .../manifold_gate/ricci_vg512_region_seed0.json | 819 +++++++++++++++++++++ 1 file changed, 819 insertions(+) create mode 100644 artifacts/manifold_gate/ricci_vg512_region_seed0.json (limited to 'artifacts/manifold_gate/ricci_vg512_region_seed0.json') diff --git a/artifacts/manifold_gate/ricci_vg512_region_seed0.json b/artifacts/manifold_gate/ricci_vg512_region_seed0.json new file mode 100644 index 0000000..fbe2ae6 --- /dev/null +++ b/artifacts/manifold_gate/ricci_vg512_region_seed0.json @@ -0,0 +1,819 @@ +{ + "protocol": "Each modality's kNN geometry evolves independently under Ollivier-Ricci flow; the identical heat-kernel channels are gated with and without the flow. Hidden pairs score orderings only.", + "meta": { + "dataset": "visual_genome_5k", + "tier": "region_closed", + "samples": 512, + "subset_seed": 0, + "bundle_channels": false, + "flow": { + "dataset": "vg", + "manifest": "artifacts/manifest.json", + "vision": "artifacts/vision.pt", + "text": "artifacts/text.pt", + "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": 0, + "neighbors": 16, + "flow_iterations": 10, + "flow_step": 0.4, + "lazy_alpha": 0.5, + "heat_times": "1.0,4.0", + "random_perms": 300, + "derangement_samples": 50, + "descent_restarts": 2, + "descent_max_steps": 200000, + "descent_objective": "mse", + "descent_verify_top": 64, + "device": "cpu", + "seed": 20260729, + "output": "artifacts/manifold_gate/ricci_vg512_region_seed0.json", + "trajectory_output": null + } + }, + "conditions": { + "diffusion_no_flow": { + "gate_a": { + "true": { + "mse": 1.6014539100876404, + "mse_channels": [ + 1.6014539100876404, + 1.507990736404536, + 1.6397135422886735 + ], + "mse_channel_mean": 1.5830527295936168, + "conditional_kl": 63.08760651902557, + "m30_total": 15.820429123980395 + }, + "similarity_shift": { + "mse": 1.8128208789485603, + "mse_channels": [ + 1.8128208789485603, + 1.845878205690379, + 1.6675622611645458 + ], + "mse_channel_mean": 1.7754204486011618, + "conditional_kl": 90.33766858405629, + "m30_total": 21.69317547470838 + }, + "random": { + "mse": { + "mean": 2.0023739337921143, + "std": 0.0580625906586647, + "min": 1.7915462255477905, + "true_z": 6.90496301651001, + "true_rank_among_random": 0, + "count": 300 + }, + "mse_channel_mean": { + "mean": 2.0025570392608643, + "std": 0.06107397750020027, + "min": 1.7756344079971313, + "true_z": 6.868789196014404, + "true_rank_among_random": 0, + "count": 300 + }, + "conditional_kl": { + "mean": 93.03399658203125, + "std": 2.193070411682129, + "min": 88.20814514160156, + "true_z": 13.65500545501709, + "true_rank_among_random": 0, + "count": 300 + }, + "m30_total": { + "mean": 22.611547470092773, + "std": 0.4970371127128601, + "min": 21.444181442260742, + "true_z": 13.663202285766602, + "true_rank_among_random": 0, + "count": 300 + } + } + }, + "gate_b": { + "pairs": 130816, + "improving_pairs": 29915, + "improving_fraction": 0.22867997798434442, + "delta_mean": 0.0023632367580343366, + "delta_min": -0.04577336576589821, + "identity_is_local_min_mse": false, + "worst_improving_swaps": [ + { + "pair": [ + 133, + 180 + ], + "delta_mse": -0.04577336576589821, + "text_cosine": -6.753513260650261, + "visual_cosine": -2.6523739190913918 + }, + { + "pair": [ + 18, + 163 + ], + "delta_mse": -0.04293332266480033, + "text_cosine": -2.4459475829014465, + "visual_cosine": -0.9823612643787966 + }, + { + "pair": [ + 18, + 180 + ], + "delta_mse": -0.04234310017681062, + "text_cosine": -3.9945642790848104, + "visual_cosine": -1.6409603795929495 + }, + { + "pair": [ + 180, + 480 + ], + "delta_mse": -0.041341194100504214, + "text_cosine": -5.4599817833668585, + "visual_cosine": -1.8022681935670093 + }, + { + "pair": [ + 18, + 434 + ], + "delta_mse": -0.0411120787708388, + "text_cosine": -3.368979666557572, + "visual_cosine": -1.104009210145915 + }, + { + "pair": [ + 133, + 452 + ], + "delta_mse": -0.04089527627753412, + "text_cosine": -6.435752203141073, + "visual_cosine": -2.234608712045037 + }, + { + "pair": [ + 18, + 452 + ], + "delta_mse": -0.0406096980771913, + "text_cosine": -3.763439468344244, + "visual_cosine": -1.5239689457059888 + }, + { + "pair": [ + 163, + 367 + ], + "delta_mse": -0.03935780762932796, + "text_cosine": -2.2556759248992497, + "visual_cosine": -1.2071738742854117 + }, + { + "pair": [ + 18, + 245 + ], + "delta_mse": -0.03891432021647693, + "text_cosine": -3.033982808205132, + "visual_cosine": -0.8205493521315517 + }, + { + "pair": [ + 367, + 434 + ], + "delta_mse": -0.03778761124898473, + "text_cosine": -2.1211615068184493, + "visual_cosine": -1.0421343214740304 + }, + { + "pair": [ + 452, + 480 + ], + "delta_mse": -0.03746113192488994, + "text_cosine": -5.142220725857669, + "visual_cosine": -1.1646527535780538 + }, + { + "pair": [ + 18, + 90 + ], + "delta_mse": -0.0373495416304652, + "text_cosine": -3.06243045204508, + "visual_cosine": -1.0703344791761449 + }, + { + "pair": [ + 18, + 285 + ], + "delta_mse": -0.03625942652997801, + "text_cosine": -3.774236195789333, + "visual_cosine": -1.109824039759465 + }, + { + "pair": [ + 180, + 367 + ], + "delta_mse": -0.03596604865533066, + "text_cosine": -3.074955810310793, + "visual_cosine": -0.23578690774565125 + }, + { + "pair": [ + 8, + 18 + ], + "delta_mse": -0.035556634646583665, + "text_cosine": -3.0884676419379278, + "visual_cosine": -0.8162738056389973 + }, + { + "pair": [ + 245, + 367 + ], + "delta_mse": -0.03533601006391431, + "text_cosine": -2.542562905971847, + "visual_cosine": -1.4112518870605946 + }, + { + "pair": [ + 18, + 137 + ], + "delta_mse": -0.035299656181634775, + "text_cosine": -3.5392329409027767, + "visual_cosine": -0.8977789730024032 + }, + { + "pair": [ + 18, + 412 + ], + "delta_mse": -0.03517567825578721, + "text_cosine": -3.6359095190009407, + "visual_cosine": -0.941861762090096 + }, + { + "pair": [ + 367, + 452 + ], + "delta_mse": -0.03516319031055127, + "text_cosine": -3.072674210304903, + "visual_cosine": -1.3916494805206137 + }, + { + "pair": [ + 18, + 459 + ], + "delta_mse": -0.03508350837047559, + "text_cosine": -3.162013729159052, + "visual_cosine": -0.7850327147304061 + } + ] + }, + "descent_from_true": { + "start_accuracy": 1.0, + "final_accuracy": 0.03125, + "final_objective": 0.6853002823500745, + "final_energy": { + "mse": 0.6853002823500745, + "mse_channels": [ + 0.6853002823500745, + 0.7124803735837811, + 0.2892207671537166 + ], + "mse_channel_mean": 0.5623338076958574, + "conditional_kl": 42.51059050915735, + "m30_total": 9.87271866653162 + }, + "accepted_moves": 740, + "moved_fraction": 