{ "a0.0_L2_s42": { "bp": { "test_acc": 0.9185, "mean_bp_cosine": 0.8935370743274689, "mean_rho": 0.9999973475933075, "mean_nudge_001": -0.016880770213901997, "mean_nudge_003": -0.046150531619787216, "mean_nudge_01": -0.11240250617265701, "bp_cosine_per_layer": [ 0.9016435146331787, 0.885430634021759 ], "rho_per_layer": [ 0.9999990463256836, 0.9999956488609314 ], "nudge_per_layer": [ -0.1530623733997345, -0.07174263894557953 ] }, "dfa": { "test_acc": 0.842, "mean_bp_cosine": 0.030407493002712727, "mean_rho": 0.008977552875876427, "mean_nudge_001": -0.00023997636890271679, "mean_nudge_003": -0.0006689573347102851, "mean_nudge_01": -0.0016241300618276, "bp_cosine_per_layer": [ -0.003507068380713463, 0.06432205438613892 ], "rho_per_layer": [ 0.009520611725747585, 0.008434494026005268 ], "nudge_per_layer": [ -0.0012172441929578781, -0.002031015930697322 ] }, "state_bridge": { "test_acc": 0.825, "mean_bp_cosine": 0.666242927312851, "mean_rho": 0.5969350039958954, "mean_nudge_001": -0.015545505564659834, "mean_nudge_003": -0.04490915546193719, "mean_nudge_01": -0.13079415168613195, "bp_cosine_per_layer": [ 0.9153801202774048, 0.4171057343482971 ], "rho_per_layer": [ 0.8286964893341064, 0.3651735186576843 ], "nudge_per_layer": [ -0.23136131465435028, -0.030226988717913628 ], "state_pred_error_per_layer": [ 438.86456298828125, 353.2518310546875 ], "mean_state_pred_error": 396.0581970214844 }, "credit_bridge": { "test_acc": 0.6675, "mean_bp_cosine": 0.16703729704022408, "mean_rho": 0.26396652311086655, "mean_nudge_001": -0.00457099755294621, "mean_nudge_003": -0.013636467512696981, "mean_nudge_01": -0.04456503316760063, "bp_cosine_per_layer": [ 0.1011175885796547, 0.23295700550079346 ], "rho_per_layer": [ 0.19994769990444183, 0.32798534631729126 ], "nudge_per_layer": [ -0.07033564895391464, -0.01879441738128662 ], "final_value_loss": 0.5091902822494507, "final_term_loss": 0.2335890076160431, "final_bridge_loss": 0.035112081968784334, "final_tgrad_loss": 0.2404891884803772 } }, "a0.0_L2_s123": { "bp": { "test_acc": 0.9205, "mean_bp_cosine": 0.8566932380199432, "mean_rho": 0.9999960958957672, "mean_nudge_001": -0.015286394860595465, "mean_nudge_003": -0.04188752360641956, "mean_nudge_01": -0.10229892283678055, "bp_cosine_per_layer": [ 0.8634263277053833, 0.8499601483345032 ], "rho_per_layer": [ 0.9999988079071045, 0.9999933838844299 ], "nudge_per_layer": [ -0.13922545313835144, -0.06537239253520966 ] }, "dfa": { "test_acc": 0.841, "mean_bp_cosine": 0.08572041988372803, "mean_rho": 0.10125503316521645, "mean_nudge_001": -0.001035464447340928, "mean_nudge_003": -0.0030792844481766224, "mean_nudge_01": -0.009957430185750127, "bp_cosine_per_layer": [ 0.048809632658958435, 0.12263120710849762 ], "rho_per_layer": [ 0.09689537435770035, 0.10561469197273254 ], "nudge_per_layer": [ -0.015226385556161404, -0.00468847481533885 ] }, "state_bridge": { "test_acc": 0.822, "mean_bp_cosine": 0.7052022516727448, "mean_rho": 0.608213409781456, "mean_nudge_001": -0.014205678016878664, "mean_nudge_003": -0.04078014753758907, "mean_nudge_01": -0.11634222697466612, "bp_cosine_per_layer": [ 0.9309895038604736, 0.47941499948501587 ], "rho_per_layer": [ 0.8439192175865173, 0.37250760197639465 ], "nudge_per_layer": [ -0.2068808674812317, -0.025803586468100548 ], "state_pred_error_per_layer": [ 436.0709533691406, 352.44915771484375 ], "mean_state_pred_error": 394.2600555419922 }, "credit_bridge": { "test_acc": 0.664, "mean_bp_cosine": 0.14723356813192368, "mean_rho": 0.27719707787036896, "mean_nudge_001": -0.0040612792945466936, "mean_nudge_003": -0.012125232955440879, "mean_nudge_01": -0.03972679190337658, "bp_cosine_per_layer": [ 0.08253486454486847, 0.21193227171897888 ], "rho_per_layer": [ 0.16999724507331848, 0.38439691066741943 ], "nudge_per_layer": [ -0.06126694008708, -0.018186643719673157 ], "final_value_loss": 0.6019191103935242, "final_term_loss": 0.28414561631679536, "final_bridge_loss": 0.04602321577072144, "final_tgrad_loss": 0.27175028200149537 } }, "a0.0_L2_s456": { "bp": { "test_acc": 0.9295, "mean_bp_cosine": 0.870507538318634, "mean_rho": 0.9999973773956299, "mean_nudge_001": -0.014503923710435629, "mean_nudge_003": -0.039223295636475086, "mean_nudge_01": -0.09240655042231083, "bp_cosine_per_layer": [ 0.8826531171798706, 0.8583619594573975 ], "rho_per_layer": [ 0.9999988079071045, 0.9999959468841553 ], "nudge_per_layer": [ -0.12340263277292252, -0.06141046807169914 ] }, "dfa": { "test_acc": 0.8485, "mean_bp_cosine": 0.051622094586491585, "mean_rho": 0.014518704265356064, "mean_nudge_001": -0.00044165633153170347, "mean_nudge_003": -0.0013033851282671094, "mean_nudge_01": -0.0040968600660562515, "bp_cosine_per_layer": [ 0.016239378601312637, 0.08700481057167053 ], "rho_per_layer": [ -0.0633457601070404, 0.09238316863775253 ], "nudge_per_layer": [ -0.004778302274644375, -0.003415417857468128 ] }, "state_bridge": { "test_acc": 0.8285, "mean_bp_cosine": 0.7213762998580933, "mean_rho": 0.644424170255661, "mean_nudge_001": -0.014933469472452998, "mean_nudge_003": -0.042839838191866875, "mean_nudge_01": -0.12188584543764591, "bp_cosine_per_layer": [ 0.9387110471725464, 0.5040415525436401 ], "rho_per_layer": [ 0.7997298240661621, 0.4891185164451599 ], "nudge_per_layer": [ -0.21233828365802765, -0.031433407217264175 ], "state_pred_error_per_layer": [ 387.8657531738281, 319.336669921875 ], "mean_state_pred_error": 353.60121154785156 }, "credit_bridge": { "test_acc": 0.7115, "mean_bp_cosine": 0.16614358872175217, "mean_rho": 0.25560564547777176, "mean_nudge_001": -0.0032864182721823454, "mean_nudge_003": -0.009816689416766167, "mean_nudge_01": -0.03221969399601221, "bp_cosine_per_layer": [ 0.08927787840366364, 0.2430092990398407 ], "rho_per_layer": [ 0.16060538589954376, 0.35060590505599976 ], "nudge_per_layer": [ -0.047705113887786865, -0.016734274104237556 ], "final_value_loss": 0.31588549439907077, "final_term_loss": 0.09000624770522117, "final_bridge_loss": 0.023082720217108728, "final_tgrad_loss": 0.20279652795791625 } }, "a0.0_L4_s42": { "bp": { "test_acc": 0.925, "mean_bp_cosine": 0.8320528566837311, "mean_rho": 0.9999772906303406, "mean_nudge_001": -0.009003315295558423, "mean_nudge_003": -0.024674441665410995, "mean_nudge_01": -0.06179796298965812, "bp_cosine_per_layer": [ 0.8599773645401001, 0.8422204256057739, 0.8250665664672852, 0.8009470701217651 ], "rho_per_layer": [ 0.9999991655349731, 0.9999953508377075, 0.9999807476997375, 0.9999338984489441 ], "nudge_per_layer": [ -0.12188813835382462, -0.07056954503059387, -0.034971628338098526, -0.019762540236115456 ] }, "dfa": { "test_acc": 0.856, "mean_bp_cosine": 0.10417622653767467, "mean_rho": 0.09206205606460571, "mean_nudge_001": -0.00045079714618623257, "mean_nudge_003": -0.0013433840940706432, "mean_nudge_01": -0.004383910913020372, "bp_cosine_per_layer": [ 0.004014266654849052, 0.16519895195960999, 0.13244280219078064, 0.11504888534545898 ], "rho_per_layer": [ 0.042726192623376846, 0.18379396200180054, 0.108277827501297, 0.03345024213194847 ], "nudge_per_layer": [ -0.005573366768658161, -0.005793239921331406, -0.003338021691888571, -0.002831015270203352 ] }, "state_bridge": { "test_acc": 0.7615, "mean_bp_cosine": 0.38204737193882465, "mean_rho": 0.3213508538901806, "mean_nudge_001": -0.006838726330897771, "mean_nudge_003": -0.01991741475649178, "mean_nudge_01": -0.059746103768702596, "bp_cosine_per_layer": [ 0.8646283149719238, 0.4090673327445984, 0.1404263973236084, 0.114067442715168 ], "rho_per_layer": [ 0.7375176548957825, 0.33617356419563293, 0.14259827136993408, 0.06911392509937286 ], "nudge_per_layer": [ -0.218303382396698, -0.013712376356124878, -0.003879058640450239, -0.0030895976815372705 ], "state_pred_error_per_layer": [ 6951.521484375, 6099.5322265625, 5877.8310546875, 3832.076416015625 ], "mean_state_pred_error": 5690.240295410156 }, "credit_bridge": { "test_acc": 0.6585, "mean_bp_cosine": 0.10922159999608994, "mean_rho": 0.21513587795197964, "mean_nudge_001": -0.00137018840177916, "mean_nudge_003": -0.004100472666323185, "mean_nudge_01": -0.013550454750657082, "bp_cosine_per_layer": [ 0.049023956060409546, 0.11911389231681824, 0.12341275811195374, 0.14533579349517822 ], "rho_per_layer": [ 0.09633419662714005, 0.23598268628120422, 0.24357257783412933, 0.28465405106544495 ], "nudge_per_layer": [ -0.026128236204385757, -0.008897590450942516, -0.009167088195681572, -0.01000890415161848 ], "final_value_loss": 0.5270826056480408, "final_term_loss": 0.22659661755561827, "final_bridge_loss": 0.04233165686130524, "final_tgrad_loss": 0.25815433073043825 } }, "a0.0_L4_s123": { "bp": { "test_acc": 0.926, "mean_bp_cosine": 0.7713210135698318, "mean_rho": 0.9999768882989883, "mean_nudge_001": -0.010334675083868206, "mean_nudge_003": -0.028318932512775064, "mean_nudge_01": -0.06953301187604666, "bp_cosine_per_layer": [ 0.808335542678833, 0.780368447303772, 0.7579712271690369, 0.7386088371276855 ], "rho_per_layer": [ 0.9999986886978149, 0.999996542930603, 0.9999827146530151, 0.9999296069145203 ], "nudge_per_layer": [ -0.13544964790344238, -0.08136264234781265, -0.039826083928346634, -0.02149367332458496 ] }, "dfa": { "test_acc": 0.8375, "mean_bp_cosine": 0.1401939271017909, "mean_rho": 0.11219900287687778, "mean_nudge_001": -0.000286264817987103, "mean_nudge_003": -0.0008422381652053446, "mean_nudge_01": -0.0026496590580791235, "bp_cosine_per_layer": [ -0.028116505593061447, 0.29581546783447266, 0.170908123254776, 0.12216862291097641 ], "rho_per_layer": [ -0.07312458753585815, 0.3015226125717163, 0.09657395631074905, 0.12382403016090393 ], "nudge_per_layer": [ 0.0038727924693375826, -0.007415304891765118, -0.0039948043413460255, -0.0030613194685429335 ] }, "state_bridge": { "test_acc": 0.788, "mean_bp_cosine": 