{ "a0.5_L2_s42": { "bp": { "test_acc": 0.8845, "mean_bp_cosine": 0.9139105975627899, "mean_rho": 0.9999974966049194, "mean_nudge_001": -0.021973546594381332, "mean_nudge_003": -0.061596330255270004, "mean_nudge_01": -0.16279328987002373, "bp_cosine_per_layer": [ 0.9165740013122559, 0.911247193813324 ], "rho_per_layer": [ 0.9999988079071045, 0.9999961853027344 ], "nudge_per_layer": [ -0.21485117077827454, -0.11073540896177292 ] }, "dfa": { "test_acc": 0.8185, "mean_bp_cosine": 0.013388294260948896, "mean_rho": -0.010654313489794731, "mean_nudge_001": 0.00010355231643188745, "mean_nudge_003": 0.00034307173336856067, "mean_nudge_01": 0.0015135867288336158, "bp_cosine_per_layer": [ -0.008949288167059422, 0.035725876688957214 ], "rho_per_layer": [ -0.042064815759658813, 0.02075618878006935 ], "nudge_per_layer": [ 0.004335631616413593, -0.0013084581587463617 ] }, "state_bridge": { "test_acc": 0.8125, "mean_bp_cosine": 0.6582441031932831, "mean_rho": 0.6169608980417252, "mean_nudge_001": -0.013940556207671762, "mean_nudge_003": -0.040253227576613426, "mean_nudge_01": -0.11712449323385954, "bp_cosine_per_layer": [ 0.8927175402641296, 0.4237706661224365 ], "rho_per_layer": [ 0.8139950037002563, 0.41992679238319397 ], "nudge_per_layer": [ -0.20875152945518494, -0.02549745701253414 ], "state_pred_error_per_layer": [ 1690.90478515625, 278.7707214355469 ], "mean_state_pred_error": 984.8377532958984 }, "credit_bridge": { "test_acc": 0.6365, "mean_bp_cosine": 0.23397326469421387, "mean_rho": 0.31064942479133606, "mean_nudge_001": -0.004609669325873256, "mean_nudge_003": -0.013776597566902637, "mean_nudge_01": -0.04530538059771061, "bp_cosine_per_layer": [ 0.12511783838272095, 0.3428286910057068 ], "rho_per_layer": [ 0.21235734224319458, 0.40894150733947754 ], "nudge_per_layer": [ -0.061930932104587555, -0.028679829090833664 ], "final_value_loss": 0.48061015357971193, "final_term_loss": 0.14492559936642646, "final_bridge_loss": 0.020528473022580148, "final_tgrad_loss": 0.31515607919692995 } }, "a0.5_L2_s123": { "bp": { "test_acc": 0.895, "mean_bp_cosine": 0.878325492143631, "mean_rho": 0.9999979734420776, "mean_nudge_001": -0.019721894524991512, "mean_nudge_003": -0.055400675162672997, "mean_nudge_01": -0.14613648504018784, "bp_cosine_per_layer": [ 0.8870643973350525, 0.8695865869522095 ], "rho_per_layer": [ 0.9999991059303284, 0.9999968409538269 ], "nudge_per_layer": [ -0.1920192688703537, -0.10025370121002197 ] }, "dfa": { "test_acc": 0.8405, "mean_bp_cosine": 0.08467924036085606, "mean_rho": 0.07321365177631378, "mean_nudge_001": -0.0010137155040865764, "mean_nudge_003": -0.0030202887719497085, "mean_nudge_01": -0.009814865421503782, "bp_cosine_per_layer": [ 0.05535196140408516, 0.11400651931762695 ], "rho_per_layer": [ 0.07898944616317749, 0.06743785738945007 ], "nudge_per_layer": [ -0.015302825719118118, -0.004326905123889446 ] }, "state_bridge": { "test_acc": 0.8305, "mean_bp_cosine": 0.659128725528717, "mean_rho": 0.5701804161071777, "mean_nudge_001": -0.012253106338903308, "mean_nudge_003": -0.035399872343987226, "mean_nudge_01": -0.10352936759591103, "bp_cosine_per_layer": [ 0.8879689574241638, 0.43028849363327026 ], "rho_per_layer": [ 0.7560213804244995, 0.38433945178985596 ], "nudge_per_layer": [ -0.18288554251194, -0.02417319267988205 ], "state_pred_error_per_layer": [ 1696.6031494140625, 278.31060791015625 ], "mean_state_pred_error": 987.4568786621094 }, "credit_bridge": { "test_acc": 0.634, "mean_bp_cosine": 0.21542657166719437, "mean_rho": 0.3815983682870865, "mean_nudge_001": -0.004454661626368761, "mean_nudge_003": -0.0132976695895195, "mean_nudge_01": -0.04354145657271147, "bp_cosine_per_layer": [ 0.11650891602039337, 0.31434422731399536 ], "rho_per_layer": [ 0.23759987950325012, 0.5255968570709229 ], "nudge_per_layer": [ -0.05865050479769707, -0.028432408347725868 ], "final_value_loss": 0.5198581983566284, "final_term_loss": 0.1660363396883011, "final_bridge_loss": 0.026915473002195357, "final_tgrad_loss": 0.32690638537406924 } }, "a0.5_L2_s456": { "bp": { "test_acc": 0.875, "mean_bp_cosine": 0.8901284635066986, "mean_rho": 0.9999979138374329, "mean_nudge_001": -0.024593621492385864, "mean_nudge_003": -0.06968314573168755, "mean_nudge_01": -0.18856997787952423, "bp_cosine_per_layer": [ 0.8992727398872375, 0.8809841871261597 ], "rho_per_layer": [ 0.999998927116394, 0.9999969005584717 ], "nudge_per_layer": [ -0.2502431571483612, -0.12689679861068726 ] }, "dfa": { "test_acc": 0.816, "mean_bp_cosine": 0.04272198863327503, "mean_rho": 0.0372689850628376, "mean_nudge_001": -0.0004823530907742679, "mean_nudge_003": -0.0014278392773121595, "mean_nudge_01": -0.004543605959042907, "bp_cosine_per_layer": [ 0.017043430358171463, 0.0684005469083786 ], "rho_per_layer": [ -0.039058439433574677, 0.11359640955924988 ], "nudge_per_layer": [ -0.005804130807518959, -0.0032830811105668545 ] }, "state_bridge": { "test_acc": 0.807, "mean_bp_cosine": 0.6757712662220001, "mean_rho": 0.6000171899795532, "mean_nudge_001": -0.01571581745520234, "mean_nudge_003": -0.04562427382916212, "mean_nudge_01": -0.13532716780900955, "bp_cosine_per_layer": [ 0.8896123766899109, 0.46193015575408936 ], "rho_per_layer": [ 0.8089620471000671, 0.3910723328590393 ], "nudge_per_layer": [ -0.23563425242900848, -0.03502008318901062 ], "state_pred_error_per_layer": [ 1470.6129150390625, 266.84222412109375 ], "mean_state_pred_error": 868.7275695800781 }, "credit_bridge": { "test_acc": 0.6605, "mean_bp_cosine": 0.23513908684253693, "mean_rho": 0.3664921522140503, "mean_nudge_001": -0.005338509101420641, "mean_nudge_003": -0.015948554500937462, "mean_nudge_01": -0.052367652766406536, "bp_cosine_per_layer": [ 0.15452352166175842, 0.31575465202331543 ], "rho_per_layer": [ 0.26963475346565247, 0.4633495509624481 ], "nudge_per_layer": [ -0.07802344858646393, -0.026711856946349144 ], "final_value_loss": 0.3851433990240097, "final_term_loss": 0.10070284021273256, "final_bridge_loss": 0.01744421122968197, "final_tgrad_loss": 0.26699634662866595 } }, "a0.5_L4_s42": { "bp": { "test_acc": 0.883, "mean_bp_cosine": 0.8371244817972183, "mean_rho": 0.9999875724315643, "mean_nudge_001": -0.0205106531502679, "mean_nudge_003": -0.05829396191984415, "mean_nudge_01": -0.15927099250257015, "bp_cosine_per_layer": [ 0.8572149276733398, 0.8503379821777344, 0.8309497833251953, 0.8099952340126038 ], "rho_per_layer": [ 0.9999990463256836, 0.9999980330467224, 0.999992311000824, 0.9999608993530273 ], "nudge_per_layer": [ -0.29344552755355835, -0.19200363755226135, -0.10158533602952957, -0.050049468874931335 ] }, "dfa": { "test_acc": 0.825, "mean_bp_cosine": 0.08056101959664375, "mean_rho": 0.07036433229222894, "mean_nudge_001": -0.0004525305703282356, "mean_nudge_003": -0.0013484152732416987, "mean_nudge_01": -0.004391533322632313, "bp_cosine_per_layer": [ 0.0060789515264332294, 0.09873200953006744, 0.10723873972892761, 0.1101943776011467 ], "rho_per_layer": [ -0.012974856421351433, 0.14552229642868042, 0.08513970673084259, 0.06377018243074417 ], "nudge_per_layer": [ -0.0025177751667797565, -0.006775799673050642, -0.004219442140311003, -0.00405311631038785 ] }, "state_bridge": { "test_acc": 0.7175, "mean_bp_cosine": 0.389427974820137, "mean_rho": 0.3547045961022377, "mean_nudge_001": -0.006397623001248576, "mean_nudge_003": -0.018759612808935344, "mean_nudge_01": -0.05771041696425527, "bp_cosine_per_layer": [ 0.81822270154953, 0.44441917538642883, 0.15419423580169678, 0.14087578654289246 ], "rho_per_layer": [ 0.7556701302528381, 0.42722654342651367, 0.13306331634521484, 0.10285839438438416 ], "nudge_per_layer": [ -0.2031538337469101, -0.01867133006453514, -0.004740326199680567, -0.00427617784589529 ], "state_pred_error_per_layer": [ 6433.45703125, 2426.59619140625, 1950.234130859375, 1929.63232421875 ], "mean_state_pred_error": 3184.9799194335938 }, "credit_bridge": { "test_acc": 0.662, "mean_bp_cosine": 0.2142251469194889, "mean_rho": 0.3778097555041313, "mean_nudge_001": -0.002453161054290831, "mean_nudge_003": -0.007344013080000877, "mean_nudge_01": -0.024294552858918905, "bp_cosine_per_layer": [ 0.08037468791007996, 0.24492917954921722, 0.2708122730255127, 0.26078444719314575 ], "rho_per_layer": [ 0.14874565601348877, 0.4311285614967346, 0.48388832807540894, 0.44747647643089294 ], "nudge_per_layer": [ -0.03933826461434364, -0.020281529054045677, -0.01922188699245453, -0.018336530774831772 ], "final_value_loss": 0.4397273305892944, "final_term_loss": 0.15641304728984834, "final_bridge_loss": 0.03557699205875397, "final_tgrad_loss": 0.24773729231357575 } }, "a0.5_L4_s123": { "bp": { "test_acc": 0.882, "mean_bp_cosine": 0.7865977436304092, "mean_rho": 0.9999855011701584, "mean_nudge_001": -0.019216455286368728, "mean_nudge_003": -0.05414565000683069, "mean_nudge_01": -0.14401183556765318, "bp_cosine_per_layer": [ 0.817541241645813, 0.7992762923240662, 0.7803024053573608, 0.749271035194397 ], "rho_per_layer": [ 0.9999992847442627, 0.9999971389770508, 0.9999915361404419, 0.9999540448188782 ], "nudge_per_layer": [ -0.25995129346847534, -0.17589232325553894, -0.09449617564678192, -0.04570754989981651 ] }, "dfa": { "test_acc": 0.8045, "mean_bp_cosine": 0.13894247123971581, "mean_rho": 0.15608793403953314, "mean_nudge_001": -0.0003833890805253759, "mean_nudge_003": -0.0011389600113034248, "mean_nudge_01": -0.0036781042581424117, "bp_cosine_per_layer": [ -0.027901431545615196, 0.2837410271167755, 0.17885790765285492, 0.12107238173484802 ], "rho_per_layer": [ -0.007988158613443375, 0.25683730840682983, 0.21096175909042358, 0.1645408272743225 ], "nudge_per_layer": [ 0.005267709027975798, -0.009674372151494026, -0.005893784575164318, -0.0044119693338871 ] }, "state_bridge": { "test_acc": 