0.96875, + "mse_local_min_certificate": true, + "trajectory": [ + { + "step": 0, + "objective": 1.5433090779993968, + "accuracy": 0.99609375 + }, + { + "step": 50, + "objective": 1.0332795075297028, + "accuracy": 0.80078125 + }, + { + "step": 100, + "objective": 0.9043116636330876, + "accuracy": 0.6171875 + }, + { + "step": 150, + "objective": 0.8440717781832804, + "accuracy": 0.484375 + }, + { + "step": 200, + "objective": 0.8032930121335065, + "accuracy": 0.375 + }, + { + "step": 250, + "objective": 0.7713698048143797, + "accuracy": 0.294921875 + }, + { + "step": 300, + "objective": 0.7501521623251177, + "accuracy": 0.228515625 + }, + { + "step": 350, + "objective": 0.7336656894405466, + "accuracy": 0.177734375 + }, + { + "step": 400, + "objective": 0.721368675673311, + "accuracy": 0.130859375 + }, + { + "step": 450, + "objective": 0.7106485549630995, + "accuracy": 0.1015625 + }, + { + "step": 500, + "objective": 0.7028514622865484, + "accuracy": 0.08984375 + }, + { + "step": 550, + "objective": 0.6961834125368588, + "accuracy": 0.072265625 + }, + { + "step": 600, + "objective": 0.6915529453807441, + "accuracy": 0.056640625 + }, + { + "step": 650, + "objective": 0.689014790855117, + "accuracy": 0.0390625 + }, + { + "step": 700, + "objective": 0.6871787621936848, + "accuracy": 0.037109375 + } + ] + }, + "descent_from_random": [ + { + "start_accuracy": 0.00390625, + "final_accuracy": 0.009765625, + "final_objective": 0.6475031218543564, + "final_energy": { + "mse": 0.6475031218543564, + "mse_channels": [ + 0.6475031218543564, + 0.6734213147417014, + 0.27414993654230096 + ], + "mse_channel_mean": 0.5316914577127863, + "conditional_kl": 43.47661595897679, + "m30_total": 9.990329435504071 + }, + "accepted_moves": 773, + "moved_fraction": 0.994140625, + "mse_local_min_certificate": true + }, + { + "start_accuracy": 0.0, + "final_accuracy": 0.001953125, + "final_objective": 0.6793331980695506, + "final_energy": { + "mse": 0.6793331980695506, + "mse_channels": [ + 0.6793331980695506, + 0.7211574117476527, + 0.27641515711669723 + ], + "mse_channel_mean": 0.5589685889779669, + "conditional_kl": 45.05801461518774, + "m30_total": 10.37026931917665 + }, + "accepted_moves": 817, + "moved_fraction": 0.99609375, + "mse_local_min_certificate": true + } + ] + }, + "ricci_flow": { + "gate_a": { + "true": { + "mse": 1.4051491852839082, + "mse_channels": [ + 1.4051491852839082, + 1.3668538710233142, + 1.5242251053368214 + ], + "mse_channel_mean": 1.432076053881348, + "conditional_kl": 88.38928639731573, + "m30_total": 20.488155650030965 + }, + "similarity_shift": { + "mse": 1.8434929645412106, + "mse_channels": [ + 1.8434929645412106, + 1.9020926823215523, + 1.6172152259653263 + ], + "mse_channel_mean": 1.7876002909426962, + "conditional_kl": 127.14432896752237, + "m30_total": 29.115851722586896 + }, + "random": { + "mse": { + "mean": 2.001699447631836, + "std": 0.026463132351636887, + "min": 1.9043169021606445, + "true_z": 22.542692184448242, + "true_rank_among_random": 0, + "count": 300 + }, + "mse_channel_mean": { + "mean": 2.00239634513855, + "std": 0.03736656904220581, + "min": 1.8651255369186401, + "true_z": 15.262847900390625, + "true_rank_among_random": 0, + "count": 300 + }, + "conditional_kl": { + "mean": 