0.3973395489156246, "mean_rho": 0.36834993585944176, "mean_nudge_001": -0.006569715158548206, "mean_nudge_003": -0.019017601298401132, "mean_nudge_01": -0.055994155758526176, "bp_cosine_per_layer": [ 0.8812471032142639, 0.4430118799209595, 0.14485391974449158, 0.12024529278278351 ], "rho_per_layer": [ 0.803270697593689, 0.4108988642692566, 0.12428833544254303, 0.13494184613227844 ], "nudge_per_layer": [ -0.20551547408103943, -0.01310029998421669, -0.0029572518542408943, -0.0024035971146076918 ], "state_pred_error_per_layer": [ 6307.859375, 4221.2177734375, 3825.8369140625, 3284.16064453125 ], "mean_state_pred_error": 4409.7686767578125 }, "credit_bridge": { "test_acc": 0.7305, "mean_bp_cosine": 0.10989296156913042, "mean_rho": 0.18325010500848293, "mean_nudge_001": -0.0012505464837886393, "mean_nudge_003": -0.00373545300681144, "mean_nudge_01": -0.012258320581167936, "bp_cosine_per_layer": [ 0.03434538468718529, 0.12379913777112961, 0.1433446854352951, 0.13808263838291168 ], "rho_per_layer": [ 0.03924814611673355, 0.21971622109413147, 0.26409071683883667, 0.20994533598423004 ], "nudge_per_layer": [ -0.020018436014652252, -0.00925234891474247, -0.010285570286214352, -0.009476927109062672 ], "final_value_loss": 0.454630423784256, "final_term_loss": 0.185297670173645, "final_bridge_loss": 0.03432008520960808, "final_tgrad_loss": 0.2350126650452614 } }, "a0.0_L4_s456": { "bp": { "test_acc": 0.9285, "mean_bp_cosine": 0.7884698659181595, "mean_rho": 0.999981090426445, "mean_nudge_001": -0.01117145799798891, "mean_nudge_003": -0.030768739874474704, "mean_nudge_01": -0.07730476558208466, "bp_cosine_per_layer": [ 0.8268355131149292, 0.7960893511772156, 0.7726784944534302, 0.758276104927063 ], "rho_per_layer": [ 0.9999993443489075, 0.999996542930603, 0.999983549118042, 0.9999449253082275 ], "nudge_per_layer": [ -0.15056422352790833, -0.08916378021240234, -0.04451071470975876, -0.024980343878269196 ] }, "dfa": { "test_acc": 0.8655, "mean_bp_cosine": 0.06126724602654576, "mean_rho": 0.09146236767992377, "mean_nudge_001": -0.00023738171148579568, "mean_nudge_003": -0.0006991284317336977, "mean_nudge_01": -0.0021681145881302655, "bp_cosine_per_layer": [ -0.012256121262907982, 0.07949584722518921, 0.0787847638130188, 0.09904449433088303 ], "rho_per_layer": [ -0.003950485959649086, 0.06165219843387604, 0.14469987154006958, 0.16344788670539856 ], "nudge_per_layer": [ 0.002696676878258586, -0.005506287328898907, -0.0027312645688652992, -0.003131583333015442 ] }, "state_bridge": { "test_acc": 0.7455, "mean_bp_cosine": 0.4014641270041466, "mean_rho": 0.3732803575694561, "mean_nudge_001": -0.007542004925198853, "mean_nudge_003": -0.022030926978914067, "mean_nudge_01": -0.06684440147364512, "bp_cosine_per_layer": [ 0.8436356782913208, 0.46039146184921265, 0.16277557611465454, 0.13905379176139832 ], "rho_per_layer": [ 0.7277628183364868, 0.48849108815193176, 0.1771586537361145, 0.09970887005329132 ], "nudge_per_layer": [ -0.2401106357574463, -0.018598388880491257, -0.004775169305503368, -0.0038934119511395693 ], "state_pred_error_per_layer": [ 3442.556884765625, 2195.27490234375, 2035.546875, 2293.3291015625 ], "mean_state_pred_error": 2491.6769409179688 }, "credit_bridge": { "test_acc": 0.644, "mean_bp_cosine": 0.1536249229684472, "mean_rho": 0.26861424930393696, "mean_nudge_001": -0.0018619927868712693, "mean_nudge_003": -0.005573431379161775, "mean_nudge_01": -0.018431853968650103, "bp_cosine_per_layer": [ 0.053475264459848404, 0.17959368228912354, 0.19590041041374207, 0.18553033471107483 ], "rho_per_layer": [ 0.04320300370454788, 0.30398592352867126, 0.35316282510757446, 0.3741052448749542 ], "nudge_per_layer": [ -0.027841825038194656, -0.015361151657998562, -0.015351004898548126, -0.015173434279859066 ], "final_value_loss": 0.7469501955509186, "final_term_loss": 0.4142983556270599, "final_bridge_loss": 0.05541581681370735, "final_tgrad_loss": 0.2772360302686691 } }, "a0.0_L8_s42": { "bp": { "test_acc": 0.906, "mean_bp_cosine": 0.8584278598427773, "mean_rho": 0.9999460652470589, "mean_nudge_001": -0.008581212867284194, "mean_nudge_003": -0.02450006199069321, "mean_nudge_01": -0.06902276305481791, "bp_cosine_per_layer": [ 0.8854324817657471, 0.8737501502037048, 0.8630377054214478, 0.8562113046646118, 0.8509326577186584, 0.8481521606445312, 0.84560626745224, 0.8443001508712769 ], "rho_per_layer": [ 0.9999991655349731, 0.9999963045120239, 0.9999881386756897, 0.9999630451202393, 0.9999364614486694, 0.9999151825904846, 0.9998926520347595, 0.9998775720596313 ], "nudge_per_layer": [ -0.21767401695251465, -0.12408202886581421, -0.06762672960758209, -0.04208396375179291, -0.031123068183660507, -0.025745777413249016, -0.022784877568483353, -0.021061642095446587 ] }, "dfa": { "test_acc": 0.8325, "mean_bp_cosine": 0.0621855678036809, "mean_rho": 0.08102096850052476, "mean_nudge_001": 7.459322478098329e-05, "mean_nudge_003": 0.00026162770882365294, "mean_nudge_01": 0.0013351911620702595, "bp_cosine_per_layer": [ -0.02968163788318634, 0.08609121292829514, 0.09377065300941467, 0.10205866396427155, 0.08807726949453354, 0.053033627569675446, 0.056919872760772705, 0.04721488058567047 ], "rho_per_layer": [ -0.09032663702964783, 0.17043988406658173, 0.18389353156089783, 0.16770029067993164, 0.1504741907119751, 0.05179755389690399, 0.050898581743240356, -0.03670964762568474 ], "nudge_per_layer": [ 0.028243789449334145, -0.00520662497729063, -0.0028999613132327795, -0.002869710559025407, -0.0023996694944798946, -0.0015899656573310494, -0.0014181910082697868, -0.0011781371431425214 ] }, "state_bridge": { "test_acc": 0.726, "mean_bp_cosine": 0.26912851445376873, "mean_rho": 0.2629122715443373, "mean_nudge_001": -0.0034038230442092754, "mean_nudge_003": -0.010012732280301861, "mean_nudge_01": -0.0311450962908566, "bp_cosine_per_layer": [ 0.8151727914810181, 0.4042307734489441, 0.16176781058311462, 0.1698867678642273, 0.14891386032104492, 0.1726662963628769, 0.1322125643491745, 0.14817725121974945 ], "rho_per_layer": [ 0.7261759042739868, 0.36915943026542664, 0.17840053141117096, 0.16105638444423676, 0.14931628108024597, 0.15696927905082703, 0.17820878326892853, 0.1840115785598755 ], "nudge_per_layer": [ -0.21425020694732666, -0.011446774937212467, -0.004181142896413803, -0.004114024806767702, -0.0038589704781770706, -0.00422251783311367, -0.003358659567311406, -0.003728472860530019 ], "state_pred_error_per_layer": [ 7398.20849609375, 5242.7607421875, 3768.9609375, 3258.12548828125, 3060.2021484375, 2786.609619140625, 3211.44921875, 2939.81005859375 ], "mean_state_pred_error": 3958.265838623047 }, "credit_bridge": { "test_acc": 0.64, "mean_bp_cosine": 0.09559969999827445, "mean_rho": 0.20091960858553648, "mean_nudge_001": -0.0007423463721352164, "mean_nudge_003": -0.0022244660067372024, "mean_nudge_01": -0.007368246268015355, "bp_cosine_per_layer": [ 0.010387556627392769, 0.06787668168544769, 0.11108389496803284, 0.11367877572774887, 0.11491300165653229, 0.1093105673789978, 0.11488444358110428, 0.12266267836093903 ], "rho_per_layer": [ 0.0504305437207222, 0.11780044436454773, 0.23789767920970917, 0.24238067865371704, 0.22429296374320984, 0.25153419375419617, 0.25647929310798645, 0.22654107213020325 ], "nudge_per_layer": [ -0.003928218502551317, -0.0057389335706830025, -0.007835019379854202, -0.008116443641483784, -0.0081653306260705, -0.00813683308660984, -0.008312376216053963, -0.00871281512081623 ], "final_value_loss": 0.6615465266227722, "final_term_loss": 0.43013124598264696, "final_bridge_loss": 0.0357342723429203, "final_tgrad_loss": 0.19568100452423096 } }, "a0.0_L8_s123": { "bp": { "test_acc": 0.913, "mean_bp_cosine": 0.8645070418715477, "mean_rho": 0.999944381415844, "mean_nudge_001": -0.007505885470891371, "mean_nudge_003": -0.021483065967913717, "mean_nudge_01": -0.06040166970342398, "bp_cosine_per_layer": [ 0.9087610840797424, 0.8890509605407715, 0.8756387233734131, 0.860000729560852, 0.8513467311859131, 0.8465403914451599, 0.8431664705276489, 0.8415512442588806 ], "rho_per_layer": [ 0.9999994039535522, 0.9999971389770508, 0.9999857544898987, 0.9999687671661377, 0.9999316930770874, 0.9999115467071533, 0.9998936653137207, 0.9998670816421509 ], "nudge_per_layer": [ -0.19297254085540771, -0.10940254479646683, -0.05867426097393036, -0.03609157353639603, -0.026526644825935364, -0.02192458137869835, -0.019491419196128845, -0.01812979206442833 ] }, "dfa": { "test_acc": 0.849, "mean_bp_cosine": 0.07409168453887105, "mean_rho": 0.0943274952005595, "mean_nudge_001": -6.366873094520997e-06, "mean_nudge_003": -1.1956191883655265e-05, "mean_nudge_01": 5.2888659411109984e-05, "bp_cosine_per_layer": [ -0.02848462387919426, 0.15189720690250397, 0.09141203761100769, 0.07272940874099731, 0.09136377274990082, 0.0857112780213356, 0.044680722057819366, 0.0834236741065979 ], "rho_per_layer": [ 0.03057212010025978, 0.16072696447372437, 0.13512857258319855, 0.0885201245546341, 0.1377822607755661, 0.04239961504936218, 0.0035119224339723587, 0.15597838163375854 ], "nudge_per_layer": [ 0.015482441522181034, -0.004561400040984154, -0.0020816801115870476, -0.0015152986161410809, -0.0019329312490299344, -0.002064004773274064, -0.001068722689524293, -0.0018352947663515806 ] }, "state_bridge": { "test_acc": 0.7725, "mean_bp_cosine": 0.2933046016842127, "mean_rho": 0.25743963569402695, "mean_nudge_001": -0.0033750918046280276, "mean_nudge_003": -0.009846233486314304, "mean_nudge_01": -0.02971906404127367, "bp_cosine_per_layer": [ 0.8679102659225464, 0.46438825130462646, 0.22033852338790894, 0.1893446445465088, 0.16738475859165192, 0.14945879578590393, 0.1267041563987732, 0.16090741753578186 ], "rho_per_layer": [ 0.6897462606430054, 0.42848390340805054, 0.19369670748710632, 0.19582855701446533, 