0.7345, "mean_bp_cosine": 0.3779419884085655, "mean_rho": 0.3288300409913063, "mean_nudge_001": -0.006349694755044766, "mean_nudge_003": -0.018683148198761046, "mean_nudge_01": -0.05818818649277091, "bp_cosine_per_layer": [ 0.8108769655227661, 0.3922353684902191, 0.15907782316207886, 0.149577796459198 ], "rho_per_layer": [ 0.6536975502967834, 0.33696871995925903, 0.15431806445121765, 0.1703358292579651 ], "nudge_per_layer": [ -0.2048409879207611, -0.01618226245045662, -0.006159749813377857, -0.005569745786488056 ], "state_pred_error_per_layer": [ 3474.9130859375, 1348.3646240234375, 1462.0654296875, 924.672607421875 ], "mean_state_pred_error": 1802.5039367675781 }, "credit_bridge": { "test_acc": 0.668, "mean_bp_cosine": 0.16247223876416683, "mean_rho": 0.2929591853171587, "mean_nudge_001": -0.001968744763871655, "mean_nudge_003": -0.005892619024962187, "mean_nudge_01": -0.019490228034555912, "bp_cosine_per_layer": [ 0.05806349962949753, 0.19388319551944733, 0.1997416615486145, 0.19820059835910797 ], "rho_per_layer": [ 0.05481833964586258, 0.3572811782360077, 0.4186326265335083, 0.3411045968532562 ], "nudge_per_layer": [ -0.0286177359521389, -0.01660466194152832, -0.01654389500617981, -0.016194619238376617 ], "final_value_loss": 0.4729928963661194, "final_term_loss": 0.15622255086898804, "final_bridge_loss": 0.034451057982444765, "final_tgrad_loss": 0.28231928570270537 } }, "a0.5_L4_s456": { "bp": { "test_acc": 0.8855, "mean_bp_cosine": 0.7703797668218613, "mean_rho": 0.999986544251442, "mean_nudge_001": -0.017331300303339958, "mean_nudge_003": -0.04883333737961948, "mean_nudge_01": -0.13096668012440205, "bp_cosine_per_layer": [ 0.792909562587738, 0.7804109454154968, 0.7618845105171204, 0.7463140487670898 ], "rho_per_layer": [ 0.9999990463256836, 0.9999976754188538, 0.9999912977218628, 0.9999581575393677 ], "nudge_per_layer": [ -0.2336917370557785, -0.1588592678308487, -0.0874030739068985, -0.04391264170408249 ] }, "dfa": { "test_acc": 0.817, "mean_bp_cosine": 0.07182160578668118, "mean_rho": 0.09282832965254784, "mean_nudge_001": -0.0003612322652770672, "mean_nudge_003": -0.0010782249155454338, "mean_nudge_01": -0.0035168152826372534, "bp_cosine_per_layer": [ -0.010789386928081512, 0.08449429273605347, 0.10027095675468445, 0.1133105605840683 ], "rho_per_layer": [ -0.022197294980287552, 0.02167895808815956, 0.18905602395534515, 0.18277563154697418 ], "nudge_per_layer": [ -0.0005262043559923768, -0.005684657022356987, -0.003974507562816143, -0.0038818921893835068 ] }, "state_bridge": { "test_acc": 0.7675, "mean_bp_cosine": 0.4189938008785248, "mean_rho": 0.348801476880908, "mean_nudge_001": -0.006437422598537523, "mean_nudge_003": -0.018775500764604658, "mean_nudge_01": -0.056744852219708264, "bp_cosine_per_layer": [ 0.8223247528076172, 0.5256583094596863, 0.17289233207702637, 0.1550998091697693 ], "rho_per_layer": [ 0.7023045420646667, 0.4341009259223938, 0.10907860845327377, 0.14972183108329773 ], "nudge_per_layer": [ -0.19577056169509888, -0.022263746708631516, -0.004863455425947905, -0.0040816450491547585 ], "state_pred_error_per_layer": [ 6500.04248046875, 2179.638671875, 1695.392333984375, 1466.414306640625 ], "mean_state_pred_error": 2960.3719482421875 }, "credit_bridge": { "test_acc": 0.617, "mean_bp_cosine": 0.19905064720660448, "mean_rho": 0.27845675125718117, "mean_nudge_001": -0.0017877554928418249, "mean_nudge_003": -0.005349543411284685, "mean_nudge_01": -0.017666283063590527, "bp_cosine_per_layer": [ 0.0613640733063221, 0.23986753821372986, 0.24101567268371582, 0.25395530462265015 ], "rho_per_layer": [ 0.07897584140300751, 0.3258228003978729, 0.35056060552597046, 0.3584677577018738 ], "nudge_per_layer": [ -0.019116222858428955, -0.01759088784456253, -0.016751645132899284, -0.017206376418471336 ], "final_value_loss": 0.4424677228450775, "final_term_loss": 0.12383701201379299, "final_bridge_loss": 0.02319883658438921, "final_tgrad_loss": 0.29543187527656556 } }, "a0.5_L8_s42": { "bp": { "test_acc": 0.8715, "mean_bp_cosine": 0.808967724442482, "mean_rho": 0.999945156276226, "mean_nudge_001": -0.011501751199830323, "mean_nudge_003": -0.03289153927471489, "mean_nudge_01": -0.09379934845492244, "bp_cosine_per_layer": [ 0.8294941186904907, 0.823101282119751, 0.815970778465271, 0.8095235824584961, 0.8066394329071045, 0.8012610673904419, 0.7956615686416626, 0.7900899648666382 ], "rho_per_layer": [ 0.999998927116394, 0.999997615814209, 0.9999938011169434, 0.9999845027923584, 0.9999604225158691, 0.9999330043792725, 0.999873697757721, 0.999819278717041 ], "nudge_per_layer": [ -0.2528620958328247, -0.17482446134090424, -0.11384513974189758, -0.07313278317451477, -0.04917728155851364, -0.03563641011714935, -0.02781001478433609, -0.02310660108923912 ] }, "dfa": { "test_acc": 0.804, "mean_bp_cosine": 0.06822564313188195, "mean_rho": 0.08298261812888086, "mean_nudge_001": -5.095545566291548e-05, "mean_nudge_003": -0.0001400069668306969, "mean_nudge_01": -0.0003324677818454802, "bp_cosine_per_layer": [ -0.029147807508707047, 0.09590217471122742, 0.0916377604007721, 0.10714393854141235, 0.10661958158016205, 0.05731643736362457, 0.06447312980890274, 0.05185993015766144 ], "rho_per_layer": [ -0.08288859575986862, 0.14039787650108337, 0.1433517336845398, 0.1548803746700287, 0.09298646450042725, 0.08800182491540909, 0.09949750453233719, 0.02763376198709011 ], "nudge_per_layer": [ 0.019184602424502373, -0.0062692309729754925, -0.0031640869565308094, -0.0033698435872793198, -0.003262510057538748, -0.0018756084609776735, -0.0022803605534136295, -0.0016227040905505419 ] }, "state_bridge": { "test_acc": 0.7265, "mean_bp_cosine": 0.2766533959656954, "mean_rho": 0.24636799190193415, "mean_nudge_001": -0.003215807260858128, "mean_nudge_003": -0.00946771033341065, "mean_nudge_01": -0.029514338413719088, "bp_cosine_per_layer": [ 0.8188973069190979, 0.4508240222930908, 0.20906424522399902, 0.16529136896133423, 0.1612900048494339, 0.1292058229446411, 0.14780420064926147, 0.1308501958847046 ], "rho_per_layer": [ 0.7345798015594482, 0.40806666016578674, 0.18922817707061768, 0.1360217034816742, 0.11187596619129181, 0.1472519189119339, 0.12442155927419662, 0.11949814856052399 ], "nudge_per_layer": [ -0.19829359650611877, -0.01426965557038784, -0.005409521050751209, -0.004344802349805832, -0.0038992268964648247, -0.0032464847899973392, -0.003532196395099163, -0.00311922375112772 ], "state_pred_error_per_layer": [ 5640.76416015625, 3768.9912109375, 2209.80126953125, 1863.6328125, 1833.0330810546875, 2052.747802734375, 1687.1876220703125, 1871.18310546875 ], "mean_state_pred_error": 2615.9176330566406 }, "credit_bridge": { "test_acc": 0.5935, "mean_bp_cosine": 0.12945402902550995, "mean_rho": 0.20130334049463272, "mean_nudge_001": -0.0008943625434767455, "mean_nudge_003": -0.0026800249179359525, "mean_nudge_01": -0.00888864736771211, "bp_cosine_per_layer": [ 0.014968650415539742, 0.08219218999147415, 0.12315872311592102, 0.1350458562374115, 0.16176681220531464, 0.1692444086074829, 0.17282617092132568, 0.17642942070960999 ], "rho_per_layer": [ 0.09865723550319672, 0.1636437624692917, 0.19447889924049377, 0.2091381549835205, 0.2243458330631256, 0.23168572783470154, 0.2512606978416443, 0.23721641302108765 ], "nudge_per_layer": [ -0.006006492767482996, -0.006066862493753433, -0.008294559083878994, -0.008802486583590508, -0.009770728647708893, -0.010544568300247192, -0.010718668811023235, -0.010904812254011631 ], "final_value_loss": 0.3977136660575867, "final_term_loss": 0.13710964921712876, "final_bridge_loss": 0.018799800185859205, "final_tgrad_loss": 0.24180421624183654 } }, "a0.5_L8_s123": { "bp": { "test_acc": 0.868, "mean_bp_cosine": 0.8070773705840111, "mean_rho": 0.9999594464898109, "mean_nudge_001": -0.01314116342109628, "mean_nudge_003": -0.037657674751244485, "mean_nudge_01": -0.10643287934362888, "bp_cosine_per_layer": [ 0.8442516326904297, 0.8351782560348511, 0.8238041400909424, 0.811212420463562, 0.800036609172821, 0.789252519607544, 0.7798817157745361, 0.7730016708374023 ], "rho_per_layer": [ 0.9999992251396179, 0.9999983906745911, 0.9999960660934448, 0.99998939037323, 0.9999715685844421, 0.9999511241912842, 0.9999076128005981, 0.9998621940612793 ], "nudge_per_layer": [ -0.281266987323761, -0.2009655237197876, -0.1321270614862442, -0.08429376780986786, -0.05586903542280197, -0.039932381361722946, -0.031076554208993912, -0.025931723415851593 ] }, "dfa": { "test_acc": 0.8, "mean_bp_cosine": 0.07366255111992359, "mean_rho": 0.07397386711090803, "mean_nudge_001": -8.963746586232446e-05, "mean_nudge_003": -0.0002625839406391606, "mean_nudge_01": -0.0007999172084964812, "bp_cosine_per_layer": [ -0.02832707017660141, 0.13987571001052856, 0.07857642322778702, 0.061437755823135376, 0.0877213403582573, 0.09552451223134995, 0.07042402029037476, 0.08406771719455719 ], "rho_per_layer": [ 0.034647777676582336, 0.1467832326889038, 0.036387018859386444, 0.061308689415454865, 0.1038040965795517, 0.09023141860961914, 0.03672575205564499, 0.08190295100212097 ], "nudge_per_layer": [ 0.012113340198993683, -0.00537356361746788, -0.0023620566353201866, -0.001645779935643077, -0.002037967788055539, -0.0027765799313783646, -0.001988787204027176, -0.002327942755073309 ] }, "state_bridge": { "test_acc": 0.718, "mean_bp_cosine": 0.2835564874112606, "mean_rho": 0.2681517219170928, "mean_nudge_001": -0.0027127867688250262, "mean_nudge_003": -0.007993465682375245, "mean_nudge_01": -0.025023294263519347, "bp_cosine_per_layer": [ 0.8429861068725586, 0.3551006317138672, 0.246289923787117, 0.21019335091114044, 0.17309927940368652, 0.14295682311058044, 0.15966862440109253, 0.13815715909004211 ], "rho_per_layer": [ 0.7732758522033691, 0.2933464050292969, 0.24295836687088013, 