128.49293518066406, + "std": 1.8770456314086914, + "min": 122.98755645751953, + "true_z": 21.36530113220215, + "true_rank_among_random": 0, + "count": 300 + }, + "m30_total": { + "mean": 29.70198631286621, + "std": 0.38021120429039, + "min": 28.589204788208008, + "true_z": 24.23345375061035, + "true_rank_among_random": 0, + "count": 300 + } + } + }, + "gate_b": { + "pairs": 130816, + "improving_pairs": 18155, + "improving_fraction": 0.13878271771037182, + "delta_mean": 0.003676335561325553, + "delta_min": -0.025990603548186376, + "identity_is_local_min_mse": false, + "worst_improving_swaps": [ + { + "pair": [ + 18, + 434 + ], + "delta_mse": -0.025990603548186376, + "text_cosine": -2.9774845097962728, + "visual_cosine": -1.133327099259649 + }, + { + "pair": [ + 434, + 480 + ], + "delta_mse": -0.025975307861517156, + "text_cosine": -3.114039178762199, + "visual_cosine": -0.3368339809459302 + }, + { + "pair": [ + 18, + 429 + ], + "delta_mse": -0.025434981356270796, + "text_cosine": -2.387755724734445, + "visual_cosine": 0.011925620802733078 + }, + { + "pair": [ + 429, + 480 + ], + "delta_mse": -0.025291857771188333, + "text_cosine": -2.5243103937003712, + "visual_cosine": 0.17385748665262277 + }, + { + "pair": [ + 18, + 150 + ], + "delta_mse": -0.02516733643851989, + "text_cosine": -2.483204015489877, + "visual_cosine": 0.015981710196079017 + }, + { + "pair": [ + 412, + 480 + ], + "delta_mse": -0.024989685441093895, + "text_cosine": -2.7300201086187283, + "visual_cosine": -0.07507519453172946 + }, + { + "pair": [ + 150, + 480 + ], + "delta_mse": -0.024933876269041855, + "text_cosine": -2.8938707356807996, + "visual_cosine": -0.2156701038567486 + }, + { + "pair": [ + 260, + 480 + ], + "delta_mse": -0.02492756324802885, + "text_cosine": -3.1107401902907696, + "visual_cosine": -0.3459107247267495 + }, + { + "pair": [ + 18, + 412 + ], + "delta_mse": -0.024631294847020656, + "text_cosine": -2.62945144632961, + "visual_cosine": -0.7703097916544893 + }, + { + "pair": [ + 18, + 274 + ], + "delta_mse": -0.02365086660992484, + "text_cosine": -1.4382738520423919, + "visual_cosine": -0.2121107390266946 + }, + { + "pair": [ + 332, + 480 + ], + "delta_mse": -0.023601975055273366, + "text_cosine": -1.7204367254837734, + "visual_cosine": 0.48217105555118694 + }, + { + "pair": [ + 274, + 480 + ], + "delta_mse": -0.023497535675822672, + "text_cosine": -1.719199473535242, + "visual_cosine": 0.6924607061677941 + }, + { + "pair": [ + 163, + 480 + ], + "delta_mse": -0.023424340811694765, + "text_cosine": -1.8859516358132917, + "visual_cosine": 0.007990501081113143 + }, + { + "pair": [ + 287, + 480 + ], + "delta_mse": -0.023356257852629833, + "text_cosine": -2.6384502626857556, + "visual_cosine": -0.06380852994038544 + }, + { + "pair": [ + 428, + 480 + ], + "delta_mse": -0.023121782695831043, + "text_cosine": -3.792657363086034, + "visual_cosine": -0.3029774968643093 + }, + { + "pair": [ + 137, + 480 + ], + "delta_mse": -0.02296646067697977, + "text_cosine": -3.363413579447069, + "visual_cosine": -0.20434528627551002 + }, + { + "pair": [ + 18, + 260 + ], + "delta_mse": -0.022905567156601137, + "text_cosine": -2.7000734700998454, + "visual_cosine": -1.2961325222003508 + }, + { + "pair": [ + 18, + 287 + ], + "delta_mse": -0.02282237359306982, + "text_cosine": -2.3081180684473432, + "visual_cosine": -0.7663805148467919 + }, + { + "pair": [ + 51, + 480 + ], + "delta_mse": -0.02267108658225231, + "text_cosine": -2.292090157483019, + "visual_cosine": -0.5664228824116138 + }, + { + "pair": [ + 245, + 480 + ], + "delta_mse": -0.022597103553432054, + "text_cosine": -2.398487490293823, + "visual_cosine": -0.019772494498080283 + } + ] + }, + "descent_from_true": { + "start_accuracy": 1.0, + "final_accuracy": 0.01171875, + "final_objective": 0.6568184917386115, + "final_energy": { + "mse": 0.6568184917386115, + "mse_channels": [ + 0.6568184917386115, + 0.7064255602473566, + 0.36858606940954197 + ], + "mse_channel_mean": 0.5772767071318367, + "conditional_kl": 47.94523600950498, + "m30_total": 10.90268418537822 + }, + "accepted_moves": 795, + "moved_fraction": 0.98828125, + "mse_local_min_certificate": true, + "trajectory": [ + { + "step": 0, + "objective": 1.3865078253833876, + "accuracy": 0.99609375 + }, + { + "step": 50, + "objective": 1.144971903790274, + "accuracy": 0.80859375 + }, + { + "step": 100, + "objective": 1.017350744716747, + "accuracy": 0.640625 + }, + { + "step": 150, + "objective": 0.9105770737296023, + "accuracy": 0.494140625 + }, + { + "step": 200, + "objective": 0.8405232866093433, + "accuracy": 0.380859375 + }, + { + "step": 250, + "objective": 0.7927776757079824, + "accuracy": 0.28515625 + }, + { + "step": 300, + "objective": 0.7572618191781307, + "accuracy": 0.2109375 + }, + { + "step": 350, + "objective": 0.7275088658442106, + "accuracy": 0.15234375 + }, + { + "step": 400, + "objective": 0.7097567099501753, + "accuracy": 0.107421875 + }, + { + "step": 450, + "objective": 0.6951604682695504, + "accuracy": 0.0859375 + }, + { + "step": 500, + "objective": 0.6839051567136586, + "accuracy": 0.0625 + }, + { + "step": 550, + "objective": 0.6747006581572619, + "accuracy": 0.0546875 + }, + { + "step": 600, + "objective": 0.6685819879324174, + "accuracy": 0.041015625 + }, + { + "step": 650, + "objective": 0.6644215622943256, + "accuracy": 0.025390625 + }, + { + "step": 700, + "objective": 0.6610114892029064, + "accuracy": 0.013671875 + }, + { + "step": 750, + "objective": 0.6590320890403556, + "accuracy": 0.013671875 + } + ] + }, + "descent_from_random": [ + { + "start_accuracy": 0.00390625, + "final_accuracy": 0.001953125, + "final_objective": 0.6502435888175013, + "final_energy": { + "mse": 0.6502435888175013, + "mse_channels": [ + 0.6502435888175013, + 0.7015187929205587, + 0.36434377499376225 + ], + "mse_channel_mean": 0.5720353855772741, + "conditional_kl": 47.14436080898222, + "m30_total": 10.729359339431447 + }, + "accepted_moves": 801, + "moved_fraction": 0.998046875, + "mse_local_min_certificate": true + }, + { + "start_accuracy": 0.0, + "final_accuracy": 0.0, + "final_objective": 0.6806427528479334, + "final_energy": { + "mse": 0.6806427528479334, + "mse_channels": [ + 0.6806427528479334, + 0.7224950679081841, + 0.3685288788678386 + ], + "mse_channel_mean": 0.5905555665413187, + "conditional_kl": 45.88555813637643, + "m30_total": 10.538397132971152 + }, + "accepted_moves": 815, + "moved_fraction": 0.9921875, + "mse_local_min_certificate": true + } + ] + } + } +} \ No newline at end of file -- cgit v1.2.3