0.14048044383525848, 0.12853404879570007, 0.09809397161006927, 0.18465319275856018 ], "nudge_per_layer": [ -0.2050759196281433, -0.01248687319457531, -0.004478689283132553, -0.003817938268184662, -0.00330835347995162, -0.002980519086122513, -0.002494834130629897, -0.003109385259449482 ], "state_pred_error_per_layer": [ 7121.279296875, 6133.4482421875, 3940.991943359375, 2496.587646484375, 2560.85595703125, 2498.587890625, 3539.93115234375, 2449.22607421875 ], "mean_state_pred_error": 3842.613525390625 }, "credit_bridge": { "test_acc": 0.656, "mean_bp_cosine": 0.092392878344981, "mean_rho": 0.17488694936037064, "mean_nudge_001": -0.0007325447659241036, "mean_nudge_003": -0.002192195039242506, "mean_nudge_01": -0.0072501768881920725, "bp_cosine_per_layer": [ 0.001700198045000434, 0.04125010222196579, 0.1112770289182663, 0.11907510459423065, 0.11768756806850433, 0.11733907461166382, 0.11329221725463867, 0.117521733045578 ], "rho_per_layer": [ 0.043069981038570404, 0.1044735386967659, 0.23828034102916718, 0.18111932277679443, 0.19782572984695435, 0.1971191018819809, 0.20729684829711914, 0.2299107313156128 ], "nudge_per_layer": [ -0.006082434207201004, -0.003400696674361825, -0.007903593592345715, -0.008138573728501797, -0.008289994671940804, -0.007994470186531544, -0.00807441771030426, -0.008117234334349632 ], "final_value_loss": 0.5009399848937989, "final_term_loss": 0.26505257573127744, "final_bridge_loss": 0.027039055764675142, "final_tgrad_loss": 0.20884835484027864 } }, "a0.0_L8_s456": { "bp": { "test_acc": 0.8965, "mean_bp_cosine": 0.8864798843860626, "mean_rho": 0.9999493733048439, "mean_nudge_001": -0.007212058277218603, "mean_nudge_003": -0.02048966591246426, "mean_nudge_01": -0.05707334354519844, "bp_cosine_per_layer": [ 0.9283311367034912, 0.9162832498550415, 0.8925424814224243, 0.8799441456794739, 0.8746041655540466, 0.8698863983154297, 0.8666130304336548, 0.863634467124939 ], "rho_per_layer": [ 0.9999991059303284, 0.9999967217445374, 0.9999891519546509, 0.9999651908874512, 0.9999410510063171, 0.9999043941497803, 0.9998966455459595, 0.9999027252197266 ], "nudge_per_layer": [ -0.17811693251132965, -0.10346511751413345, -0.05678309500217438, -0.035222865641117096, -0.025761041790246964, -0.021198198199272156, -0.01872769370675087, -0.017311803996562958 ] }, "dfa": { "test_acc": 0.833, "mean_bp_cosine": 0.10069436486810446, "mean_rho": 0.11729028820991516, "mean_nudge_001": -0.00045204991511127446, "mean_nudge_003": -0.0013379494193941355, "mean_nudge_01": -0.004276915889931843, "bp_cosine_per_layer": [ 0.020978286862373352, 0.16464680433273315, 0.15073026716709137, 0.08116275072097778, 0.08070477098226547, 0.0867747962474823, 0.10854049026966095, 0.11201675236225128 ], "rho_per_layer": [ 0.032643549144268036, 0.2071959227323532, 0.2057623416185379, 0.09326457977294922, 0.06205561012029648, 0.1389445811510086, 0.11340519785881042, 0.08505052328109741 ], "nudge_per_layer": [ -0.012655803002417088, -0.005697320215404034, -0.004094945732504129, -0.0018561023753136396, -0.0020380420610308647, -0.0022426587529480457, -0.002856952603906393, -0.0027735023759305477 ] }, "state_bridge": { "test_acc": 0.7495, "mean_bp_cosine": 0.246165806427598, "mean_rho": 0.20948901865631342, "mean_nudge_001": -0.0035059251604252495, "mean_nudge_003": -0.010245608449622523, "mean_nudge_01": -0.03114599542459473, "bp_cosine_per_layer": [ 0.8467778563499451, 0.3549606502056122, 0.1711018681526184, 0.12375819683074951, 0.1153455525636673, 0.12219913303852081, 0.12895643711090088, 0.10622675716876984 ], "rho_per_layer": [ 0.7416437268257141, 0.2796669900417328, 0.14060860872268677, 0.09820520132780075, 0.10454020649194717, 0.08740514516830444, 0.10375625640153885, 0.12008601427078247 ], "nudge_per_layer": [ -0.2212277054786682, -0.010068328119814396, -0.00393392750993371, -0.002955838106572628, -0.0027290289290249348, -0.0028208147268742323, -0.002979037119075656, -0.002453283406794071 ], "state_pred_error_per_layer": [ 10942.7939453125, 10293.740234375, 7482.4482421875, 5008.9619140625, 4863.818359375, 4492.451171875, 3565.2744140625, 3837.47998046875 ], "mean_state_pred_error": 6310.871032714844 }, "credit_bridge": { "test_acc": 0.617, "mean_bp_cosine": 0.09351247036829591, "mean_rho": 0.15350601403042674, "mean_nudge_001": -0.0010894104561884888, "mean_nudge_003": -0.0032592675706837326, "mean_nudge_01": -0.010775732051115483, "bp_cosine_per_layer": [ 0.03283574804663658, 0.05836774408817291, 0.11590129137039185, 0.10100162029266357, 0.1079024076461792, 0.10276803374290466, 0.11581481993198395, 0.11350809782743454 ], "rho_per_layer": [ 0.06095704063773155, 0.13836193084716797, 0.1803440898656845, 0.1439131498336792, 0.14064353704452515, 0.15858328342437744, 0.22606094181537628, 0.17918413877487183 ], "nudge_per_layer": [ -0.023668643087148666, -0.006166210863739252, -0.009575091302394867, -0.00891521479934454, -0.00932026281952858, -0.008899858221411705, -0.009868312627077103, -0.009792262688279152 ], "final_value_loss": 1.2871107803344726, "final_term_loss": 0.9274139808654785, "final_bridge_loss": 0.0815536917924881, "final_tgrad_loss": 0.2781431001186371 } }, "a0.0_L12_s42": { "bp": { "test_acc": 0.9135, "mean_bp_cosine": 0.8262566576401392, "mean_rho": 0.9998530795176824, "mean_nudge_001": -0.005460524485291292, "mean_nudge_003": -0.015294013021048158, "mean_nudge_01": -0.04075813665986061, "bp_cosine_per_layer": [ 0.8936765193939209, 0.875713586807251, 0.8523612022399902, 0.835299551486969, 0.8243429064750671, 0.8148068189620972, 0.809612512588501, 0.8055970668792725, 0.8028815984725952, 0.8011909127235413, 0.8001247644424438, 0.799472451210022 ], "rho_per_layer": [ 0.9999988079071045, 0.999997615814209, 0.9999872446060181, 0.9999661445617676, 0.9999244213104248, 0.9998762011528015, 0.9998204708099365, 0.9997811317443848, 0.9997683167457581, 0.9997478723526001, 0.9997052550315857, 0.9996634721755981 ], "nudge_per_layer": [ -0.15880361199378967, -0.10216463357210159, -0.058997806161642075, -0.0368044376373291, -0.02622675709426403, -0.02074911817908287, -0.017529264092445374, -0.015481802634894848, -0.014148212037980556, -0.013247357681393623, -0.012655021622776985, -0.012289617210626602 ] }, "dfa": { "test_acc": 0.8405, "mean_bp_cosine": 0.06366231696059306, "mean_rho": 0.0974447913467884, "mean_nudge_001": -0.00019441865091115082, "mean_nudge_003": -0.0005761026429051223, "mean_nudge_01": -0.001826463839582478, "bp_cosine_per_layer": [ 0.022819485515356064, 0.10611869394779205, 0.08763782680034637, 0.06361521780490875, 0.08818645775318146, 0.07573625445365906, 0.07574683427810669, 0.025106821209192276, 0.029594242572784424, 0.09018446505069733, 0.04191981628537178, 0.05728168785572052 ], "rho_per_layer": [ -0.0498071163892746, 0.18554872274398804, 0.19663047790527344, 0.14150168001651764, 0.06663499772548676, 0.12039551883935928, 0.12434913218021393, 0.05157296732068062, 0.09405694901943207, 0.07376686483621597, 0.061272408813238144, 0.10341489315032959 ], "nudge_per_layer": [ 0.000284646637737751, -0.005279931705445051, -0.002400151453912258, -0.0017087131273001432, -0.0019207102013751864, -0.0020927595905959606, -0.0017691099783405662, -0.0008208724320866168, -0.0012333606136962771, -0.0026537002995610237, -0.0007275568204931915, -0.0015953464899212122 ] }, "state_bridge": { "test_acc": 0.675, "mean_bp_cosine": 0.16806975565850735, "mean_rho": 0.14302178906897703, "mean_nudge_001": -0.0021900876042006225, "mean_nudge_003": -0.006470390944741666, "mean_nudge_01": -0.020386384644856054, "bp_cosine_per_layer": [ 0.7431880235671997, 0.2796655595302582, 0.13151727616786957, 0.09881292283535004, 0.0924338698387146, 0.10159576684236526, 0.09764175117015839, 0.09642879664897919, 0.09804226458072662, 0.09603762626647949, 0.09261822700500488, 0.08885498344898224 ], "rho_per_layer": [ 0.611870527267456, 0.24307268857955933, 0.10120554268360138, 0.09459854662418365, 0.09521515667438507, 0.07265714555978775, 0.07570740580558777, 0.0650622695684433, 0.07738275825977325, 0.11346932500600815, 0.08958262950181961, 0.07643747329711914 ], "nudge_per_layer": [ -0.20912763476371765, -0.00893338117748499, -0.0035844063386321068, -0.0027289288118481636, -0.0024399380199611187, -0.0026743747293949127, -0.0025294050574302673, -0.002557961270213127, -0.0026802942156791687, -0.002547960262745619, -0.002460954710841179, -0.0023713763803243637 ], "state_pred_error_per_layer": [ 20223.798828125, 14219.4013671875, 11048.2333984375, 10263.419921875, 8570.201171875, 7224.33740234375, 7281.2197265625, 8034.87548828125, 6686.9033203125, 6885.98291015625, 6667.048828125, 6627.78125 ], "mean_state_pred_error": 9477.766967773438 }, "credit_bridge": { "test_acc": 0.6635, "mean_bp_cosine": 0.07870348079207663, "mean_rho": 0.1263093726399044, "mean_nudge_001": -0.000519128576949394, "mean_nudge_003": -0.0015524110640399158, "mean_nudge_01": -0.005132454059397181, "bp_cosine_per_layer": [ 9.498395957052708e-05, 0.029415149241685867, 0.062070880085229874, 0.07393385469913483, 0.07970248907804489, 0.08624569326639175, 0.10002343356609344, 0.10488344728946686, 0.10920242965221405, 0.10451158881187439, 0.09929930418729782, 0.09505851566791534 ], "rho_per_layer": [ 0.025425152853131294, 0.0574759915471077, 0.13417154550552368, 0.10034967213869095, 0.17468321323394775, 0.17164339125156403, 0.15925100445747375, 0.08668951690196991, 0.1521437168121338, 0.14672580361366272, 0.1825145184993744, 0.12463894486427307 ], "nudge_per_layer": [ 0.001420565415173769, -0.0026360084302723408, -0.00476992130279541, -0.005131378769874573, -0.005615689791738987, -0.006001932080835104, -0.006417973432689905, -0.006630052346736193, -0.006984136067330837, -0.0066003259271383286, -0.00623705331236124, -0.005985542666167021 ], "final_value_loss": 0.9863593735218048, "final_term_loss": 0.7168336891174316, "final_bridge_loss": 0.05245816183909774, "final_tgrad_loss": 0.21706753172874452 } }, "a0.0_L12_s123": { "bp": { "test_acc": 0.9145, "mean_bp_cosine": 0.7868696798880895, "mean_rho": 0.9998756001393, "mean_nudge_001": -0.005259950548255195, "mean_nudge_003": -0.014897353908357521, "mean_nudge_01": -0.04119087569415569, "bp_cosine_per_layer": [ 0.832116425037384, 0.8141677379608154, 0.8023761510848999, 0.7934653759002686, 0.7857377529144287, 0.7797414064407349, 0.7767306566238403, 0.7744640111923218, 0.7725205421447754, 0.7711977958679199, 0.7703278064727783, 0.7695904970169067 ], "rho_per_layer": [ 0.9999991655349731, 0.9999972581863403, 0.9999898672103882, 0.999970555305481, 0.9999305009841919, 0.9999109506607056, 0.9998676180839539, 0.999832808971405, 0.9997924566268921, 0.9997202754020691, 0.9997268319129944, 0.9997689127922058 ], "nudge_per_layer": [ -0.17892791330814362, -0.10439710319042206, -0.05659153312444687, -0.034265220165252686, -0.02416030317544937, -0.01895919442176819, -0.015955276787281036, -0.01404004916548729, -0.01277724839746952, -0.011922692880034447, -0.011335920542478561, -0.010958053171634674 ] }, "dfa": { "test_acc": 0.841, "mean_bp_cosine": 0.10045036673545837, "mean_rho": 0.06510928040370345, "mean_nudge_001": -0.000200933368432743, "mean_nudge_003": -0.000600688594810587, "mean_nudge_01": -0.001976225830730982, "bp_cosine_per_layer": [ 0.009534038603305817, 0.2771005630493164, 0.12846891582012177, 0.12727834284305573, 0.09264838695526123, 0.08494879305362701, 0.1454543173313141, 0.07647830247879028, 0.055110231041908264, 0.08491981029510498, 0.054209403693675995, 0.06925329566001892 ], "rho_per_layer": [ -0.030354971066117287, 0.19958093762397766, 0.10430362075567245, 0.06749340891838074, 0.08917459845542908, 0.06985980272293091, 0.08990536630153656, -0.0003178045153617859, -0.05352415516972542, 0.12398940324783325, 0.05844062566757202, 0.06276053190231323 ], "nudge_per_layer": [ 0.0008368014823645353, -0.00641840323805809, -0.0027151124086230993, -0.002792379120364785, -0.0018645611125975847, -0.0017843465320765972, -0.002957880962640047, -0.0012784524587914348, -0.0007560949889011681, -0.0015581869520246983, -0.0010259757982566953, -0.0014001178788021207 ] }, "state_bridge": { "test_acc": 0.675, "mean_bp_cosine": 0.21027903320888677, "mean_rho": 0.16716737858951092, "mean_nudge_001": -0.002424063318661259, "mean_nudge_003": -0.007164865601225756, "mean_nudge_01": -0.02264038595603779, "bp_cosine_per_layer": [ 0.7589784860610962, 0.34465640783309937, 0.20695620775222778, 0.15412747859954834, 0.14041899144649506, 0.152847021818161, 0.1464981734752655, 0.13855373859405518, 0.13026131689548492, 0.11798880249261856, 0.11249798536300659, 0.11956378817558289 ], "rho_per_layer": [ 0.6762608289718628, 0.29351934790611267, 0.15623700618743896, 0.07772233337163925, 0.09808109700679779, 0.14218339323997498, 0.09985582530498505, 0.07779418677091599, 0.10923020541667938, 0.10118008404970169, 0.08211226016283035, 0.09183197468519211 ], "nudge_per_layer": [ -0.22324970364570618, -0.01123635284602642, -0.0056420015171170235, -0.0040974924340844154, -0.0037027131766080856, -0.003983458504080772, -0.0038626876194030046, -0.0036268262192606926, -0.003363188588991761, -0.0030670221894979477, -0.0028224957641214132, -0.0030306889675557613 ], "state_pred_error_per_layer": [ 10215.58203125, 8926.0654296875, 5285.6044921875, 5192.0830078125, 4575.35791015625, 4063.82666015625, 4966.6884765625, 5378.6435546875, 4702.37890625, 4026.143310546875, 3985.35595703125, 3780.156494140625 ], "mean_state_pred_error": 5424.8238525390625 }, "credit_bridge": { "test_acc": 0.721, "mean_bp_cosine": 0.020079606105961528, "mean_rho": 0.0501531979146724, "mean_nudge_001": -0.00015450727914867457, "mean_nudge_003": -0.0004599024711448389, "mean_nudge_01": -0.001498352525231894, "bp_cosine_per_layer": [ -0.0035465662367641926, -0.001160490675829351, 0.00486466521397233, 0.012821277603507042, 0.012568866834044456, 0.019143100827932358, 0.022381767630577087, 0.024246837943792343, 0.027966029942035675, 0.03333558514714241, 0.04247187077999115, 0.04586232826113701 ], "rho_per_layer": [ 0.01447216235101223, -0.01648387312889099, 0.027175014838576317, 0.03045208379626274, 0.05211577191948891, -0.0074118818156421185, 0.0508866012096405, 0.01983593963086605, 0.03707783669233322, 0.12354743480682373, 0.16596439480781555, 0.10420688986778259 ], "nudge_per_layer": [ 0.002342540305107832, -0.0004765552293974906, -0.0007624666905030608, -0.0012423819862306118, -0.0012568822130560875, -0.0016762344166636467, -0.0018269497668370605, -0.0020500137470662594, -0.0022837575525045395, -0.0026187507901340723, -0.0029940432868897915, -0.0031347349286079407 ], "final_value_loss": 0.37253577736616134, "final_term_loss": 0.18445234730243684, "final_bridge_loss": 0.0251624245762825, "final_tgrad_loss": 0.16292100727558137 } }, "a0.0_L12_s456": { "bp": { "test_acc": 0.9115, "mean_bp_cosine": 0.7958964506785074, "mean_rho": 0.9998697290817896, "mean_nudge_001": -0.0046546936306792, "mean_nudge_003": -0.013125322526320815, "mean_nudge_01": -0.03561708889901638, "bp_cosine_per_layer": [ 0.841245174407959, 0.830803632736206, 0.8195754885673523, 0.8066799640655518, 0.7983194589614868, 0.7908304929733276, 0.7860588431358337, 0.7811387777328491, 0.7771316170692444, 0.7744728326797485, 0.7727954983711243, 0.7717056274414062 ], "rho_per_layer": [ 0.9999991059303284, 0.9999973177909851, 0.9999889731407166, 0.9999734163284302, 0.9999427795410156, 0.9998763799667358, 0.9998629093170166, 0.9998323321342468, 0.9997599720954895, 0.9997479319572449, 0.9997410774230957, 0.9997145533561707 ], "nudge_per_layer": [ -0.1470111906528473, -0.09042022377252579, -0.050197783857584, -0.030754856765270233, -0.021816428750753403, -0.017199307680130005, -0.014493460766971111, -0.012762569822371006, -0.01160583458840847, -0.010828070342540741, -0.010316012427210808, -0.00999932736158371 ] }, "dfa": { "test_acc": 0.8275, "mean_bp_cosine": 0.109907116740942, "mean_rho": 0.10316232999321073, "mean_nudge_001": -0.00034242031688336283, "mean_nudge_003": -0.001020948082441464, "mean_nudge_01": -0.0033549320845243833, "bp_cosine_per_layer": [ 0.0384417325258255, 0.27492010593414307, 0.17608770728111267, 0.12729278206825256, 0.09575799107551575, 0.13158489763736725, 0.07343505322933197, 0.08012130856513977, 0.08938327431678772, 0.09414967149496078, 0.06517850607633591, 0.07253237068653107 ], "rho_per_layer": [ -0.00336202559992671, 0.18004602193832397, 0.18598641455173492, 0.15388232469558716, 0.058810293674468994, 0.14123386144638062, 0.05320374667644501, 0.10196533054113388, 0.09802494943141937, 0.10918877273797989, 0.09344048798084259, 0.0655277818441391 ], "nudge_per_layer": [ -0.010087584145367146, -0.006617182400077581, -0.004152917303144932, -0.0031182561069726944, -0.0022262579295784235, -0.0028664623387157917, -0.0016081882640719414, -0.0017378728371113539, -0.0020772137213498354, -0.002457245485857129, -0.0015545324422419071, -0.0017554720398038626 ] }, "state_bridge": { "test_acc": 0.684, "mean_bp_cosine": 0.21628577137986818, "mean_rho": 0.17074896084765592, "mean_nudge_001": -0.0019644595328524397, "mean_nudge_003": -0.0058011185271122185, "mean_nudge_01": -0.018295541075834382, "bp_cosine_per_layer": [ 0.7363088726997375, 0.32443851232528687, 0.21702790260314941, 0.21007081866264343, 0.16975754499435425, 0.1722516119480133, 0.16474759578704834, 0.12138545513153076, 0.1178494393825531, 0.13036376237869263, 0.11493153870105743, 0.1162962019443512 ], "rho_per_layer": [ 0.6429969072341919, 0.2531689405441284, 0.22036504745483398, 0.17542049288749695, 0.15819020569324493, 0.14438438415527344, 0.07042165100574493, 0.11963094770908356, 0.1070300042629242, 0.07622774690389633, 0.035559043288230896, 0.045592159032821655 ], "nudge_per_layer": [ -0.1725819855928421, -0.009430286474525928, -0.0054127369076013565, -0.005186344031244516, -0.004062540829181671, -0.0039931111969053745, -0.003901706077158451, -0.003083862829953432, -0.002966922242194414, -0.0033692335709929466, -0.002796228276565671, -0.002761534880846739 ], "state_pred_error_per_layer": [ 8107.484375, 6015.64697265625, 3272.651123046875, 3025.219970703125, 3498.379150390625, 2962.442626953125, 2732.457275390625, 3394.069091796875, 3050.444091796875, 3397.53271484375, 3068.22607421875, 3145.807373046875 ], "mean_state_pred_error": 3805.8634033203125 }, "credit_bridge": { "test_acc": 0.669, "mean_bp_cosine": 0.048155296875241525, "mean_rho": 0.07334851861620943, "mean_nudge_001": -0.00035685058537637815, "mean_nudge_003": -0.0010696607399343823, "mean_nudge_01": -0.0035326933624067656, "bp_cosine_per_layer": [ 0.0015124600613489747, 0.020085982978343964, 0.03127610310912132, 0.03517032414674759, 0.042219605296850204, 0.05027471110224724, 0.05683741718530655, 0.057938531041145325, 0.06581299006938934, 0.06761687994003296, 0.07107532024383545, 0.07804323732852936 ], "rho_per_layer": [ 0.03147998824715614, -0.015333062037825584, 0.020876668393611908, 0.07304719090461731, 0.0599740669131279, 0.08874240517616272, 0.05341971293091774, 0.07721589505672455, 0.08801877498626709, 0.10527393221855164, 0.16833831369876862, 0.1291283369064331 ], "nudge_per_layer": [ -0.004048495087772608, -0.0012588155223056674, -0.0019918864127248526, -0.002725755563005805, -0.0028244066052138805, -0.003425109200179577, -0.0036919033154845238, -0.003999053034931421, -0.0042928787879645824, -0.004538967739790678, -0.004742524586617947, -0.004852524492889643 ], "final_value_loss": 0.8257113872528076, "final_term_loss": 0.5509136886596679, "final_bridge_loss": 