0.22880345582962036, 0.1602117419242859, 0.11042392998933792, 0.18763291835784912, 0.14856110513210297 ], "nudge_per_layer": [ -0.15847395360469818, -0.010836342349648476, -0.007060050964355469, -0.005947217810899019, -0.004753550048917532, -0.004264200571924448, -0.004713424481451511, -0.0041376142762601376 ], "state_pred_error_per_layer": [ 3389.93017578125, 4477.634765625, 1038.2586669921875, 911.949951171875, 702.6972045898438, 659.9431762695312, 824.164794921875, 821.9293212890625 ], "mean_state_pred_error": 1603.3135070800781 }, "credit_bridge": { "test_acc": 0.6185, "mean_bp_cosine": 0.1525184395723045, "mean_rho": 0.23177338391542435, "mean_nudge_001": -0.0011540915293153375, "mean_nudge_003": -0.0034589636488817632, "mean_nudge_01": -0.01147624314762652, "bp_cosine_per_layer": [ 0.024487923830747604, 0.1551763117313385, 0.17440856993198395, 0.18776318430900574, 0.163284569978714, 0.17855775356292725, 0.162883460521698, 0.17358574271202087 ], "rho_per_layer": [ 0.024775028228759766, 0.25605684518814087, 0.2314496785402298, 0.2697882056236267, 0.24329833686351776, 0.3152206242084503, 0.23843368887901306, 0.2751646637916565 ], "nudge_per_layer": [ -0.010280786082148552, -0.011451397091150284, -0.01173271145671606, -0.012431012466549873, -0.01105603575706482, -0.011893808841705322, -0.011241357773542404, -0.011722835712134838 ], "final_value_loss": 0.3728706964492798, "final_term_loss": 0.12118487646579743, "final_bridge_loss": 0.015310661105066539, "final_tgrad_loss": 0.23637515988349914 } }, "a0.5_L8_s456": { "bp": { "test_acc": 0.8555, "mean_bp_cosine": 0.8527287542819977, "mean_rho": 0.9999570250511169, "mean_nudge_001": -0.01194569593644701, "mean_nudge_003": -0.033871308725792915, "mean_nudge_01": -0.09340489702299237, "bp_cosine_per_layer": [ 0.8810831904411316, 0.8730225563049316, 0.8656145334243774, 0.8578245639801025, 0.8490546345710754, 0.8397257328033447, 0.8311692476272583, 0.8243355751037598 ], "rho_per_layer": [ 0.9999987483024597, 0.9999980330467224, 0.9999955892562866, 0.9999854564666748, 0.9999719858169556, 0.999948263168335, 0.9998974204063416, 0.9998607039451599 ], "nudge_per_layer": [ -0.23731596767902374, -0.17453062534332275, -0.11795195937156677, -0.07674264907836914, -0.05135927349328995, -0.036901768296957016, -0.028667811304330826, -0.02376912161707878 ] }, "dfa": { "test_acc": 0.7985, "mean_bp_cosine": 0.09374871966429055, "mean_rho": 0.12165694544091821, "mean_nudge_001": -0.0004589786858559819, "mean_nudge_003": -0.0013758796121692285, "mean_nudge_01": -0.004529613710474223, "bp_cosine_per_layer": [ 0.027731137350201607, 0.10604047030210495, 0.13601002097129822, 0.07596071809530258, 0.0985877588391304, 0.08992618322372437, 0.1115533858537674, 0.10418008267879486 ], "rho_per_layer": [ -0.01300876960158348, 0.12486347556114197, 0.1899794042110443, 0.08801613748073578, 0.11934874951839447, 0.15905101597309113, 0.1374979466199875, 0.167507603764534 ], "nudge_per_layer": [ -0.011210349388420582, -0.00580610241740942, -0.004417200107127428, -0.002277099061757326, -0.003015844151377678, -0.0030763214454054832, -0.003333096392452717, -0.003100896719843149 ] }, "state_bridge": { "test_acc": 0.7505, "mean_bp_cosine": 0.25108483247458935, "mean_rho": 0.22314902767539024, "mean_nudge_001": -0.0032180442212847993, "mean_nudge_003": -0.009445599382161163, "mean_nudge_01": -0.029192803369369358, "bp_cosine_per_layer": [ 0.8145060539245605, 0.39495280385017395, 0.15602949261665344, 0.14954066276550293, 0.1189848780632019, 0.14297258853912354, 0.11834269762039185, 0.11334948241710663 ], "rho_per_layer": [ 0.7131592035293579, 0.31938445568084717, 0.13574811816215515, 0.13510501384735107, 0.12680554389953613, 0.13444143533706665, 0.1260460913181305, 0.09450235962867737 ], "nudge_per_layer": [ -0.20040717720985413, -0.012472910806536674, -0.003969438374042511, -0.003768536727875471, -0.0031205681152641773, -0.00363306887447834, -0.0031901695765554905, -0.002980557270348072 ], "state_pred_error_per_layer": [ 7208.7744140625, 4043.077392578125, 2431.6259765625, 2036.907958984375, 1977.547119140625, 1708.2327880859375, 1901.497314453125, 1821.6331787109375 ], "mean_state_pred_error": 2891.1620178222656 }, "credit_bridge": { "test_acc": 0.555, "mean_bp_cosine": 0.12328627426177263, "mean_rho": 0.16509935073554516, "mean_nudge_001": -0.000985160811978858, "mean_nudge_003": -0.002952177223050967, "mean_nudge_01": -0.009791445219889283, "bp_cosine_per_layer": [ 0.03383665531873703, 0.10378465056419373, 0.1522345095872879, 0.12050338089466095, 0.12787620723247528, 0.13963481783866882, 0.14998120069503784, 0.1584387719631195 ], "rho_per_layer": [ 0.02018044888973236, 0.14589767158031464, 0.20607107877731323, 0.18045580387115479, 0.1894514560699463, 0.16447019577026367, 0.1924649477005005, 0.2218032032251358 ], "nudge_per_layer": [ -0.00981426052749157, -0.00849075522273779, -0.010712197050452232, -0.009203735738992691, -0.009092864580452442, -0.00978156179189682, -0.010470103472471237, -0.010766083374619484 ], "final_value_loss": 0.5412774887084961, "final_term_loss": 0.23150972747802734, "final_bridge_loss": 0.02621319934129715, "final_tgrad_loss": 0.28355456256866457 } }, "a0.5_L12_s42": { "bp": { "test_acc": 0.8585, "mean_bp_cosine": 0.8183901458978653, "mean_rho": 0.9998798171679179, "mean_nudge_001": -0.011188714299350977, "mean_nudge_003": -0.03214278844340394, "mean_nudge_01": -0.09301998885348439, "bp_cosine_per_layer": [ 0.8657882213592529, 0.8541754484176636, 0.8436384201049805, 0.832263708114624, 0.8208613991737366, 0.8122134208679199, 0.8057717084884644, 0.8017721176147461, 0.7986597418785095, 0.7967146635055542, 0.7952107787132263, 0.7936121225357056 ], "rho_per_layer": [ 0.9999991655349731, 0.9999977350234985, 0.9999939203262329, 0.9999861121177673, 0.9999700784683228, 0.9999415874481201, 0.9999051094055176, 0.9998413324356079, 0.9998086094856262, 0.9997647404670715, 0.9996839165687561, 0.9996654987335205 ], "nudge_per_layer": [ -0.3461707830429077, -0.23528623580932617, -0.151905819773674, -0.0986800491809845, -0.06758347898721695, -0.04991145431995392, -0.039162103086709976, -0.032274406403303146, -0.02771918475627899, -0.024551689624786377, -0.02231007255613804, -0.020684588700532913 ] }, "dfa": { "test_acc": 0.791, "mean_bp_cosine": 0.07122930279001594, "mean_rho": 0.10592638049274683, "mean_nudge_001": -0.00047647209915642935, "mean_nudge_003": -0.0014203774189809337, "mean_nudge_01": -0.004644906090106815, "bp_cosine_per_layer": [ 0.023988625034689903, 0.0961650088429451, 0.10208466649055481, 0.07022920250892639, 0.09604307264089584, 0.07800711691379547, 0.08064866811037064, 0.015484225004911423, 0.04617477208375931, 0.10530653595924377, 0.06217034161090851, 0.07844939827919006 ], "rho_per_layer": [ -0.021668296307325363, 0.13302525877952576, 0.14676621556282043, 0.11482817679643631, 0.09230026602745056, 0.15572097897529602, 0.084853395819664, 0.06741442531347275, 0.09585727006196976, 0.17402535676956177, 0.1249062716960907, 0.10308724641799927 ], "nudge_per_layer": [ -0.013704460114240646, -0.009605048224329948, -0.004514003172516823, -0.00345102627761662, -0.003412249032407999, -0.0038682736922055483, -0.0031317053362727165, -0.0012646813411265612, -0.002444072160869837, -0.004538315813988447, -0.0027466074097901583, -0.0030584305059164762 ] }, "state_bridge": { "test_acc": 0.684, "mean_bp_cosine": 0.17019233045478663, "mean_rho": 0.14254611171782017, "mean_nudge_001": -0.0020112845547070415, "mean_nudge_003": -0.005932022473037553, "mean_nudge_01": -0.01860854132488991, "bp_cosine_per_layer": [ 0.7669721841812134, 0.3159201741218567, 0.12931227684020996, 0.11285121738910675, 0.11607236415147781, 0.09898597002029419, 0.08882047981023788, 0.08661012351512909, 0.08416090905666351, 0.07891363650560379, 0.08571723103523254, 0.07797139883041382 ], "rho_per_layer": [ 0.6671731472015381, 0.26826971769332886, 0.10340800136327744, 0.06843797862529755, 0.0669264942407608, 0.08176875114440918, 0.10683348774909973, 0.06641782820224762, 0.09755225479602814, 0.04477599635720253, 0.0847611278295517, 0.05422855541110039 ], "nudge_per_layer": [ -0.18878120183944702, -0.009807860478758812, -0.003219514386728406, -0.0027990485541522503, -0.0029557342641055584, -0.002479594200849533, -0.0022394107654690742, -0.0022681057453155518, -0.0022236446384340525, -0.0021239176858216524, -0.002271223347634077, -0.0021332399919629097 ], "state_pred_error_per_layer": [ 11168.2578125, 8443.55078125, 4301.74609375, 3486.31884765625, 3134.353515625, 2733.8369140625, 2666.431640625, 2571.673095703125, 2489.172119140625, 2732.81298828125, 3179.04296875, 3116.75927734375 ], "mean_state_pred_error": 4168.663004557292 }, "credit_bridge": { "test_acc": 0.571, "mean_bp_cosine": 0.08207109943032265, "mean_rho": 0.14231220919949314, "mean_nudge_001": -0.0008072109461257545, "mean_nudge_003": -0.0024178701957377293, "mean_nudge_01": -0.008001433801837265, "bp_cosine_per_layer": [ 0.009321734309196472, 0.04962325841188431, 0.06880976259708405, 0.07808464020490646, 0.08887439966201782, 0.0946938544511795, 0.09941184520721436, 0.09979166835546494, 0.10197117179632187, 0.10147081315517426, 0.09914548695087433, 0.0936545580625534 ], "rho_per_layer": [ -0.0036096302792429924, 0.0626683160662651, 0.10237888991832733, 0.12722331285476685, 0.138666570186615, 0.14094725251197815, 0.1720128357410431, 0.18820315599441528, 0.1971924901008606, 0.19901405274868011, 0.18742212653160095, 0.1956271380186081 ], "nudge_per_layer": [ -0.00799381174147129, -0.0042757005430758, -0.005744780879467726, -0.006799482740461826, -0.0076793828047811985, -0.008442584425210953, -0.008797119371592999, -0.009013657458126545, -0.009450232610106468, -0.009471941739320755, -0.009278533980250359, -0.009069977328181267 ], "final_value_loss": 0.95497522315979, "final_term_loss": 0.6288936994552612, "final_bridge_loss": 0.0344168305516243, "final_tgrad_loss": 0.2916646807193756 } }, "a0.5_L12_s123": { "bp": { "test_acc": 0.873, "mean_bp_cosine": 0.7415736317634583, "mean_rho": 0.9999099324146906, "mean_nudge_001": -0.010182377610666057, "mean_nudge_003": -0.029109801825446386, "mean_nudge_01": -0.08149818579355876, "bp_cosine_per_layer": [ 0.8060421943664551, 0.7944625616073608, 0.776944637298584, 0.7588614821434021, 0.7418972849845886, 0.7322722673416138, 0.724165678024292, 0.7197043299674988, 0.7152823209762573, 0.7119243144989014, 0.709496021270752, 0.7078304886817932 ], "rho_per_layer": [ 0.9999996423721313, 0.9999984502792358, 0.9999962449073792, 0.999989926815033, 0.9999754428863525, 0.9999579191207886, 0.999927818775177, 0.999887228012085, 0.9998751878738403, 0.9997915029525757, 0.9997637867927551, 0.9997560381889343 ], "nudge_per_layer": [ -0.2872876822948456, -0.20931334793567657, -0.13937950134277344, -0.09144100546836853, -0.06199805811047554, -0.04483566805720329, -0.03463221341371536, -0.02817366272211075, -0.023888790979981422, -0.02094285935163498, -0.01881779544055462, -0.01726764440536499 ] }, "dfa": { "test_acc": 0.7975, "mean_bp_cosine": 0.06331516286203016, "mean_rho": 0.04216622615543505, "mean_nudge_001": -0.00018036465129019538, "mean_nudge_003": -0.0005390862012670065, "mean_nudge_01": -0.0017679805799465005, "bp_cosine_per_layer": [ 0.004078918602317572, 0.1059243381023407, 0.07574640959501266, 0.07308551669120789, 0.066571444272995, 0.056897446513175964, 0.12165610492229462, 0.06462406367063522, 0.04925639182329178, 0.04602370783686638, 0.035759277641773224, 0.06015833467245102 ], "rho_per_layer": [ -0.011486321687698364, 0.04376016557216644, 0.037421341985464096, 0.05220063030719757, 0.09364941716194153, -0.03169674053788185, 0.15961039066314697, 0.10634914040565491, 0.03384825587272644, 0.02390148863196373, -0.010392685420811176, 0.00882963091135025 ], "nudge_per_layer": [ -0.0026711553800851107, -0.003443576628342271, -0.001918894238770008, -0.0020233127288520336, -0.0016027885721996427, -0.0010300527792423964, -0.002917288802564144, -0.001525502186268568, -0.0009703689720481634, -0.0010098961647599936, -0.0008095581433735788, -0.0012933723628520966 ] }, "state_bridge": { "test_acc": 0.69, "mean_bp_cosine": 0.17323703753451505, "mean_rho": 0.15179013274610043, "mean_nudge_001": -0.0019852635256635645, "mean_nudge_003": -0.005834519392616737, "mean_nudge_01": -0.018113662886510912, "bp_cosine_per_layer": [ 0.7613614797592163, 0.3896995782852173, 0.1309610903263092, 0.10979019105434418, 0.09164053201675415, 0.07316795736551285, 0.09407100081443787, 0.08964815735816956, 0.08292470127344131, 0.08648289740085602, 0.0870373547077179, 0.08205951005220413 ], "rho_per_layer": [ 0.6734859943389893, 0.37621644139289856, 0.08932416141033173, 0.11710754036903381, 0.07942430675029755, 0.058268360793590546, 0.05698401480913162, 0.12727776169776917, 0.03869014233350754, 0.06775026768445969, 0.0798446387052536, 0.05710796266794205 ], "nudge_per_layer": [ -0.1828688532114029, -0.012644743546843529, -0.003011793363839388, -0.0024378912057727575, -0.0020753229036927223, -0.0017659795703366399, -0.0021767604630440474, -0.0021774054039269686, -0.0019914493896067142, -0.0021301559172570705, -0.0020812181755900383, -0.0020023814868181944 ], "state_pred_error_per_layer": [ 12496.853515625, 8106.66162109375, 5059.9169921875, 4050.405517578125, 3454.0888671875, 2936.005859375, 2830.2294921875, 2863.870361328125, 3118.476806640625, 3786.361083984375, 3251.337890625, 3255.708740234375 ], "mean_state_pred_error": 4600.826395670573 }, "credit_bridge": { "test_acc": 0.6545, "mean_bp_cosine": -0.010937775252386928, "mean_rho": 0.02020453056320548, "mean_nudge_001": 7.376461204936884e-05, "mean_nudge_003": 0.00022662243100057822, "mean_nudge_01": 0.0008031492385877451, "bp_cosine_per_layer": [ -0.022733785212039948, -0.008966443128883839, -0.020323125645518303, -0.018136776983737946, -0.01592089980840683, -0.019174236804246902, -0.014491843990981579, -0.016727179288864136, -0.008660020306706429, -0.007111798040568829, 0.0011803526431322098, 0.019812453538179398 ], "rho_per_layer": [ -0.05155084282159805, 0.03821132332086563, 0.07940525561571121, 0.014752240851521492, 0.016678672283887863, 0.04385560005903244, 0.03276895359158516, -0.0006132312119007111, -0.024450641125440598, -0.029637619853019714, 0.058109357953071594, 0.06492529809474945 ], "nudge_per_layer": [ 0.00777015183120966, 0.00029987754533067346, 0.00037500812322832644, 0.00043640032527036965, 0.0005594875547103584, 0.0007026067469269037, 0.00045700964983552694, 0.000550219090655446, 0.00020889028382953256, 0.00013992507592774928, -0.0004261335707269609, -0.0014356517931446433 ], "final_value_loss": 1.470289986038208, "final_term_loss": 0.35005189373493195, "final_bridge_loss": 0.9054676934719086, "final_tgrad_loss": 0.2147704033613205 } }, "a0.5_L12_s456": { "bp": { "test_acc": 0.869, "mean_bp_cosine": 0.74041115740935, "mean_rho": 0.9998869399229685, "mean_nudge_001": -0.007502871051353092, "mean_nudge_003": -0.02132872895648082, "mean_nudge_01": -0.05926296177009741, "bp_cosine_per_layer": [ 0.802544355392456, 0.7848162055015564, 0.7700465321540833, 0.7570744752883911, 0.7440193891525269, 0.733696699142456, 0.7256457805633545, 0.7199540734291077, 0.7157886028289795, 0.7124584913253784, 0.7102035284042358, 0.7086857557296753 ], "rho_per_layer": [ 0.9999985694885254, 0.9999974966049194, 0.9999951720237732, 0.9999867677688599, 0.9999691843986511, 0.9999383687973022, 0.9999032020568848, 0.9998719692230225, 0.9998380541801453, 0.9998047351837158, 0.9996931552886963, 0.9996466040611267 ], "nudge_per_layer": [ -0.20057430863380432, -0.1480334997177124, -0.10051561146974564, -0.0672251284122467, -0.04664912819862366, -0.03438428044319153, -0.026901239529252052, -0.02211749367415905, -0.018912356346845627, -0.01670587807893753, -0.015132123604416847, -0.01400449313223362 ] }, "dfa": { "test_acc": 0.8, "mean_bp_cosine": 0.11594532305995624, "mean_rho": 0.11397512660672267, "mean_nudge_001": -0.00042300097387245234, "mean_nudge_003": -0.0012634664356786136, "mean_nudge_01": -0.0041745746857486665, "bp_cosine_per_layer": [ 0.04262668639421463, 0.2711353898048401, 0.1898173838853836, 0.1387583613395691, 0.09531620889902115, 0.1376194804906845, 0.07653278112411499, 0.08707999438047409, 0.09030534327030182, 0.10221391171216965, 0.06503984332084656, 0.09489849209785461 ], "rho_per_layer": [ 0.0572824701666832, 0.20349135994911194, 0.20036441087722778, 0.17148703336715698, 0.0623050183057785, 0.14095279574394226, 0.05262026563286781, 0.12087561190128326, 0.10731191188097, 0.061124786734580994, 0.07995927333831787, 0.10992658138275146 ], "nudge_per_layer": [ -0.009360449388623238, -0.00823313370347023, -0.005617598537355661, -0.004333253484219313, -0.0029935596976429224, -0.0036506117321550846, -0.002341690007597208, -0.002613792195916176, -0.0027945097535848618, -0.003059752518311143, -0.002222315641120076, -0.002874229568988085 ] }, "state_bridge": { "test_acc": 0.702, "mean_bp_cosine": 0.18659504254659018, "mean_rho": 0.1786113940179348, "mean_nudge_001": -0.002035917636627952, "mean_nudge_003": -0.0060201529073917, "mean_nudge_01": -0.019059499570478994, "bp_cosine_per_layer": [ 0.7615377902984619, 0.3628881573677063, 0.1397334486246109, 0.13061293959617615, 0.11024273186922073, 0.09393852204084396, 0.10003305226564407, 0.10794281214475632, 0.10362739115953445, 0.10100233554840088, 0.1030983105301857, 0.12448301911354065 ], "rho_per_layer": [ 0.7019654512405396, 0.3051794767379761, 0.12004730850458145, 0.11974964290857315, 0.1315283179283142, 0.11543487757444382, 0.08415813744068146, 0.14185138046741486, 0.12905646860599518, 0.0691840872168541, 0.12997986376285553, 0.09520171582698822 ], "nudge_per_layer": [ -0.1852446049451828, -0.011355580762028694, -0.003981029614806175, -0.0038526845164597034, -0.0031567115802317858, -0.0028176922351121902, -0.0029711187817156315, -0.0031051787082105875, -0.0029434349853545427, -0.002931142458692193, -0.002905802335590124, -0.0034490139223635197 ], "state_pred_error_per_layer": [ 8110.80078125, 4497.13623046875, 2448.651123046875, 2213.206298828125, 2038.592041015625, 2361.478515625, 2122.5341796875, 2178.11474609375, 2310.2265625, 2081.50390625, 2114.170654296875, 2407.947998046875 ], "mean_state_pred_error": 2907.0302530924478 }, "credit_bridge": { "test_acc": 0.6375, "mean_bp_cosine": 0.09317115658389714, "mean_rho": 0.17472073963532844, "mean_nudge_001": -0.000727248170733219, "mean_nudge_003": -0.0021776299763587303, "mean_nudge_01": -0.007225943821443555, "bp_cosine_per_layer": [ -0.0012992812553420663, 0.028360359370708466, 0.07222752273082733, 0.06384990364313126, 0.0745362713932991, 0.10235601663589478, 0.11958298832178116, 0.12767711281776428, 0.12978672981262207, 0.13033980131149292, 0.131376713514328, 0.13925974071025848 ], "rho_per_layer": [ -0.036700520664453506, 0.08258875459432602, 0.18529893457889557, 0.1350553333759308, 0.14971259236335754, 0.23370781540870667, 0.15428856015205383, 0.2475280910730362, 0.20684200525283813, 0.26022571325302124, 0.23770767450332642, 0.24039392173290253 ], "nudge_per_layer": [ 0.0004691533395089209, -0.002260619541630149, -0.007056917063891888, -0.005040735937654972, -0.005857650190591812, -0.007526098750531673, -0.008702869527041912, -0.009377602487802505, -0.009931675158441067, -0.01026402972638607, -0.010433407500386238, -0.010728873312473297 ], "final_value_loss": 0.6376882638454437, "final_term_loss": 0.3444120800256729, "final_bridge_loss": 