0.049323998257517813, "final_tgrad_loss": 0.22547370266914368 } }, "a0.25_L2_s42": { "bp": { "test_acc": 0.908, "mean_bp_cosine": 0.8980661034584045, "mean_rho": 0.9999964535236359, "mean_nudge_001": -0.01618960639461875, "mean_nudge_003": -0.04483883548527956, "mean_nudge_01": -0.11487101018428802, "bp_cosine_per_layer": [ 0.9068200588226318, 0.8893121480941772 ], "rho_per_layer": [ 0.9999985098838806, 0.9999943971633911 ], "nudge_per_layer": [ -0.15680165588855743, -0.07294036448001862 ] }, "dfa": { "test_acc": 0.847, "mean_bp_cosine": 0.026606750674545765, "mean_rho": -0.0022891219705343246, "mean_nudge_001": -2.3714659619145095e-05, "mean_nudge_003": -1.6182020772248507e-05, "mean_nudge_01": 0.0005980018759146333, "bp_cosine_per_layer": [ -0.008382977917790413, 0.06159647926688194 ], "rho_per_layer": [ -0.022914212197065353, 0.018335968255996704 ], "nudge_per_layer": [ 0.0033093439415097237, -0.002113340189680457 ] }, "state_bridge": { "test_acc": 0.827, "mean_bp_cosine": 0.6775506436824799, "mean_rho": 0.6271905303001404, "mean_nudge_001": -0.015066297259181738, "mean_nudge_003": -0.04329914692789316, "mean_nudge_01": -0.12377128005027771, "bp_cosine_per_layer": [ 0.9188902378082275, 0.4362110495567322 ], "rho_per_layer": [ 0.8401009440422058, 0.41428011655807495 ], "nudge_per_layer": [ -0.21947401762008667, -0.02806854248046875 ], "state_pred_error_per_layer": [ 753.1610107421875, 333.87890625 ], "mean_state_pred_error": 543.5199584960938 }, "credit_bridge": { "test_acc": 0.6455, "mean_bp_cosine": 0.15288914740085602, "mean_rho": 0.2540942281484604, "mean_nudge_001": -0.003834493225440383, "mean_nudge_003": -0.011456333566457033, "mean_nudge_01": -0.03762407600879669, "bp_cosine_per_layer": [ 0.07500731945037842, 0.23077097535133362 ], "rho_per_layer": [ 0.12894567847251892, 0.37924277782440186 ], "nudge_per_layer": [ -0.05425257235765457, -0.020995579659938812 ], "final_value_loss": 0.5630231554031372, "final_term_loss": 0.2533654543399811, "final_bridge_loss": 0.03314736911058426, "final_tgrad_loss": 0.27651033034324646 } }, "a0.25_L2_s123": { "bp": { "test_acc": 0.92, "mean_bp_cosine": 0.8535160422325134, "mean_rho": 0.9999972283840179, "mean_nudge_001": -0.01451590727083385, "mean_nudge_003": -0.03988213837146759, "mean_nudge_01": -0.09831684827804565, "bp_cosine_per_layer": [ 0.8569881916046143, 0.8500438928604126 ], "rho_per_layer": [ 0.9999989867210388, 0.9999954700469971 ], "nudge_per_layer": [ -0.13159069418907166, -0.06504300236701965 ] }, "dfa": { "test_acc": 0.853, "mean_bp_cosine": 0.0877363346517086, "mean_rho": 0.0626654140651226, "mean_nudge_001": -0.0009488673531450331, "mean_nudge_003": -0.0028224033303558826, "mean_nudge_01": -0.009134252090007067, "bp_cosine_per_layer": [ 0.05345926433801651, 0.1220134049654007 ], "rho_per_layer": [ 0.09254007041454315, 0.03279075771570206 ], "nudge_per_layer": [ -0.0142319705337286, -0.004036533646285534 ] }, "state_bridge": { "test_acc": 0.834, "mean_bp_cosine": 0.6950189173221588, "mean_rho": 0.5849489569664001, "mean_nudge_001": -0.012975628254935145, "mean_nudge_003": -0.03723907680250704, "mean_nudge_01": -0.10619704332202673, "bp_cosine_per_layer": [ 0.9177796840667725, 0.47225815057754517 ], "rho_per_layer": [ 0.7912815809249878, 0.3786163330078125 ], "nudge_per_layer": [ -0.187799870967865, -0.02459421567618847 ], "state_pred_error_per_layer": [ 654.3871459960938, 278.6759338378906 ], "mean_state_pred_error": 466.5315399169922 }, "credit_bridge": { "test_acc": 0.638, "mean_bp_cosine": 0.18947000801563263, "mean_rho": 0.3388983681797981, "mean_nudge_001": -0.004155236994847655, "mean_nudge_003": -0.012411902891471982, "mean_nudge_01": -0.04074052348732948, "bp_cosine_per_layer": [ 0.09752383828163147, 0.2814161777496338 ], "rho_per_layer": [ 0.2172888070344925, 0.46050792932510376 ], "nudge_per_layer": [ -0.05797416716814041, -0.023506879806518555 ], "final_value_loss": 0.6248516966819763, "final_term_loss": 0.2734043194293976, "final_bridge_loss": 0.04374015842676163, "final_tgrad_loss": 0.3077072183132172 } }, "a0.25_L2_s456": { "bp": { "test_acc": 0.914, "mean_bp_cosine": 0.8777631521224976, "mean_rho": 0.9999973177909851, "mean_nudge_001": -0.017619559541344643, "mean_nudge_003": -0.04847448691725731, "mean_nudge_01": -0.1200898252427578, "bp_cosine_per_layer": [ 0.8855119943618774, 0.8700143098831177 ], "rho_per_layer": [ 0.9999988079071045, 0.9999958276748657 ], "nudge_per_layer": [ -0.16111940145492554, -0.07906024903059006 ] }, "dfa": { "test_acc": 0.842, "mean_bp_cosine": 0.049672482535243034, "mean_rho": 0.03196176886558533, "mean_nudge_001": -0.0005511407216545194, "mean_nudge_003": -0.0016261094715446234, "mean_nudge_01": -0.0050985191483050585, "bp_cosine_per_layer": [ 0.015632983297109604, 0.08371198177337646 ], "rho_per_layer": [ -0.06864285469055176, 0.1325663924217224 ], "nudge_per_layer": [ -0.006782663054764271, -0.003414375241845846 ] }, "state_bridge": { "test_acc": 0.817, "mean_bp_cosine": 0.7210330367088318, "mean_rho": 0.6920621693134308, "mean_nudge_001": -0.01618554489687085, "mean_nudge_003": -0.04673728672787547, "mean_nudge_01": -0.13591178134083748, "bp_cosine_per_layer": [ 0.9367842674255371, 0.5052818059921265 ], "rho_per_layer": [ 0.846778392791748, 0.5373459458351135 ], "nudge_per_layer": [ -0.23458413779735565, -0.037239424884319305 ], "state_pred_error_per_layer": [ 626.5816650390625, 277.74237060546875 ], "mean_state_pred_error": 452.1620178222656 }, "credit_bridge": { "test_acc": 0.6685, "mean_bp_cosine": 0.21126148104667664, "mean_rho": 0.3547997772693634, "mean_nudge_001": -0.004930314142256975, "mean_nudge_003": -0.014735064469277859, "mean_nudge_01": -0.048451079055666924, "bp_cosine_per_layer": [ 0.13380715250968933, 0.28871580958366394 ], "rho_per_layer": [ 0.2647465467453003, 0.4448530077934265 ], "nudge_per_layer": [ -0.07276535779237747, -0.024136800318956375 ], "final_value_loss": 0.4186974630832672, "final_term_loss": 0.13192469900846482, "final_bridge_loss": 0.02328252377510071, "final_tgrad_loss": 0.2634902417182922 } }, "a0.25_L4_s42": { "bp": { "test_acc": 0.911, "mean_bp_cosine": 0.841001495718956, "mean_rho": 0.9999824017286301, "mean_nudge_001": -0.012893156963400543, "mean_nudge_003": -0.035831862362101674, "mean_nudge_01": -0.09177740290760994, "bp_cosine_per_layer": [ 0.868922770023346, 0.8429951667785645, 0.8289605379104614, 0.8231275081634521 ], "rho_per_layer": [ 0.999998927116394, 0.9999972581863403, 0.9999882578849792, 0.9999451637268066 ], "nudge_per_layer": [ -0.17674864828586578, -0.10683682560920715, -0.05415143817663193, -0.029372699558734894 ] }, "dfa": { "test_acc": 0.8515, "mean_bp_cosine": 0.0922011851798743, "mean_rho": 0.09930456057190895, "mean_nudge_001": -0.0006074781995266676, "mean_nudge_003": -0.001813438575481996, "mean_nudge_01": -0.005950940365437418, "bp_cosine_per_layer": [ 0.006489354185760021, 0.12851163744926453, 0.11991241574287415, 0.11389133334159851 ], "rho_per_layer": [ 0.029786743223667145, 0.18639126420021057, 0.11502038687467575, 0.06601984798908234 ], "nudge_per_layer": [ -0.01056693121790886, -0.006519604474306107, -0.003515135496854782, -0.003202090272679925 ] }, "state_bridge": { "test_acc": 0.738, "mean_bp_cosine": 0.3966497778892517, "mean_rho": 0.33668526262044907, "mean_nudge_001": -0.006818186957389116, "mean_nudge_003": -0.020020907453726977, "mean_nudge_01": -0.0618970247451216, "bp_cosine_per_layer": [ 0.8448630571365356, 0.4407959580421448, 0.15707260370254517, 0.14386749267578125 ], "rho_per_layer": [ 0.7033247947692871, 0.34020179510116577, 0.13395404815673828, 0.1692604124546051 ], "nudge_per_layer": [ -0.2203540802001953, -0.01797248050570488, -0.004883656278252602, -0.004377881996333599 ], "state_pred_error_per_layer": [ 3192.55322265625, 2613.6337890625, 1881.11669921875, 2102.51953125 ], "mean_state_pred_error": 2447.455810546875 }, "credit_bridge": { "test_acc": 0.662, "mean_bp_cosine": 0.1593644004315138, "mean_rho": 0.2909943610429764, "mean_nudge_001": -0.0019232924096286297, "mean_nudge_003": -0.005755707039497793, "mean_nudge_01": -0.019015976693481207, "bp_cosine_per_layer": [ 0.07093458622694016, 0.19623729586601257, 0.18378585577011108, 0.18649986386299133 ], "rho_per_layer": [ 0.14815130829811096, 0.336488276720047, 0.35506749153137207, 0.3242703676223755 ], "nudge_per_layer": [ -0.033366985619068146, -0.015239425003528595, -0.013842864893376827, -0.01361463125795126 ], "final_value_loss": 0.5039997359514237, "final_term_loss": 0.20441836426854135, "final_bridge_loss": 0.03893857387006283, "final_tgrad_loss": 0.26064279942512514 } }, "a0.25_L4_s123": { "bp": { "test_acc": 0.917, "mean_bp_cosine": 0.7935318648815155, "mean_rho": 0.9999762326478958, "mean_nudge_001": -0.01081152242841199, "mean_nudge_003": -0.029669071780517697, "mean_nudge_01": -0.0743116931989789, "bp_cosine_per_layer": [ 0.8279672861099243, 0.8114954233169556, 0.7863935232162476, 0.7482712268829346 ], "rho_per_layer": [ 0.9999987483024597, 0.9999956488609314, 0.999983012676239, 0.9999275207519531 ], "nudge_per_layer": [ -0.14138484001159668, -0.08725307881832123, -0.04476301744580269, -0.023845836520195007 ] }, "dfa": { "test_acc": 0.824, "mean_bp_cosine": 0.1441014357842505, "mean_rho": 0.14396781660616398, "mean_nudge_001": -0.0003178600236424245, "mean_nudge_003": -0.0009397821850143373, "mean_nudge_01": -0.0029927556752227247, "bp_cosine_per_layer": [ -0.030194921419024467, 0.3003618121147156, 0.18209302425384521, 0.12414582818746567 ], "rho_per_layer": [ -0.038011275231838226, 0.29912513494491577, 0.1735759675502777, 0.14118143916130066 ], "nudge_per_layer": [ 0.005076530389487743, -0.00841662660241127, -0.005011391825973988, -0.0036195346619933844 ] }, "state_bridge": { "test_acc": 0.733, "mean_bp_cosine": 0.3913041353225708, "mean_rho": 0.3523585610091686, "mean_nudge_001": -0.0061569800454890355, "mean_nudge_003": -0.01808436232386157, "mean_nudge_01": -0.055880151805467904, "bp_cosine_per_layer": [ 0.821143627166748, 0.36722686886787415, 0.20176959037780762, 0.1750764548778534 ], "rho_per_layer": [ 0.6905952095985413, 0.36183467507362366, 0.19890159368515015, 0.15810276567935944 ], "nudge_per_layer": [ -0.1973220407962799, -0.013631552457809448, -0.0068134344182908535, -0.0057535795494914055 ], "state_pred_error_per_layer": [ 3017.568359375, 2069.30517578125, 2103.28662109375, 1370.3270263671875 ], "mean_state_pred_error": 2140.121795654297 }, "credit_bridge": { "test_acc": 0.7085, "mean_bp_cosine": 0.11074696853756905, "mean_rho": 0.20390365412458777, "mean_nudge_001": -0.0010930995340459049, "mean_nudge_003": -0.0032710169325582683, "mean_nudge_01": -0.010791003005579114, "bp_cosine_per_layer": [ 0.021101564168930054, 0.1092284768819809, 0.14931762218475342, 0.16334021091461182 ], "rho_per_layer": [ 0.026919251307845116, 0.23688018321990967, 0.28421375155448914, 0.2676014304161072 ], "nudge_per_layer": [ -0.008425744250416756, -0.009595570154488087, -0.012630455195903778, -0.012512242421507835 ], "final_value_loss": 0.44552071504592894, "final_term_loss": 0.1639193995475769, "final_bridge_loss": 0.022096455523371698, "final_tgrad_loss": 0.2595048580646515 } }, "a0.25_L4_s456": { "bp": { "test_acc": 0.9185, "mean_bp_cosine": 0.7816510498523712, "mean_rho": 0.9999814331531525, "mean_nudge_001": -0.01180829742224887, "mean_nudge_003": -0.032709411811083555, "mean_nudge_01": -0.08326426614075899, "bp_cosine_per_layer": [ 0.8029586672782898, 0.7900986075401306, 0.7753629684448242, 0.7581839561462402 ], "rho_per_layer": [ 0.999998927116394, 0.9999964237213135, 0.999983549118042, 0.9999468326568604 ], "nudge_per_layer": [ -0.15638579428195953, -0.09745092689990997, -0.05102477967739105, -0.028195563703775406 ] }, "dfa": { "test_acc": 0.8515, "mean_bp_cosine": 0.06627230579033494, "mean_rho": 0.10299420775845647, "mean_nudge_001": -0.000254730271990411, "mean_nudge_003": -0.0007513021846534684, "mean_nudge_01": -0.0023495194036513567, "bp_cosine_per_layer": [ -0.014162426814436913, 0.0798005759716034, 0.09311279654502869, 0.10633827745914459 ], "rho_per_layer": [ 0.041398823261260986, 0.02247910387814045, 0.1714327037334442, 0.17666620016098022 ], "nudge_per_layer": [ 0.002492401283234358, -0.005343085154891014, -0.0033098761923611164, -0.003237517550587654 ] }, "state_bridge": { "test_acc": 0.7085, "mean_bp_cosine": 0.37615348026156425, "mean_rho": 0.3445914164185524, "mean_nudge_001": -0.00598156735941302, "mean_nudge_003": -0.017677294556051493, "mean_nudge_01": -0.055878938641399145, "bp_cosine_per_layer": [ 0.8312708735466003, 0.37615716457366943, 0.14436809718608856, 0.15281778573989868 ], "rho_per_layer": [ 0.7588610649108887, 0.34236449003219604, 0.14024245738983154, 0.13689765334129333 ], "nudge_per_layer": [ -0.19714727997779846, -0.01597677543759346, -0.004914070479571819, -0.005477628670632839 ], "state_pred_error_per_layer": [ 2233.37158203125, 1546.82080078125, 996.9107666015625, 942.33251953125 ], "mean_state_pred_error": 1429.8589172363281 }, "credit_bridge": { "test_acc": 0.6185, "mean_bp_cosine": 0.1528007835149765, "mean_rho": 0.2638061139732599, "mean_nudge_001": -0.0015675137401558459, "mean_nudge_003": -0.004692548653110862, "mean_nudge_01": -0.015524385962635279, "bp_cosine_per_layer": [ 0.048861026763916016, 0.17950567603111267, 0.17905107140541077, 0.20378535985946655 ], "rho_per_layer": [ 0.07314518839120865, 0.3309200406074524, 0.2789418697357178, 0.3722173571586609 ], "nudge_per_layer": [ -0.017692390829324722, -0.014619983732700348, -0.014469930902123451, -0.015315238386392593 ], "final_value_loss": 0.64283842253685, "final_term_loss": 0.27786159224510193, "final_bridge_loss": 0.049283357608318326, "final_tgrad_loss": 0.31569347660541536 } }, "a0.25_L8_s42": { "bp": { "test_acc": 0.8955, "mean_bp_cosine": 0.8411818519234657, "mean_rho": 0.999948225915432, "mean_nudge_001": -0.008358085062354803, "mean_nudge_003": -0.023856630141381174, "mean_nudge_01": -0.06703015509992838, "bp_cosine_per_layer": [ 0.8706932663917542, 0.8596300482749939, 0.8512309789657593, 0.842391848564148, 0.8334972858428955, 0.8275372385978699, 0.823634147644043, 0.8208400011062622 ], "rho_per_layer": [ 0.9999992847442627, 0.9999973773956299, 0.9999889731407166, 0.9999737739562988, 0.999952495098114, 0.9999229311943054, 0.9998778104782104, 0.999873161315918 ], "nudge_per_layer": [ -0.19971789419651031, -0.12236367166042328, -0.07083894312381744, -0.044348303228616714, -0.031799670308828354, -0.02545534446835518, -0.02191336825489998, -0.01980404555797577 ] }, "dfa": { "test_acc": 0.835, "mean_bp_cosine": 0.06675009592436254, "mean_rho": 0.0799849673639983, "mean_nudge_001": -6.642583684879355e-06, "mean_nudge_003": -9.95869777398184e-06, "mean_nudge_01": 0.00010420403850730509, "bp_cosine_per_layer": [ -0.029614916071295738, 0.08655504137277603, 0.09347832202911377, 0.10665939003229141, 0.10298001766204834, 0.055703092366456985, 0.06406168639659882, 0.054178133606910706 ], "rho_per_layer": [ -0.10576558113098145, 0.13653026521205902, 0.13744692504405975, 0.169493168592453, 0.1142939031124115, 0.09325758367776871, 0.1042378693819046, -0.00961439497768879 ], "nudge_per_layer": [ 0.021624702960252762, -0.006033358629792929, -0.003085250034928322, -0.0032903538085520267, -0.0030274176970124245, -0.0016871094703674316, -0.0020643905736505985, -0.0016031904378905892 ] }, "state_bridge": { "test_acc": 0.7705, "mean_bp_cosine": 0.2720195781439543, "mean_rho": 0.24409680720418692, "mean_nudge_001": -0.0037333294421841856, "mean_nudge_003": -0.01090175670105964, "mean_nudge_01": -0.03293398043024354, "bp_cosine_per_layer": [ 0.8497925996780396, 0.4692588150501251, 0.1727459728717804, 0.1467132717370987, 0.13944602012634277, 0.1262509524822235, 0.13373470306396484, 0.13821429014205933 ], "rho_per_layer": [ 0.7623768448829651, 0.4223487377166748, 0.18194836378097534, 0.12129966914653778, 0.11680327355861664, 0.15588274598121643, 0.1260674148797989, 0.06604740768671036 ], "nudge_per_layer": [ -0.23203837871551514, -0.013907128013670444, -0.0036564678885042667, -0.0029728016816079617, -0.0027978671714663506, -0.0026036568451672792, -0.002694307826459408, -0.0028012352995574474 ], "state_pred_error_per_layer": [ 10530.408203125, 8501.392578125, 5819.005859375, 4775.642578125, 4112.55224609375, 3853.83984375, 4342.64697265625, 3851.865478515625 ], "mean_state_pred_error": 5723.419219970703 }, "credit_bridge": { "test_acc": 0.616, "mean_bp_cosine": 0.11742978217080235, "mean_rho": 0.20278298296034336, "mean_nudge_001": -0.000763369575906836, "mean_nudge_003": -0.0022864671773277223, "mean_nudge_01": -0.007572929258458316, "bp_cosine_per_layer": [ 0.0006081201136112213, 0.13602128624916077, 0.1264307200908661, 0.13147516548633575, 0.14721693098545074, 0.1274956315755844, 0.13849349319934845, 0.1316969096660614 ], "rho_per_layer": [ 0.0121258944272995, 0.21904222667217255, 0.23992177844047546, 0.23384322226047516, 0.24397125840187073, 0.24230614304542542, 0.23810306191444397, 0.1929502785205841 ], "nudge_per_layer": [ 0.0015163691714406013, -0.009462166577577591, -0.008223006501793861, -0.008610539138317108, -0.009565019980072975, -0.00855704490095377, -0.008995135314762592, -0.008686890825629234 ], "final_value_loss": 0.5891867377281189, "final_term_loss": 0.3338063308954239, "final_bridge_loss": 0.026477533173561097, "final_tgrad_loss": 0.2289028725862503 } }, "a0.25_L8_s123": { "bp": { "test_acc": 0.896, "mean_bp_cosine": 0.8579939901828766, "mean_rho": 0.9999475553631783, "mean_nudge_001": -0.008832237246679142, "mean_nudge_003": -0.02530999999726191, "mean_nudge_01": -0.07218505791388452, "bp_cosine_per_layer": [ 0.8995364308357239, 0.8806806802749634, 0.8653636574745178, 0.852218508720398, 0.8451849222183228, 0.8416627645492554, 0.8401176333427429, 0.8391873240470886 ], "rho_per_layer": [ 0.9999992251396179, 0.9999973773956299, 0.9999878406524658, 0.9999729990959167, 0.9999438524246216, 0.9999128580093384, 0.9999063611030579, 0.9998599290847778 ], "nudge_per_layer": [ -0.2192329615354538, -0.13158643245697021, -0.07524827122688293, -0.04681030660867691, -0.03351406753063202, -0.026811176910996437, -0.02318240888416767, -0.02109483815729618 ] }, "dfa": { "test_acc": 0.833, "mean_bp_cosine": 0.07527017407119274, "mean_rho": 0.08970394870266318, "mean_nudge_001": -3.044791992579121e-05, "mean_nudge_003": -8.396412886213511e-05, "mean_nudge_01": -0.00019736881949938834, "bp_cosine_per_layer": [ -0.030230171978473663, 0.1515132635831833, 0.08784028142690659, 0.07231537997722626, 0.09554086625576019, 0.09082843363285065, 0.05412551760673523, 0.0802278220653534 ], "rho_per_layer": [ 0.04074572026729584, 0.1662997603416443, 0.08309999853372574, 0.07389070093631744, 0.14882703125476837, 0.04519672691822052, 0.0252661295235157, 0.13430552184581757 ], "nudge_per_layer": [ 0.013973418623209, -0.0047885943204164505, -0.0020089128520339727, -0.0015227446565404534, -0.002120924647897482, -0.0020366935059428215, -0.001262089004740119, -0.0018124101916328073 ] }, "state_bridge": { "test_acc": 0.753, "mean_bp_cosine": 0.26880230009555817, "mean_rho": 0.23677901271730661, "mean_nudge_001": -0.0032272498192469357, "mean_nudge_003": -0.009375294612254947, "mean_nudge_01": -0.027950568503001705, "bp_cosine_per_layer": [ 0.8559742569923401, 0.48324936628341675, 0.20313377678394318, 0.15002062916755676, 0.13207414746284485, 0.11473802477121353, 0.10703189671039581, 0.10419630259275436 ], "rho_per_layer": [ 0.712527871131897, 0.4307812452316284, 