0.029638116455078124, "final_tgrad_loss": 0.2636380691766739 } }, "a1.0_L2_s42": { "bp": { "test_acc": 0.7915, "mean_bp_cosine": 0.9376864433288574, "mean_rho": 0.9999988675117493, "mean_nudge_001": -0.040167851373553276, "mean_nudge_003": -0.11654911190271378, "mean_nudge_01": -0.34635232388973236, "bp_cosine_per_layer": [ 0.937772274017334, 0.9376006126403809 ], "rho_per_layer": [ 0.9999990463256836, 0.9999986886978149 ], "nudge_per_layer": [ -0.4167234003543854, -0.27598124742507935 ] }, "dfa": { "test_acc": 0.7025, "mean_bp_cosine": 0.007199164479970932, "mean_rho": 0.005297745577991009, "mean_nudge_001": -7.51367933844449e-05, "mean_nudge_003": -0.00021808056044392288, "mean_nudge_01": -0.000639150632196106, "bp_cosine_per_layer": [ -0.005922339856624603, 0.020320668816566467 ], "rho_per_layer": [ -0.003393532708287239, 0.013989023864269257 ], "nudge_per_layer": [ 5.659737507812679e-05, -0.0013348986394703388 ] }, "state_bridge": { "test_acc": 0.6735, "mean_bp_cosine": 0.5210442394018173, "mean_rho": 0.48512911796569824, "mean_nudge_001": -0.011768221855163574, "mean_nudge_003": -0.034833950689062476, "mean_nudge_01": -0.11062158830463886, "bp_cosine_per_layer": [ 0.6727198958396912, 0.3693685829639435 ], "rho_per_layer": [ 0.6028042435646057, 0.36745399236679077 ], "nudge_per_layer": [ -0.19608698785305023, -0.025156188756227493 ], "state_pred_error_per_layer": [ 4083.197509765625, 262.47991943359375 ], "mean_state_pred_error": 2172.8387145996094 }, "credit_bridge": { "test_acc": 0.5125, "mean_bp_cosine": 0.5678956210613251, "mean_rho": 0.5935650169849396, "mean_nudge_001": -0.00795893557369709, "mean_nudge_003": -0.023799872025847435, "mean_nudge_01": -0.07842501997947693, "bp_cosine_per_layer": [ 0.39126425981521606, 0.7445269823074341 ], "rho_per_layer": [ 0.44417524337768555, 0.7429547905921936 ], "nudge_per_layer": [ -0.08887787163257599, -0.06797216832637787 ], "final_value_loss": 0.6212269179344178, "final_term_loss": 0.05304660900235176, "final_bridge_loss": 0.010462266248464584, "final_tgrad_loss": 0.5577180406570434 } }, "a1.0_L2_s123": { "bp": { "test_acc": 0.809, "mean_bp_cosine": 0.9559402465820312, "mean_rho": 0.9999986886978149, "mean_nudge_001": -0.03185446374118328, "mean_nudge_003": -0.09129266440868378, "mean_nudge_01": -0.258349284529686, "bp_cosine_per_layer": [ 0.9615377187728882, 0.9503427743911743 ], "rho_per_layer": [ 0.9999992847442627, 0.9999980926513672 ], "nudge_per_layer": [ -0.3078651428222656, -0.20883342623710632 ] }, "dfa": { "test_acc": 0.7455, "mean_bp_cosine": 0.05108090303838253, "mean_rho": 0.05254313163459301, "mean_nudge_001": -0.0008275873260572553, "mean_nudge_003": -0.0024727254058234394, "mean_nudge_01": -0.008127598091959953, "bp_cosine_per_layer": [ 0.057878345251083374, 0.044283460825681686 ], "rho_per_layer": [ 0.047237616032361984, 0.057848647236824036 ], "nudge_per_layer": [ -0.013432648032903671, -0.0028225481510162354 ] }, "state_bridge": { "test_acc": 0.704, "mean_bp_cosine": 0.5101080536842346, "mean_rho": 0.45375654101371765, "mean_nudge_001": -0.00980611308477819, "mean_nudge_003": -0.02897635381668806, "mean_nudge_01": -0.09152827039361, "bp_cosine_per_layer": [ 0.6596157550811768, 0.3606003522872925 ], "rho_per_layer": [ 0.5883278250694275, 0.3191852569580078 ], "nudge_per_layer": [ -0.16100725531578064, -0.02204928547143936 ], "state_pred_error_per_layer": [ 4266.72509765625, 268.82733154296875 ], "mean_state_pred_error": 2267.7762145996094 }, "credit_bridge": { "test_acc": 0.536, "mean_bp_cosine": 0.5000061094760895, "mean_rho": 0.5169470459222794, "mean_nudge_001": -0.007283595856279135, "mean_nudge_003": -0.0217662351205945, "mean_nudge_01": -0.07154248096048832, "bp_cosine_per_layer": [ 0.3330497741699219, 0.6669624447822571 ], "rho_per_layer": [ 0.38148215413093567, 0.652411937713623 ], "nudge_per_layer": [ -0.08540114760398865, -0.05768381431698799 ], "final_value_loss": 0.6294826040267945, "final_term_loss": 0.06573917962312699, "final_bridge_loss": 0.01474373927861452, "final_tgrad_loss": 0.5489996829986572 } }, "a1.0_L2_s456": { "bp": { "test_acc": 0.7905, "mean_bp_cosine": 0.9266799688339233, "mean_rho": 0.9999988377094269, "mean_nudge_001": -0.03671902138739824, "mean_nudge_003": -0.10590140148997307, "mean_nudge_01": -0.30589545518159866, "bp_cosine_per_layer": [ 0.9291538000106812, 0.9242061376571655 ], "rho_per_layer": [ 0.9999991059303284, 0.9999985694885254 ], "nudge_per_layer": [ -0.36666756868362427, -0.24512334167957306 ] }, "dfa": { "test_acc": 0.73, "mean_bp_cosine": 0.03006170690059662, "mean_rho": 0.039368635043501854, "mean_nudge_001": -0.0004903670051135123, "mean_nudge_003": -0.0014640121953561902, "mean_nudge_01": -0.004785260884091258, "bp_cosine_per_layer": [ 0.02131858840584755, 0.03880482539534569 ], "rho_per_layer": [ 0.006325956434011459, 0.07241131365299225 ], "nudge_per_layer": [ -0.006775922141969204, -0.002794599626213312 ] }, "state_bridge": { "test_acc": 0.6755, "mean_bp_cosine": 0.5186701416969299, "mean_rho": 0.48068559169769287, "mean_nudge_001": -0.011591368587687612, "mean_nudge_003": -0.03431171411648393, "mean_nudge_01": -0.10901244357228279, "bp_cosine_per_layer": [ 0.669547975063324, 0.3677923083305359 ], "rho_per_layer": [ 0.6219363808631897, 0.33943480253219604 ], "nudge_per_layer": [ -0.19162335991859436, -0.026401527225971222 ], "state_pred_error_per_layer": [ 3936.27783203125, 288.300537109375 ], "mean_state_pred_error": 2112.2891845703125 }, "credit_bridge": { "test_acc": 0.544, "mean_bp_cosine": 0.5161235481500626, "mean_rho": 0.5938436686992645, "mean_nudge_001": -0.0068406532518565655, "mean_nudge_003": -0.020453338511288166, "mean_nudge_01": -0.0673653893172741, "bp_cosine_per_layer": [ 0.35489675402641296, 0.6773503422737122 ], "rho_per_layer": [ 0.4646533727645874, 0.7230339646339417 ], "nudge_per_layer": [ -0.07572861015796661, -0.059002168476581573 ], "final_value_loss": 0.6262330022811889, "final_term_loss": 0.049621654653549195, "final_bridge_loss": 0.011661755262315273, "final_tgrad_loss": 0.5649495909690857 } }, "a1.0_L4_s42": { "bp": { "test_acc": 0.7875, "mean_bp_cosine": 0.882385790348053, "mean_rho": 0.999995619058609, "mean_nudge_001": -0.03646231605671346, "mean_nudge_003": -0.10610398091375828, "mean_nudge_01": -0.31568005308508873, "bp_cosine_per_layer": [ 0.8902335166931152, 0.8886877298355103, 0.8827626705169678, 0.8678592443466187 ], "rho_per_layer": [ 0.9999993443489075, 0.9999990463256836, 0.9999973773956299, 0.9999867081642151 ], "nudge_per_layer": [ -0.49256381392478943, -0.4056996703147888, -0.2497977763414383, -0.11465895175933838 ] }, "dfa": { "test_acc": 0.7075, "mean_bp_cosine": 0.060902527533471584, "mean_rho": 0.052925301948562264, "mean_nudge_001": -0.00048208528460236266, "mean_nudge_003": -0.0014429198927246034, "mean_nudge_01": -0.004763618926517665, "bp_cosine_per_layer": [ 0.026215966790914536, 0.05776425451040268, 0.07666948437690735, 0.08296040445566177 ], "rho_per_layer": [ 0.04998716711997986, 0.09949691593647003, 0.049676910042762756, 0.012540214695036411 ], "nudge_per_layer": [ -0.008544353768229485, -0.0038241599686443806, -0.0033461027778685093, -0.003339859191328287 ] }, "state_bridge": { "test_acc": 0.6625, "mean_bp_cosine": 0.3484594225883484, "mean_rho": 0.3289090357720852, "mean_nudge_001": -0.005907032289542258, "mean_nudge_003": -0.01748002166277729, "mean_nudge_01": -0.05545389896724373, "bp_cosine_per_layer": [ 0.6404350996017456, 0.4214136302471161, 0.18068093061447144, 0.15130802989006042 ], "rho_per_layer": [ 0.5499319434165955, 0.368219256401062, 0.21391229331493378, 0.1835726499557495 ], "nudge_per_layer": [ -0.18587297201156616, -0.0235043503344059, -0.00688221026211977, -0.005556063260883093 ], "state_pred_error_per_layer": [ 10066.107421875, 1273.9140625, 696.594482421875, 612.3014526367188 ], "mean_state_pred_error": 3162.2293548583984 }, "credit_bridge": { "test_acc": 0.558, "mean_bp_cosine": 0.45794273912906647, "mean_rho": 0.5322257168591022, "mean_nudge_001": -0.003842960635665804, "mean_nudge_003": -0.011505396338179708, "mean_nudge_01": -0.038075629621744156, "bp_cosine_per_layer": [ 0.17809242010116577, 0.5349531173706055, 0.5522423982620239, 0.5664830207824707 ], "rho_per_layer": [ 0.22843755781650543, 0.5910820364952087, 0.6707864999771118, 0.638596773147583 ], "nudge_per_layer": [ -0.044362872838974, -0.03722546622157097, -0.03515344485640526, -0.0355607345700264 ], "final_value_loss": 0.38718592133522034, "final_term_loss": 0.06566130868196488, "final_bridge_loss": 0.01777836950570345, "final_tgrad_loss": 0.30374624242782594 } }, "a1.0_L4_s123": { "bp": { "test_acc": 0.786, "mean_bp_cosine": 0.8366721421480179, "mean_rho": 0.9999945610761642, "mean_nudge_001": -0.03740503964945674, "mean_nudge_003": -0.1080933827906847, "mean_nudge_01": -0.3170367181301117, "bp_cosine_per_layer": [ 0.8610074520111084, 0.8541143536567688, 0.8321727514266968, 0.7993940114974976 ], "rho_per_layer": [ 0.9999990463256836, 0.9999986886978149, 0.9999971389770508, 0.9999833703041077 ], "nudge_per_layer": [ -0.492245614528656, -0.40577083826065063, -0.25304561853408813, -0.117084801197052 ] }, "dfa": { "test_acc": 0.682, "mean_bp_cosine": 0.049554066732525826, "mean_rho": 0.06381266703829169, "mean_nudge_001": -0.00014734033902641386, "mean_nudge_003": -0.0004330205119913444, "mean_nudge_01": -0.001343857147730887, "bp_cosine_per_layer": [ -0.008613362908363342, 0.039367884397506714, 0.07619941979646683, 0.0912623256444931 ], "rho_per_layer": [ -0.007624803110957146, 0.008791688829660416, 0.13103844225406647, 0.12304534018039703 ], "nudge_per_layer": [ 0.005509334150701761, -0.003195145633071661, -0.0036095697432756424, -0.004080047365278006 ] }, "state_bridge": { "test_acc": 0.6305, "mean_bp_cosine": 0.3143155351281166, "mean_rho": 0.30846646428108215, "mean_nudge_001": -0.005087231853394769, "mean_nudge_003": -0.015093521040398628, "mean_nudge_01": -0.04834054387174547, "bp_cosine_per_layer": [ 0.5847928524017334, 0.3501134514808655, 0.17105570435523987, 0.15130013227462769 ], "rho_per_layer": [ 0.5387442708015442, 0.3080189526081085, 0.17941737174987793, 0.20768526196479797 ], "nudge_per_layer": [ -0.16164088249206543, -0.01902196928858757, -0.006823995150625706, -0.005875328555703163 ], "state_pred_error_per_layer": [ 10103.57421875, 1132.8861083984375, 636.5859375, 537.355224609375 ], "mean_state_pred_error": 3102.600372314453 }, "credit_bridge": { "test_acc": 0.502, "mean_bp_cosine": 0.39079025387763977, "mean_rho": 0.4630395397543907, "mean_nudge_001": -0.0038928183494135737, "mean_nudge_003": -0.011659935116767883, "mean_nudge_01": -0.03865493927150965, "bp_cosine_per_layer": [ 0.1632033884525299, 0.4602273404598236, 0.4661305546760559, 0.47359973192214966 ], "rho_per_layer": [ 0.20324298739433289, 0.5429707169532776, 0.5653814077377319, 0.5405630469322205 ], "nudge_per_layer": [ -0.04415123537182808, -0.037312425673007965, -0.03657253086566925, -0.036583565175533295 ], "final_value_loss": 0.5799021621704101, "final_term_loss": 0.1424343505382538, "final_bridge_loss": 0.03182844363451004, "final_tgrad_loss": 0.4056393639087677 } }, "a1.0_L4_s456": { "bp": { "test_acc": 0.792, "mean_bp_cosine": 0.8552748262882233, "mean_rho": 0.9999953359365463, "mean_nudge_001": -0.03656455362215638, "mean_nudge_003": -0.10595990065485239, "mean_nudge_01": -0.31139715015888214, "bp_cosine_per_layer": [ 0.8668075799942017, 0.8621845245361328, 0.8550785779953003, 0.8370286226272583 ], "rho_per_layer": [ 0.9999995827674866, 0.9999990463256836, 0.999997615814209, 0.9999850988388062 ], "nudge_per_layer": [ -0.4802546799182892, -0.3984469175338745, -0.2503964900970459, -0.11649051308631897 ] }, "dfa": { "test_acc": 0.68, "mean_bp_cosine": 0.052356213971506804, "mean_rho": 0.05650564655661583, "mean_nudge_001": -0.0003577733878046274, "mean_nudge_003": -0.0010662981367204338, "mean_nudge_01": -0.0034739490947686136, "bp_cosine_per_layer": [ -0.001797341974452138, 0.06668862700462341, 0.05448886379599571, 0.09004470705986023 ], "rho_per_layer": [ -0.01287321001291275, 0.007983170449733734, 0.10568365454673767, 0.12522897124290466 ], "nudge_per_layer": [ -0.0016110537108033895, -0.004868227988481522, -0.0030549662187695503, -0.004361548461019993 ] }, "state_bridge": { "test_acc": 0.651, "mean_bp_cosine": 0.35348402336239815, "mean_rho": 0.3312904089689255, "mean_nudge_001": -0.005755784470238723, "mean_nudge_003": -0.017025350069161505, "mean_nudge_01": -0.05395022651646286, "bp_cosine_per_layer": [ 0.6501579284667969, 0.4189426302909851, 0.1800653487443924, 0.1647701859474182 ], "rho_per_layer": [ 0.5907484292984009, 0.36159276962280273, 0.19337056577205658, 0.1794498711824417 ], "nudge_per_layer": [ -0.1790994107723236, -0.023556631058454514, -0.006848426535725594, -0.0062964376993477345 ], "state_pred_error_per_layer": [ 7107.7138671875, 890.22314453125, 493.5357360839844, 407.69677734375 ], "mean_state_pred_error": 2224.792381286621 }, "credit_bridge": { "test_acc": 0.487, "mean_bp_cosine": 0.5049829035997391, "mean_rho": 0.5321159400045872, "mean_nudge_001": -0.004336995130870491, "mean_nudge_003": -0.012986946385353804, "mean_nudge_01": -0.04301031678915024, "bp_cosine_per_layer": [ 0.20670706033706665, 0.594096302986145, 0.6002212762832642, 0.6189069747924805 ], "rho_per_layer": [ 0.24109239876270294, 0.6317664384841919, 0.6270265579223633, 0.6285783648490906 ], "nudge_per_layer": [ -0.03731346130371094, -0.04525235667824745, -0.044518306851387024, -0.04495714232325554 ], "final_value_loss": 0.46226135368347165, "final_term_loss": 0.049428026616573334, "final_bridge_loss": 0.022721626716852187, "final_tgrad_loss": 0.39011170201301576 } }, "a1.0_L8_s42": { "bp": { "test_acc": 0.763, "mean_bp_cosine": 0.8278333768248558, "mean_rho": 0.9999853745102882, "mean_nudge_001": -0.029990217997692525, "mean_nudge_003": -0.08752790885046124, "mean_nudge_01": -0.2652519349940121, "bp_cosine_per_layer": [ 0.85625159740448, 0.8527897596359253, 0.847720742225647, 0.8408411145210266, 0.831807017326355, 0.8158122301101685, 0.7960221171379089, 0.7814224362373352 ], "rho_per_layer": [ 0.999999463558197, 0.9999992847442627, 0.9999988079071045, 0.9999978542327881, 0.999994695186615, 0.9999883770942688, 0.9999692440032959, 0.9999352693557739 ], "nudge_per_layer": [ -0.5478029251098633, -0.4737934172153473, -0.3800549805164337, -0.2769272029399872, -0.18745023012161255, -0.12213317304849625, -0.07957549393177032, -0.05427805706858635 ] }, "dfa": { "test_acc": 0.657, "mean_bp_cosine": 0.046988741494715214, "mean_rho": 0.0527694383636117, "mean_nudge_001": -0.00015309322225220967, "mean_nudge_003": -0.00045631649845745414, "mean_nudge_01": -0.0014890892343828455, "bp_cosine_per_layer": [ -0.01574046164751053, 0.07191511988639832, 0.043905604630708694, 0.07308061420917511, 0.07518148422241211, 0.026394926011562347, 0.06854978203773499, 0.03262286260724068 ], "rho_per_layer": [ -0.02709903195500374, 0.1010747104883194, 0.06438975036144257, 0.07812117040157318, 0.0766223892569542, 0.026489075273275375, 0.09218208491802216, 0.010375358164310455 ], "nudge_per_layer": [ 0.007441862486302853, -0.0052239359356462955, -0.0022090799175202847, -0.003076428547501564, -0.0031523762736469507, -0.0013899998739361763, -0.0030318819917738438, -0.0012708738213405013 ] }, "state_bridge": { "test_acc": 0.65, "mean_bp_cosine": 0.2592074144631624, "mean_rho": 0.23237370885908604, "mean_nudge_001": -0.0031611585363862105, "mean_nudge_003": -0.009385082448716275, "mean_nudge_01": -0.030122902884613723, "bp_cosine_per_layer": [ 0.6320836544036865, 0.4503885507583618, 0.22882479429244995, 0.17507168650627136, 0.15456035733222961, 0.15327030420303345, 0.14079461991786957, 0.1386653482913971 ], "rho_per_layer": [ 0.5491607189178467, 0.3870346248149872, 0.23711450397968292, 0.15141981840133667, 0.12508505582809448, 0.15914584696292877, 0.13365353643894196, 0.11637556552886963 ], "nudge_per_layer": [ -0.18924641609191895, -0.02186501771211624, -0.007085676770657301, -0.0053551034070551395, -0.00462403055280447, -0.004574076272547245, -0.004170445259660482, -0.004062457010149956 ], "state_pred_error_per_layer": [ 9406.7255859375, 1991.255126953125, 1233.173095703125, 996.4456787109375, 824.75537109375, 800.2864990234375, 821.3734130859375, 840.202392578125 ], "mean_state_pred_error": 2114.277145385742 }, "credit_bridge": { "test_acc": 0.5165, "mean_bp_cosine": 0.41811549477279186, "mean_rho": 0.45575808361172676, "mean_nudge_001": -0.002801764407195151, "mean_nudge_003": -0.00839249708224088, "mean_nudge_01": -0.02782702073454857, "bp_cosine_per_layer": [ 0.159218892455101, 0.41163527965545654, 0.4353415369987488, 0.4608742594718933, 0.46223077178001404, 0.4579800069332123, 0.4774817228317261, 0.48016148805618286 ], "rho_per_layer": [ 0.16694357991218567, 0.4445253610610962, 0.4925374388694763, 0.48059922456741333, 0.4989780783653259, 0.5061379075050354, 0.5181043148040771, 0.5382387638092041 ], "nudge_per_layer": [ -0.030580628663301468, -0.027534715831279755, -0.027298064902424812, -0.027803048491477966, -0.027431834489107132, -0.02679138071835041, -0.027646536007523537, -0.02752995677292347 ], "final_value_loss": 0.35056146993637083, "final_term_loss": 0.0784306697010994, "final_bridge_loss": 0.0149920065253973, "final_tgrad_loss": 0.25713879618644714 } }, "a1.0_L8_s123": { "bp": { "test_acc": 0.7715, "mean_bp_cosine": 0.8317327573895454, "mean_rho": 0.9999875128269196, "mean_nudge_001": -0.028257266676519066, "mean_nudge_003": -0.08214446459896863, "mean_nudge_01": -0.24747182056307793, "bp_cosine_per_layer": [ 0.8705248832702637, 0.8637571334838867, 0.8509291410446167, 0.8393645286560059, 0.8308560848236084, 0.8178726434707642, 0.800449013710022, 0.780108630657196 ], "rho_per_layer": [ 0.9999995231628418, 0.9999995827674866, 0.9999989867210388, 0.9999984502792358, 0.9999953508377075, 0.9999901056289673, 0.9999750852584839, 0.9999430179595947 ], "nudge_per_layer": [ -0.5067576169967651, -0.4360688626766205, -0.3518902063369751, -0.26053690910339355, -0.1787266880273819, -0.1170111671090126, -0.07650406658649445, -0.052279047667980194 ] }, "dfa": { "test_acc": 0.67, "mean_bp_cosine": 0.06255943619180471, "mean_rho": 0.0582913524704054, "mean_nudge_001": -0.00016377134670619853, "mean_nudge_003": -0.00048249874453176744, "mean_nudge_01": -0.0015042584564071149, "bp_cosine_per_layer": [ -0.006531993858516216, 0.06646430492401123, 0.03470785170793533, 0.06075863167643547, 0.08359120786190033, 0.10361115634441376, 0.07729409635066986, 0.08058023452758789 ], "rho_per_layer": [ -0.009454472921788692, 0.027615420520305634, 0.016344428062438965, 0.07737872004508972, 0.07998150587081909, 0.10012705624103546, 0.06766189634799957, 0.10667626559734344 ], "nudge_per_layer": [ 0.008429424837231636, -0.004321999382227659, -0.0012797827366739511, -0.0022665630094707012, -0.003274995367974043, -0.0036867314483970404, -0.0025471593253314495, -0.0030862612184137106 ] }, "state_bridge": { "test_acc": 0.63, "mean_bp_cosine": 0.2676183916628361, "mean_rho": 0.249815184623003, "mean_nudge_001": -0.0025806196208577603, "mean_nudge_003": -0.007662267831619829, "mean_nudge_01": -0.024620095442514867, "bp_cosine_per_layer": [ 0.6138853430747986, 0.41074037551879883, 0.25203365087509155, 0.20834164321422577, 0.16624599695205688, 0.15929004549980164, 0.15449269115924835, 0.175917387008667 ], "rho_per_layer": [ 0.5534218549728394, 0.3554393947124481, 