0.20181137323379517, 0.13881561160087585, 0.15075963735580444, 0.09288278967142105, 0.0890430212020874, 0.0776105523109436 ], "nudge_per_layer": [ -0.19413721561431885, -0.013532894663512707, -0.004060069564729929, -0.0029658200219273567, -0.00256463885307312, -0.0022291922941803932, -0.00204793200828135, -0.002066785003989935 ], "state_pred_error_per_layer": [ 8381.830078125, 6536.470703125, 4195.43359375, 3240.16162109375, 3085.78857421875, 3078.50390625, 3156.81787109375, 3267.9345703125 ], "mean_state_pred_error": 4367.867614746094 }, "credit_bridge": { "test_acc": 0.651, "mean_bp_cosine": 0.131322234403342, "mean_rho": 0.19494653795845807, "mean_nudge_001": -0.0009437191256438382, "mean_nudge_003": -0.002826296287821606, "mean_nudge_01": -0.009364197961986065, "bp_cosine_per_layer": [ 0.021486874669790268, 0.13597777485847473, 0.14020901918411255, 0.16125136613845825, 0.15609152615070343, 0.1541982889175415, 0.14447104930877686, 0.13689197599887848 ], "rho_per_layer": [ 0.021052071824669838, 0.17873916029930115, 0.25896143913269043, 0.18253415822982788, 0.20761561393737793, 0.24419330060482025, 0.21560567617416382, 0.25087088346481323 ], "nudge_per_layer": [ -0.006693178787827492, -0.009655885398387909, -0.009345939382910728, -0.010369300842285156, -0.010154420509934425, -0.010046296752989292, -0.009472687728703022, -0.009175874292850494 ], "final_value_loss": 0.41045001599788666, "final_term_loss": 0.17502211028933526, "final_bridge_loss": 0.01536440534889698, "final_tgrad_loss": 0.2200635038137436 } }, "a0.25_L8_s456": { "bp": { "test_acc": 0.8915, "mean_bp_cosine": 0.8808343783020973, "mean_rho": 0.9999449253082275, "mean_nudge_001": -0.00886416298453696, "mean_nudge_003": -0.025221005314961076, "mean_nudge_01": -0.07025118661113083, "bp_cosine_per_layer": [ 0.9125592708587646, 0.9000785946846008, 0.8909372687339783, 0.8829550743103027, 0.8725745677947998, 0.8661020994186401, 0.8620940446853638, 0.8593741059303284 ], "rho_per_layer": [ 0.999998927116394, 0.9999973177909851, 0.9999907612800598, 0.9999756813049316, 0.9999436140060425, 0.9999102354049683, 0.9998818635940552, 0.9998610019683838 ], "nudge_per_layer": [ -0.20531992614269257, -0.12906500697135925, -0.07575342804193497, -0.04753156006336212, -0.033830925822257996, -0.02684028074145317, -0.02297976426780224, -0.020688600838184357 ] }, "dfa": { "test_acc": 0.8165, "mean_bp_cosine": 0.1053345538675785, "mean_rho": 0.1329206523951143, "mean_nudge_001": -0.0005203891905694036, "mean_nudge_003": -0.0015467511984752491, "mean_nudge_01": -0.0049866703047882766, "bp_cosine_per_layer": [ 0.023487910628318787, 0.16097302734851837, 0.15321409702301025, 0.08171598613262177, 0.09132305532693863, 0.0943714827299118, 0.12134475260972977, 0.11624611914157867 ], "rho_per_layer": [ 0.006953349336981773, 0.2090909481048584, 0.23601846396923065, 0.09328415989875793, 0.10916170477867126, 0.12872634828090668, 0.15839380025863647, 0.12173644453287125 ], "nudge_per_layer": [ -0.013351024128496647, -0.0065501369535923, -0.004914519842714071, -0.002402292797341943, -0.002674146555364132, -0.0028969072736799717, -0.0036479653790593147, -0.0034563695080578327 ] }, "state_bridge": { "test_acc": 0.765, "mean_bp_cosine": 0.24095587618649006, "mean_rho": 0.20889087952673435, "mean_nudge_001": -0.0035492044535203604, "mean_nudge_003": -0.01036601059968234, "mean_nudge_01": -0.03139855920744594, "bp_cosine_per_layer": [ 0.8715300559997559, 0.44268321990966797, 0.14538408815860748, 0.1153520941734314, 0.09680549800395966, 0.08768410980701447, 0.08654746413230896, 0.08166047930717468 ], "rho_per_layer": [ 0.7710200548171997, 0.4253605008125305, 0.09275925159454346, 0.09712537378072739, 0.11123719066381454, 0.028249293565750122, 0.05002531409263611, 0.09535005688667297 ], "nudge_per_layer": [ -0.22555799782276154, -0.012845459394156933, -0.003021553158760071, -0.0024172987323254347, -0.002025863155722618, -0.0018953735707327724, -0.0018195084994658828, -0.0016054193256422877 ], "state_pred_error_per_layer": [ 7857.92041015625, 7327.2666015625, 4158.3056640625, 3231.240234375, 3079.712890625, 2843.65966796875, 3014.824951171875, 2904.82080078125 ], "mean_state_pred_error": 4302.218902587891 }, "credit_bridge": { "test_acc": 0.6845, "mean_bp_cosine": 0.08619831164833158, "mean_rho": 0.20396122278179973, "mean_nudge_001": -0.0008905812101147603, "mean_nudge_003": -0.0026642234588507563, "mean_nudge_01": -0.008798376395134255, "bp_cosine_per_layer": [ -0.005484581924974918, 0.03739434853196144, 0.0736197978258133, 0.10366753488779068, 0.11266068369150162, 0.12653732299804688, 0.11692580580711365, 0.12426558136940002 ], "rho_per_layer": [ -0.0015162499621510506, 0.12250564247369766, 0.22542724013328552, 0.204342320561409, 0.2217923402786255, 0.28867796063423157, 0.33211782574653625, 0.23834270238876343 ], "nudge_per_layer": [ 0.0002549269702285528, -0.006913396995514631, -0.008808376267552376, -0.010325662791728973, -0.010591026395559311, -0.011654841713607311, -0.010967787355184555, -0.011380846612155437 ], "final_value_loss": 0.966836922454834, "final_term_loss": 0.6684003219604492, "final_bridge_loss": 0.09051647346019745, "final_tgrad_loss": 0.20792012577056884 } }, "a0.25_L12_s42": { "bp": { "test_acc": 0.897, "mean_bp_cosine": 0.826698362827301, "mean_rho": 0.9998699675003687, "mean_nudge_001": -0.006519351474707946, "mean_nudge_003": -0.0185935211678346, "mean_nudge_01": -0.05259064872128268, "bp_cosine_per_layer": [ 0.8824102282524109, 0.865839958190918, 0.8526839017868042, 0.839972198009491, 0.826134204864502, 0.8190910220146179, 0.8134430646896362, 0.8095068335533142, 0.8061133027076721, 0.8034266233444214, 0.8016149997711182, 0.8001440167427063 ], "rho_per_layer": [ 0.9999990463256836, 0.9999975562095642, 0.9999887347221375, 0.9999736547470093, 0.9999515414237976, 0.9999052882194519, 0.9998592138290405, 0.999845027923584, 0.99979567527771, 0.9997392892837524, 0.999713659286499, 0.9996709227561951 ], "nudge_per_layer": [ -0.21559840440750122, -0.13125374913215637, -0.0758625715970993, -0.04728776216506958, -0.03328181058168411, -0.025909576565027237, -0.021545151248574257, -0.01875336468219757, -0.016917269676923752, -0.015656160190701485, -0.014799138531088829, -0.01422282587736845 ] }, "dfa": { "test_acc": 0.8165, "mean_bp_cosine": 0.06692749153201778, "mean_rho": 0.09511666395701468, "mean_nudge_001": -0.0003660954941248444, "mean_nudge_003": -0.0010870555027698476, "mean_nudge_01": -0.003503491034886489, "bp_cosine_per_layer": [ 0.02035588026046753, 0.09721152484416962, 0.10017214715480804, 0.06879279017448425, 0.0898706465959549, 0.0775061622262001, 0.0738426148891449, 0.022470442578196526, 0.03698011115193367, 0.09759337455034256, 0.05344103276729584, 0.06489317119121552 ], "rho_per_layer": [ -0.041805852204561234, 0.14336195588111877, 0.21263040602207184, 0.15326815843582153, 0.04533267021179199, 0.07981619238853455, 0.08665720373392105, 0.004602088592946529, 0.0780038833618164, 0.11769488453865051, 0.1415681391954422, 0.12027023732662201 ], "nudge_per_layer": [ -0.007749770302325487, -0.008321389555931091, -0.003876405069604516, -0.0026999912224709988, -0.0026630363427102566, -0.0033310302533209324, -0.0026011697482317686, -0.0011158722918480635, -0.0018408913165330887, -0.0036767907440662384, -0.0019704087171703577, -0.0021951368544250727 ] }, "state_bridge": { "test_acc": 0.687, "mean_bp_cosine": 0.1969081535935402, "mean_rho": 0.18958334562679133, "mean_nudge_001": -0.0026059118196523436, "mean_nudge_003": -0.007709211970601852, "mean_nudge_01": -0.02444164449116215, "bp_cosine_per_layer": [ 0.7618155479431152, 0.3189862072467804, 0.20356883108615875, 0.1471748650074005, 0.14541830122470856, 0.13144558668136597, 0.11054451763629913, 0.12045299261808395, 0.11467338353395462, 0.1017053946852684, 0.10614258050918579, 0.10096963495016098 ], "rho_per_layer": [ 0.6406711339950562, 0.31470298767089844, 0.1861359179019928, 0.17725443840026855, 0.11286133527755737, 0.1387007087469101, 0.07976903766393661, 0.11419158428907394, 0.13525697588920593, 0.13865813612937927, 0.10792528837919235, 0.1288726031780243 ], "nudge_per_layer": [ -0.24276113510131836, -0.011681582778692245, -0.006372842937707901, -0.004580066539347172, -0.004361784551292658, -0.004005158320069313, -0.0033329487778246403, -0.003609130624681711, -0.003437052248045802, -0.003070915350690484, -0.0031537071336060762, -0.0029334095306694508 ], "state_pred_error_per_layer": [ 13069.6103515625, 12902.267578125, 5325.7177734375, 4598.6796875, 4871.38818359375, 4525.17333984375, 4230.712890625, 4117.212890625, 3694.05078125, 3737.547607421875, 3966.26025390625, 4157.70654296875 ], "mean_state_pred_error": 5766.360656738281 }, "credit_bridge": { "test_acc": 0.6145, "mean_bp_cosine": 0.07906571677691925, "mean_rho": 0.138323282978187, "mean_nudge_001": -0.000623912480174719, "mean_nudge_003": -0.0018697016639634967, "mean_nudge_01": -0.006201692313576738, "bp_cosine_per_layer": [ -0.00034100248012691736, 0.04912012815475464, 0.07342620939016342, 0.08266763389110565, 0.08488726615905762, 0.08897420018911362, 0.09187051653862, 0.09561659395694733, 0.09157264232635498, 0.0947732925415039, 0.09517554938793182, 0.10104557126760483 ], "rho_per_layer": [ 0.03049650974571705, 0.10809889435768127, 0.08979646861553192, 0.17959633469581604, 0.1275440901517868, 0.17255432903766632, 0.16415373980998993, 0.11534757167100906, 0.17073199152946472, 0.15983673930168152, 0.15877583622932434, 0.18294689059257507 ], "nudge_per_layer": [ -0.003660556860268116, -0.004647519439458847, -0.005617665126919746, -0.006360755302011967, -0.006202561780810356, -0.006613034754991531, -0.006794797722250223, -0.006956290453672409, -0.006790010258555412, -0.006838815752416849, -0.006875438615679741, -0.007062861695885658 ], "final_value_loss": 0.5342490439891815, "final_term_loss": 0.2775654150247574, "final_bridge_loss": 0.01819542294666171, "final_tgrad_loss": 0.2384882021665573 } }, "a0.25_L12_s123": { "bp": { "test_acc": 0.902, "mean_bp_cosine": 0.7854474087556204, "mean_rho": 0.9998569438854853, "mean_nudge_001": -0.005986864319614445, "mean_nudge_003": -0.01699222985189408, "mean_nudge_01": -0.04738136318822702, "bp_cosine_per_layer": [ 0.8318638801574707, 0.8205357789993286, 0.8115277886390686, 0.8032979965209961, 0.7886118292808533, 0.7801527976989746, 0.7755030393600464, 0.769822359085083, 0.7656118869781494, 0.7624454498291016, 0.7592873573303223, 0.7567087411880493 ], "rho_per_layer": [ 0.9999992251396179, 0.9999969005584717, 0.9999886155128479, 0.9999699592590332, 0.9999390244483948, 0.9998932480812073, 0.9998554587364197, 0.9998190999031067, 0.9997619986534119, 0.9997045397758484, 0.9997037649154663, 0.9996514916419983 ], "nudge_per_layer": [ -0.19001588225364685, -0.1191760003566742, -0.0696173831820488, -0.04359997808933258, -0.030619269236922264, -0.023702502250671387, -0.019644131883978844, -0.01702706888318062, -0.015273808501660824, -0.014067228883504868, -0.01321301981806755, -0.012620084919035435 ] }, "dfa": { "test_acc": 0.8475, "mean_bp_cosine": 0.08017303561791778, "mean_rho": 0.06949969536314408, "mean_nudge_001": -0.00017117233619501349, "mean_nudge_003": -0.0005107432225486264, "mean_nudge_01": -0.0016712878035226215, "bp_cosine_per_layer": [ 0.007017737254500389, 0.1750856339931488, 0.09811605513095856, 0.09870222210884094, 0.06923168152570724, 0.07072796672582626, 0.1299426257610321, 0.06372487545013428, 0.05202930420637131, 0.07090489566326141, 0.04892899468541145, 0.07766443490982056 ], "rho_per_layer": [ -0.05082311853766441, 0.15127798914909363, 0.10723739117383957, 0.1141807958483696, 0.10590988397598267, 0.05159565061330795, 0.11685939878225327, 0.06997300684452057, 0.00848393514752388, 0.09840574860572815, 0.04575839638710022, 0.015137266367673874 ], "nudge_per_layer": [ -9.43206250667572e-05, -0.004689650610089302, -0.0020264536142349243, -0.0020407852716743946, -0.0012656663311645389, -0.0012923413887619972, -0.0026080894749611616, -0.0013762509915977716, -0.0008721412159502506, -0.0014046088326722383, -0.0009981195908039808, -0.0013870256952941418 ] }, "state_bridge": { "test_acc": 0.653, "mean_bp_cosine": 0.212141719336311, "mean_rho": 0.16903599786261717, "mean_nudge_001": -0.002121228181446592, "mean_nudge_003": -0.006256894756612989, "mean_nudge_01": -0.019624086019272607, "bp_cosine_per_layer": [ 0.7411977648735046, 0.3599514067173004, 0.24466437101364136, 0.18324854969978333, 0.1438049077987671, 0.1288401186466217, 0.12390943616628647, 0.11832019686698914, 0.1285572201013565, 0.12467152625322342, 0.12941977381706238, 0.11911536008119583 ], "rho_per_layer": [ 0.637965977191925, 0.3033628761768341, 0.22177954018115997, 0.1946793794631958, 0.08981841057538986, 0.07293770462274551, 0.06646254658699036, 0.08072252571582794, 0.1001453697681427, 0.10873826593160629, 0.08548736572265625, 0.06633201241493225 ], "nudge_per_layer": [ -0.19135718047618866, -0.010539980605244637, -0.006164146587252617, -0.004623022396117449, -0.003351652529090643, -0.002855603815987706, -0.002745934296399355, -0.0025949627161026, -0.002952038776129484, -0.002784580923616886, -0.002810355508700013, -0.0027095736004412174 ], "state_pred_error_per_layer": [ 11398.169921875, 8994.236328125, 6843.07958984375, 5338.4423828125, 5392.61279296875, 5254.96435546875, 5057.08544921875, 3968.3759765625, 3609.97021484375, 3456.01708984375, 3801.56005859375, 3800.938232421875 ], "mean_state_pred_error": 5576.287699381511 }, "credit_bridge": { "test_acc": 0.6165, "mean_bp_cosine": 0.062155287557592, "mean_rho": 0.1365320294474562, "mean_nudge_001": -0.0005568450481708472, "mean_nudge_003": -0.001666992757236585, "mean_nudge_01": -0.005523480668974419, "bp_cosine_per_layer": [ 0.022855455055832863, 0.05216163024306297, 0.06243491545319557, 0.06105152890086174, 0.04914259910583496, 0.0686616599559784, 0.07184358686208725, 0.06793813407421112, 0.07173632830381393, 0.07284142076969147, 0.07256414741277695, 0.07263204455375671 ], "rho_per_layer": [ 0.035705771297216415, 0.06484436988830566, 0.15210987627506256, 0.16708198189735413, 0.14389753341674805, 0.17870034277439117, 0.1870872974395752, 0.1476944386959076, 0.15697413682937622, 0.11093902587890625, 0.14925891160964966, 0.1440906673669815 ], "nudge_per_layer": [ -0.010396174155175686, -0.0040237451903522015, -0.004916047677397728, -0.004709047265350819, -0.003890307154506445, -0.005256335251033306, -0.00541608827188611, -0.005244302563369274, -0.005511538591235876, -0.005561722908169031, -0.005636853165924549, -0.0057196058332920074 ], "final_value_loss": 0.7472030250549316, "final_term_loss": 0.48659993648529054, "final_bridge_loss": 0.02176231173723936, "final_tgrad_loss": 0.23884078254699706 } }, "a0.25_L12_s456": { "bp": { "test_acc": 0.899, "mean_bp_cosine": 0.7671962430079778, "mean_rho": 0.9998593876759211, "mean_nudge_001": -0.004748132584306101, "mean_nudge_003": -0.013334368025728812, "mean_nudge_01": -0.03616663611804446, "bp_cosine_per_layer": [ 0.8408721685409546, 0.8110387921333313, 0.7906323671340942, 0.7719681262969971, 0.7599710822105408, 0.7539124488830566, 0.7502779364585876, 0.7483201026916504, 0.7469743490219116, 0.7455896139144897, 0.7440224885940552, 0.7427754402160645 ], "rho_per_layer": [ 0.9999985694885254, 0.9999968409538269, 0.9999892711639404, 0.9999746680259705, 0.9999452233314514, 0.9998985528945923, 0.999815821647644, 0.9998071789741516, 0.9997591972351074, 0.9997697472572327, 0.9996993541717529, 0.9996582269668579 ], "nudge_per_layer": [ -0.1375543624162674, -0.0896252989768982, -0.05422336980700493, -0.034489020705223083, -0.024439619854092598, -0.019029833376407623, -0.015829749405384064, -0.013777680695056915, -0.012396320700645447, -0.011455517262220383, -0.01080663874745369, -0.01037222146987915 ] }, "dfa": { "test_acc": 0.8205, "mean_bp_cosine": 0.11264437561233838, "mean_rho": 0.10325007668385904, "mean_nudge_001": -0.0003942390236867747, "mean_nudge_003": -0.001173070602817461, "mean_nudge_01": -0.003865133059055855, "bp_cosine_per_layer": [ 0.03703948110342026, 0.2760871648788452, 0.18428769707679749, 0.1331849992275238, 0.09719057381153107, 0.1258918195962906, 0.0748991146683693, 0.08436713367700577, 0.09110807627439499, 0.10076140612363815, 0.06335102021694183, 0.08356402069330215 ], "rho_per_layer": [ 0.048133689910173416, 0.19754433631896973, 0.16992174088954926, 0.13566580414772034, 0.05106702446937561, 0.13118207454681396, 0.057789772748947144, 0.09156530350446701, 0.05846566706895828, 0.07805749773979187, 0.11393842101097107, 0.10566958785057068 ], "nudge_per_layer": [ -0.010506195947527885, -0.007814407348632812, -0.005055932328104973, -0.003603969234973192, -0.0024205115623772144, -0.0030835624784231186, -0.0020989018958061934, -0.002160410862416029, -0.0025065483059734106, -0.0029421483632177114, -0.0018882867880165577, -0.0023007215932011604 ] }, "state_bridge": { "test_acc": 0.645, "mean_bp_cosine": 0.17685540641347566, "mean_rho": 0.14925638431062302, "mean_nudge_001": -0.001906786697266701, "mean_nudge_003": -0.005651665274247837, "mean_nudge_01": -0.018038961667722713, "bp_cosine_per_layer": [ 0.7252062559127808, 0.29899632930755615, 0.17116883397102356, 0.13035137951374054, 0.11515486240386963, 0.09652335941791534, 0.0831163302063942, 0.08587749302387238, 0.1138496845960617, 0.09370164573192596, 0.10970423370599747, 0.09861446917057037 ], "rho_per_layer": [ 0.5776556730270386, 0.24244043231010437, 0.15723995864391327, 0.11212021112442017, 0.13527166843414307, 0.09345141798257828, 0.05218992009758949, 0.09958191215991974, 0.09601333737373352, 0.05242425203323364, 0.08741243183612823, 0.08527539670467377 ], "nudge_per_layer": [ -0.1766481101512909, -0.00916180107742548, -0.004768161568790674, -0.00345130218192935, -0.0031899111345410347, -0.002747050253674388, -0.002392321825027466, -0.0024352066684514284, -0.0032566720619797707, -0.0024684183299541473, -0.003083357820287347, -0.0028652269393205643 ], "state_pred_error_per_layer": [ 14567.1396484375, 8843.26171875, 5969.56787109375, 5045.4873046875, 4699.3388671875, 4805.94287109375, 4535.58935546875, 4793.4423828125, 4549.9345703125, 4690.251953125, 4812.1875, 4947.22509765625 ], "mean_state_pred_error": 6021.614095052083 }, "credit_bridge": { "test_acc": 0.5665, "mean_bp_cosine": 0.0795192973067363, "mean_rho": 0.11809263161073129, "mean_nudge_001": -0.0007323032235338663, "mean_nudge_003": -0.002197402771950389, "mean_nudge_01": -0.007294998116170366, "bp_cosine_per_layer": [ 0.0279831662774086, 0.034206733107566833, 0.06341560930013657, 0.0695083886384964, 0.09047738462686539, 0.09644618630409241, 0.1011449545621872, 0.09714356064796448, 0.09739723801612854, 0.09588810801506042, 0.09328337758779526, 0.08733686059713364 ], "rho_per_layer": [ -0.01078212633728981, 0.08389643579721451, 0.09591226279735565, 0.09822964668273926, 0.15539337694644928, 0.077084481716156, 0.18135306239128113, 0.13923753798007965, 0.15523496270179749, 0.1566530466079712, 0.16011789441108704, 0.12478099763393402 ], "nudge_per_layer": [ -0.014675050042569637, -0.0035726604983210564, -0.005193591117858887, -0.005878218449652195, -0.007073326967656612, -0.007276617921888828, -0.0072667524218559265, -0.007385652512311935, -0.00723686721175909, -0.007413622923195362, -0.0073149921372532845, -0.007252625189721584 ], "final_value_loss": 0.8463798692464829, "final_term_loss": 0.5014169258773327, "final_bridge_loss": 0.03549532132707536, "final_tgrad_loss": 0.3094676312446594 } } }