0.28831934928894043, 0.1869889795780182, 0.14766299724578857, 0.176305890083313, 0.1294843554496765, 0.16089865565299988 ], "nudge_per_layer": [ -0.14630448818206787, -0.017482426017522812, -0.00796731747686863, -0.006267632823437452, -0.00492177763953805, -0.00457757618278265, -0.004429594613611698, -0.00500995060428977 ], "state_pred_error_per_layer": [ 7478.478515625, 1555.7657470703125, 846.884521484375, 607.9534301757812, 659.867919921875, 614.796630859375, 546.969482421875, 518.796142578125 ], "mean_state_pred_error": 1603.6890487670898 }, "credit_bridge": { "test_acc": 0.5695, "mean_bp_cosine": 0.33955336920917034, "mean_rho": 0.4137213062494993, "mean_nudge_001": -0.002328026050236076, "mean_nudge_003": -0.006974202115088701, "mean_nudge_01": -0.023127662017941475, "bp_cosine_per_layer": [ 0.08390675485134125, 0.2993828356266022, 0.37482553720474243, 0.3876585066318512, 0.3966418504714966, 0.3863479495048523, 0.3938036561012268, 0.39385986328125 ], "rho_per_layer": [ 0.09506039321422577, 0.3888479173183441, 0.42724472284317017, 0.5000925064086914, 0.4708487391471863, 0.47097086906433105, 0.48436442017555237, 0.4723408818244934 ], "nudge_per_layer": [ -0.027181372046470642, -0.02298363298177719, -0.02282092720270157, -0.022664468735456467, -0.02277727797627449, -0.02207607962191105, -0.022199643775820732, -0.02231789380311966 ], "final_value_loss": 0.32433377566337585, "final_term_loss": 0.09466714213788509, "final_bridge_loss": 0.018076158130913973, "final_tgrad_loss": 0.21159047515392304 } }, "a1.0_L8_s456": { "bp": { "test_acc": 0.753, "mean_bp_cosine": 0.8595180436968803, "mean_rho": 0.9999836012721062, "mean_nudge_001": -0.029901663539931178, "mean_nudge_003": -0.08636575657874346, "mean_nudge_01": -0.25306460773572326, "bp_cosine_per_layer": [ 0.8814424872398376, 0.8780966401100159, 0.8716025352478027, 0.8659577369689941, 0.8611791133880615, 0.8505971431732178, 0.840730607509613, 0.8265380859375 ], "rho_per_layer": [ 0.9999989867210388, 0.9999986886978149, 0.9999985694885254, 0.9999977350234985, 0.9999950528144836, 0.9999887943267822, 0.9999659657478333, 0.9999250173568726 ], "nudge_per_layer": [ -0.4892617166042328, -0.43737056851387024, -0.36532458662986755, -0.27772778272628784, -0.1931857019662857, -0.12626537680625916, -0.08131371438503265, -0.05406741425395012 ] }, "dfa": { "test_acc": 0.6545, "mean_bp_cosine": 0.07594231329858303, "mean_rho": 0.08383850008249283, "mean_nudge_001": -0.00040534641811973415, "mean_nudge_003": -0.0011938292009290308, "mean_nudge_01": -0.0038653643568977714, "bp_cosine_per_layer": [ 0.025392260402441025, 0.05978121981024742, 0.09360896795988083, 0.06215091794729233, 0.08870920538902283, 0.07948237657546997, 0.10833052545785904, 0.0900830328464508 ], "rho_per_layer": [ -0.015408031642436981, 0.08876338601112366, 0.15722069144248962, 0.07516318559646606, 0.12132002413272858, 0.08764688670635223, 0.0905974954366684, 0.06540436297655106 ], "nudge_per_layer": [ -0.002068763133138418, -0.004839141853153706, -0.004507332108914852, -0.002498403424397111, -0.004324691370129585, -0.004049480427056551, -0.004787375684827566, -0.0038477268535643816 ] }, "state_bridge": { "test_acc": 0.6315, "mean_bp_cosine": 0.22162932250648737, "mean_rho": 0.200162704102695, "mean_nudge_001": -0.002804466668749228, "mean_nudge_003": -0.008329451040481217, "mean_nudge_01": -0.026790971111040562, "bp_cosine_per_layer": [ 0.6046093106269836, 0.3773959279060364, 0.1675240397453308, 0.13364183902740479, 0.12548640370368958, 0.11966440826654434, 0.11708953976631165, 0.12762311100959778 ], "rho_per_layer": [ 0.5598001480102539, 0.35775649547576904, 0.14680343866348267, 0.103827565908432, 0.12472978234291077, 0.08794894069433212, 0.12868158519268036, 0.09175367653369904 ], "nudge_per_layer": [ -0.17137765884399414, -0.018192056566476822, -0.005168645642697811, -0.004143164958804846, -0.0039602769538760185, -0.0037884372286498547, -0.003734107594937086, -0.00396342109888792 ], "state_pred_error_per_layer": [ 11751.603515625, 2306.73095703125, 1321.091796875, 1033.674560546875, 973.6993408203125, 865.9056396484375, 869.378173828125, 839.1387329101562 ], "mean_state_pred_error": 2495.1528396606445 }, "credit_bridge": { "test_acc": 0.5275, "mean_bp_cosine": 0.33450845070183277, "mean_rho": 0.39851769525557756, "mean_nudge_001": -0.0025428086810279638, "mean_nudge_003": -0.007617499912157655, "mean_nudge_01": -0.025270385667681694, "bp_cosine_per_layer": [ 0.1268395036458969, 0.30569249391555786, 0.34143707156181335, 0.3805205225944519, 0.3833070397377014, 0.38344305753707886, 0.3742453455924988, 0.3805825710296631 ], "rho_per_layer": [ 0.10492167621850967, 0.4008791148662567, 0.42003583908081055, 0.4331238567829132, 0.4291571080684662, 0.4853004217147827, 0.4488537311553955, 0.46586981415748596 ], "nudge_per_layer": [ -0.02585173025727272, -0.023091215640306473, -0.0244273878633976, -0.026221776381134987, -0.026039035990834236, -0.025849176570773125, -0.025189466774463654, -0.02549329586327076 ], "final_value_loss": 0.41436073665618894, "final_term_loss": 0.08663651337623596, "final_bridge_loss": 0.013127461233735085, "final_tgrad_loss": 0.3145967625617981 } }, "a1.0_L12_s42": { "bp": { "test_acc": 0.761, "mean_bp_cosine": 0.8604618857304255, "mean_rho": 0.9999715387821198, "mean_nudge_001": -0.027060116563613217, "mean_nudge_003": -0.07916912420963247, "mean_nudge_01": -0.2410698818663756, "bp_cosine_per_layer": [ 0.9098777174949646, 0.9046642780303955, 0.8960509300231934, 0.8827707171440125, 0.8719829320907593, 0.8605888485908508, 0.8495617508888245, 0.8414114117622375, 0.8343594074249268, 0.8291183114051819, 0.8250235915184021, 0.8201327323913574 ], "rho_per_layer": [ 0.9999990463256836, 0.9999990463256836, 0.9999988675117493, 0.9999977946281433, 0.9999968409538269, 0.9999942779541016, 0.9999903440475464, 0.9999786615371704, 0.9999630451202393, 0.9999446272850037, 0.9999108910560608, 0.9998850226402283 ], "nudge_per_layer": [ -0.6227189302444458, -0.5321269631385803, -0.43762683868408203, -0.34503036737442017, -0.26150017976760864, -0.1957385241985321, -0.14543747901916504, -0.10888919234275818, -0.08308196067810059, -0.06497863680124283, -0.05229911208152771, -0.043410398066043854 ] }, "dfa": { "test_acc": 0.6555, "mean_bp_cosine": 0.07142843399196863, "mean_rho": 0.08962296570340793, "mean_nudge_001": -0.00027090053966579336, "mean_nudge_003": -0.0007314661440129081, "mean_nudge_01": -0.0026503523016193262, "bp_cosine_per_layer": [ 0.019931811839342117, 0.08357247710227966, 0.07596792280673981, 0.07454836368560791, 0.08405900746583939, 0.09273561835289001, 0.08532143384218216, 0.014841631054878235, 0.05702097713947296, 0.11365848034620285, 0.0635562390089035, 0.09192724525928497 ], "rho_per_layer": [ -0.021908380091190338, 0.08644437789916992, 0.133758544921875, 0.11560394614934921, 0.09019643068313599, 0.10901004821062088, 0.05401141196489334, 0.06626297533512115, 0.10668720304965973, 0.11482831090688705, 0.07995637506246567, 0.14062434434890747 ], "nudge_per_layer": [ 0.012093277648091316, -0.00814887322485447, -0.0036774289328604937, -0.0036643892526626587, -0.0034756590612232685, -0.0045063551515340805, -0.003735880134627223, -0.00101565255317837, -0.0033070247154682875, -0.005390047561377287, -0.0029281126335263252, -0.004048082046210766 ] }, "state_bridge": { "test_acc": 0.612, "mean_bp_cosine": 0.2312017778555552, "mean_rho": 0.2218381396184365, "mean_nudge_001": -0.0020883519706937173, "mean_nudge_003": -0.0062129807386857765, "mean_nudge_01": -0.02010280774751057, "bp_cosine_per_layer": [ 0.6137553453445435, 0.458263099193573, 0.30694735050201416, 0.21984770894050598, 0.18884725868701935, 0.1519574671983719, 0.14165323972702026, 0.13691727817058563, 0.1416233777999878, 0.126474529504776, 0.13544593751430511, 0.15268874168395996 ], "rho_per_layer": [ 0.5528259873390198, 0.43722641468048096, 0.3177984952926636, 0.20145072042942047, 0.17441785335540771, 0.10548915714025497, 0.16265320777893066, 0.1351865828037262, 0.16532285511493683, 0.1285737156867981, 0.14820542931556702, 0.13290725648403168 ], "nudge_per_layer": [ -0.16301177442073822, -0.02157333306968212, -0.011153988540172577, -0.00765608623623848, -0.006283854134380817, -0.005166183691471815, -0.0045799510553479195, -0.004474613815546036, -0.004547867923974991, -0.003950329497456551, -0.004098072648048401, -0.004737637937068939 ], "state_pred_error_per_layer": [ 9663.3720703125, 2355.42919921875, 1189.673095703125, 946.2671508789062, 890.0131225585938, 836.83837890625, 765.7700805664062, 776.2135009765625, 909.9769287109375, 744.7386474609375, 809.6233520507812, 764.1907958984375 ], "mean_state_pred_error": 1721.0088602701824 }, "credit_bridge": { "test_acc": 0.551, "mean_bp_cosine": 0.24796992043654123, "mean_rho": 0.34521655660743517, "mean_nudge_001": -0.0019821600387028107, "mean_nudge_003": -0.005940369796007872, "mean_nudge_01": -0.01972930335129301, "bp_cosine_per_layer": [ 0.050826266407966614, 0.1324397623538971, 0.19395378232002258, 0.22954219579696655, 0.24835197627544403, 0.26261672377586365, 0.2866933345794678, 0.3069983422756195, 0.3148944675922394, 0.31817084550857544, 0.3155951499938965, 0.31555619835853577 ], "rho_per_layer": [ 0.024354537948966026, 0.23902058601379395, 0.29458120465278625, 0.31519168615341187, 0.3208061456680298, 0.38384467363357544, 0.38460177183151245, 0.4478553831577301, 0.41032421588897705, 0.452781081199646, 0.43433815240859985, 0.434899240732193 ], "nudge_per_layer": [ -0.018513288348913193, -0.014132995158433914, -0.016322217881679535, -0.017585258930921555, -0.018846090883016586, -0.02027801051735878, -0.021141095086932182, -0.02164183184504509, -0.02207053080201149, -0.022233182564377785, -0.02206714265048504, -0.021919995546340942 ], "final_value_loss": 0.40481987361907956, "final_term_loss": 0.14867231681346893, "final_bridge_loss": 0.015895452450215817, "final_tgrad_loss": 0.24025210301876068 } }, "a1.0_L12_s123": { "bp": { "test_acc": 0.7615, "mean_bp_cosine": 0.8364224582910538, "mean_rho": 0.9999674806992213, "mean_nudge_001": -0.021432781708426774, "mean_nudge_003": -0.061891427962109447, "mean_nudge_01": -0.18090539829184613, "bp_cosine_per_layer": [ 0.86488276720047, 0.8622424006462097, 0.8600971698760986, 0.85722416639328, 0.8523166179656982, 0.8457985520362854, 0.8359642028808594, 0.828606367111206, 0.8221874237060547, 0.8125513792037964, 0.80171799659729, 0.7934804558753967 ], "rho_per_layer": [ 0.9999992251396179, 0.9999991655349731, 0.9999990463256836, 0.9999980330467224, 0.9999971389770508, 0.9999936819076538, 0.9999881386756897, 0.9999800324440002, 0.9999632835388184, 0.999937891960144, 0.9999018311500549, 0.9998522996902466 ], "nudge_per_layer": [ -0.4491935074329376, -0.38857996463775635, -0.3270212411880493, -0.26439201831817627, -0.20410630106925964, -0.15283547341823578, -0.11351937055587769, -0.08454076200723648, -0.06403174996376038, -0.04978282377123833, -0.03986590355634689, -0.032995663583278656 ] }, "dfa": { "test_acc": 0.6485, "mean_bp_cosine": 0.05831230893575897, "mean_rho": 0.0407225100013117, "mean_nudge_001": -0.00021713114256272092, "mean_nudge_003": -0.0006509826756276501, "mean_nudge_01": -0.002172412224657213, "bp_cosine_per_layer": [ 0.003139778506010771, 0.07545783370733261, 0.07123857736587524, 0.04737270250916481, 0.07089244574308395, 0.05843178555369377, 0.10959646850824356, 0.04730183631181717, 0.05721074715256691, 0.06935717165470123, 0.04019004851579666, 0.04955831170082092 ], "rho_per_layer": [ -0.020620837807655334, 0.0424807034432888, 0.07906489074230194, -0.014874707907438278, 0.06078936904668808, 0.006801994517445564, 0.11016394942998886, 0.07577681541442871, 0.01940302737057209, 0.0600065141916275, 0.020656054839491844, 0.04902234673500061 ], "nudge_per_layer": [ 0.00014946935698390007, -0.004455030895769596, -0.0028790468350052834, -0.0015581330517306924, -0.0028997263871133327, -0.0020245155319571495, -0.003806713502854109, -0.0014502943959087133, -0.0017623959574848413, -0.002352846320718527, -0.001400630921125412, -0.001629082253202796 ] }, "state_bridge": { "test_acc": 0.6055, "mean_bp_cosine": 0.23132580270369849, "mean_rho": 0.21027085681756338, "mean_nudge_001": -0.002061755420678916, "mean_nudge_003": -0.006128282761589314, "mean_nudge_01": -0.019752548774704337, "bp_cosine_per_layer": [ 0.5538393259048462, 0.4287089705467224, 0.25277969241142273, 0.2402075231075287, 0.180599182844162, 0.15653815865516663, 0.1736883521080017, 0.18825414776802063, 0.1335570216178894, 0.13271775841712952, 0.15632027387619019, 0.17869922518730164 ], "rho_per_layer": [ 0.4586557149887085, 0.4002187252044678, 0.23149724304676056, 0.21227607131004333, 0.1842716634273529, 0.16792945563793182, 0.21390791237354279, 0.19318881630897522, 0.08300350606441498, 0.10752588510513306, 0.12254554033279419, 0.14822974801063538 ], "nudge_per_layer": [ -0.16128812730312347, -0.02071414887905121, -0.008381091989576817, -0.007560182362794876, -0.005691731348633766, -0.0046913521364331245, -0.005232630763202906, -0.0054755983874201775, -0.0038832086138427258, -0.004123975522816181, -0.004747895523905754, -0.005240642465651035 ], "state_pred_error_per_layer": [ 8762.341796875, 2348.0966796875, 1278.38525390625, 1081.769287109375, 948.214599609375, 985.5714111328125, 775.6072387695312, 713.37109375, 927.2667846679688, 1123.047119140625, 1035.65966796875, 864.7014770507812 ], "mean_state_pred_error": 1737.002700805664 }, "credit_bridge": { "test_acc": 0.4905, "mean_bp_cosine": 0.14494075424348316, "mean_rho": 0.18026241473853588, "mean_nudge_001": -0.0010125784683623351, "mean_nudge_003": -0.003034341447346378, "mean_nudge_01": -0.010073745266709011, "bp_cosine_per_layer": [ 0.0054779257625341415, 0.004163481295108795, -0.01823507994413376, 0.06908583641052246, 0.09582538902759552, 0.15172168612480164, 0.18814444541931152, 0.21233659982681274, 0.2397850900888443, 0.26321661472320557, 0.2640886902809143, 0.2636783719062805 ], "rho_per_layer": [ 0.05151631310582161, 0.03990761935710907, 0.02668120339512825, 0.08933386206626892, 0.12927217781543732, 0.1893344521522522, 0.20202970504760742, 0.27141058444976807, 0.2724490761756897, 0.2844354808330536, 0.2797800898551941, 0.32699841260910034 ], "nudge_per_layer": [ -0.0015225817915052176, -0.0009501088643446565, 0.00033305209944956005, -0.005592360161244869, -0.00755771528929472, -0.010863769799470901, -0.012987833470106125, -0.014405487105250359, -0.015937766060233116, -0.017140284180641174, -0.017137283459305763, -0.01712280511856079 ], "final_value_loss": 0.42473253202438355, "final_term_loss": 0.10016471980810165, "final_bridge_loss": 0.011516085395216941, "final_tgrad_loss": 0.3130517265796661 } }, "a1.0_L12_s456": { "bp": { "test_acc": 0.762, "mean_bp_cosine": 0.8058839539686838, "mean_rho": 0.999967540303866, "mean_nudge_001": -0.023895990879585344, "mean_nudge_003": -0.06970183023562034, "mean_nudge_01": -0.21104640668878952, "bp_cosine_per_layer": [ 0.8338037729263306, 0.8328856229782104, 0.8300410509109497, 0.823561429977417, 0.8153207302093506, 0.8093391060829163, 0.8022103309631348, 0.797042727470398, 0.7909072041511536, 0.7847579121589661, 0.7778322100639343, 0.7729053497314453 ], "rho_per_layer": [ 0.9999991059303284, 0.999998927116394, 0.9999983906745911, 0.9999978542327881, 0.999996542930603, 0.9999936819076538, 0.9999876022338867, 0.9999791383743286, 0.9999581575393677, 0.9999358057975769, 0.9998983144760132, 0.9998669624328613 ], "nudge_per_layer": [ -0.5344969034194946, -0.45930254459381104, -0.3820700943470001, -0.30436116456985474, -0.2331882268190384, -0.17455099523067474, -0.12991222739219666, -0.09728521853685379, -0.07412658631801605, -0.05795694515109062, -0.04662308096885681, -0.03868289291858673 ] }, "dfa": { "test_acc": 0.6705, "mean_bp_cosine": 0.07614415294180314, "mean_rho": 0.07346192157516877, "mean_nudge_001": -0.0004569464314651365, "mean_nudge_003": -0.0013685211645982538, "mean_nudge_01": -0.004532526101684198, "bp_cosine_per_layer": [ 0.04563574492931366, 0.07466404139995575, 0.07062304764986038, 0.07941032946109772, 0.08478259295225143, 0.1176375150680542, 0.051365386694669724, 0.07094398140907288, 0.06898803263902664, 0.07310269773006439, 0.0635182335972786, 0.11305823177099228 ], "rho_per_layer": [ 0.026268895715475082, 0.03733495995402336, 0.07775422930717468, 0.12396883964538574, 0.05687103420495987, 0.13703036308288574, -0.0030320845544338226, 0.08171480149030685, 0.06336408108472824, 0.09909991919994354, 0.06059148907661438, 0.12057653069496155 ], "nudge_per_layer": [ -0.015672076493501663, -0.0053095086477696896, -0.0031822575256228447, -0.003722358262166381, -0.0036155935376882553, -0.004913205746561289, -0.0017889357404783368, -0.002959918463602662, -0.003063073381781578, -0.003171245101839304, -0.0024787387810647488, -0.004513401538133621 ] }, "state_bridge": { "test_acc": 0.6015, "mean_bp_cosine": 0.1852269663165013, "mean_rho": 0.18596469921370348, "mean_nudge_001": -0.001999816272776419, "mean_nudge_003": -0.005956100358162075, "mean_nudge_01": -0.01933680862809221, "bp_cosine_per_layer": [ 0.5561584234237671, 0.39794063568115234, 0.20452159643173218, 0.14641982316970825, 0.11315252631902695, 0.12314423173666, 0.11100500077009201, 0.11256138980388641, 0.12550771236419678, 0.11898934841156006, 0.11000233888626099, 0.10332056879997253 ], "rho_per_layer": [ 0.4663659632205963, 0.38622361421585083, 0.20298895239830017, 0.1521472930908203, 0.10572830587625504, 0.14532539248466492, 0.1235864907503128, 0.12295880913734436, 0.1263602077960968, 0.1123446375131607, 0.14707674086093903, 0.1404699832201004 ], "nudge_per_layer": [ -0.16852372884750366, -0.020911922678351402, -0.007518916856497526, -0.005010381806641817, -0.0037449360825121403, -0.004001074004918337, -0.0037657374050468206, -0.0037060989998281, -0.00406123511493206, -0.003861584234982729, -0.0035726726055145264, -0.0033634149003773928 ], "state_pred_error_per_layer": [ 16351.89453125, 3139.08837890625, 1824.6005859375, 1391.949462890625, 1295.1900634765625, 1149.4605712890625, 981.4422607421875, 921.3729248046875, 954.6389770507812, 1032.970703125, 1065.58984375, 1272.3818359375 ], "mean_state_pred_error": 2615.048344930013 }, "credit_bridge": { "test_acc": 0.5585, "mean_bp_cosine": 0.3251700059821208, "mean_rho": 0.4226210874815782, "mean_nudge_001": -0.0023732566623948514, "mean_nudge_003": -0.007109481530884902, "mean_nudge_01": -0.02357900080581506, "bp_cosine_per_layer": [ 0.05994441360235214, 0.19127817451953888, 0.25874069333076477, 0.31905949115753174, 0.31442493200302124, 0.3214789927005768, 0.368133008480072, 0.38988351821899414, 0.4050426781177521, 0.416469931602478, 0.4253390431404114, 0.4322451949119568 ], "rho_per_layer": [ 0.1191493421792984, 0.31643322110176086, 0.4170604646205902, 0.42330044507980347, 0.4319503903388977, 0.42732322216033936, 0.46659332513809204, 0.487338662147522, 0.496512234210968, 0.4787842035293579, 0.5010949969291687, 0.5059125423431396 ], "nudge_per_layer": [ -0.015865826979279518, -0.018151750788092613, -0.021288521587848663, -0.02352731302380562, -0.023306310176849365, -0.02329857647418976, -0.025182321667671204, -0.025520823895931244, -0.026067661121487617, -0.026535863056778908, -0.027021951973438263, -0.02718108892440796 ], "final_value_loss": 0.3557061047077179, "final_term_loss": 0.06590610226392746, "final_bridge_loss": 0.010093027733266353, "final_tgrad_loss": 0.2797069767475128 } } }