{ "2": { "dfa": { "log": { "train_loss": [ 2.0704764087677003, 2.049675481185913, 2.0393172933197024, 2.0385254068756105, 2.0333104403305056, 2.037317512435913, 2.035111841583252, 2.033784921951294, 2.0276396758270265, 2.0287634895324707, 2.0255478504943847, 2.0249928087997437, 2.0251559844207763, 2.026155082015991, 2.0227601895141603, 2.0238794365692137, 2.0220592249298095, 2.0211007302474977, 2.0220025036621094, 2.0187504733657837, 2.022589133682251, 2.0209386083221434, 2.022369726409912, 2.019175484046936, 2.0184574550628662, 2.020228290786743, 2.0170737294769285, 2.0200500258636476, 2.0192940770721437, 2.017934571380615, 2.018111932220459, 2.019744345703125, 2.0181864041137696, 2.018692120895386, 2.019279464416504, 2.020759454421997, 2.0175593240356444, 2.0189450312805177, 2.018363641014099, 2.0204193601989746, 2.018774921836853, 2.020020353240967, 2.020256604042053, 2.018686157913208, 2.017777068939209, 2.01642233127594, 2.0193701077270507, 2.0175296588897704, 2.0180459953308105, 2.0163426136016844, 2.0169748413085937, 2.017522889404297, 2.0171759409332277, 2.0182337244415285, 2.0202614056396486, 2.0193526766967773, 2.0176297589874266, 2.0176475778961183, 2.017951414642334, 2.0159765984725952, 2.0174861521911622, 2.0162583628082276, 2.0189596444702147, 2.017551812057495, 2.0174899821472168, 2.017485563354492, 2.0167667126464845, 2.017466169166565, 2.017418462524414, 2.0175614359283447, 2.016368521194458, 2.017146753158569, 2.016855639877319, 2.016222254562378, 2.016020330963135, 2.0157443789672853, 2.014490191497803, 2.0179072814559937, 2.0176085803985595, 2.016617900123596, 2.016193874168396, 2.0163377814483643, 2.017562115097046, 2.017051735992432, 2.015485999984741, 2.0135854721450808, 2.0146356997299195, 2.0149522621154787, 2.0146646865844726, 2.0153391693878175, 2.0160440145874023, 2.0161946767425536, 2.014449896469116, 2.015591969642639, 2.01485006980896, 2.01346685836792, 2.0139751796722414, 2.0144011250305174, 2.014900064163208, 2.016459233779907 ], "train_acc": [ 0.23802, 0.24392, 0.24906, 0.25122, 0.25396, 0.25104, 0.2552, 0.25244, 0.2577, 0.25632, 0.25834, 0.2557, 0.25944, 0.25876, 0.26044, 0.25964, 0.2593, 0.26036, 0.26106, 0.26136, 0.26018, 0.25898, 0.25858, 0.2613, 0.26334, 0.26364, 0.26186, 0.26224, 0.26166, 0.26282, 0.26364, 0.26358, 0.2645, 0.26294, 0.26592, 0.26282, 0.26398, 0.26414, 0.26538, 0.26506, 0.26578, 0.26148, 0.26312, 0.26586, 0.26478, 0.26574, 0.26416, 0.26488, 0.26366, 0.26614, 0.2656, 0.26508, 0.2675, 0.26694, 0.26532, 0.26614, 0.26596, 0.26538, 0.2652, 0.26606, 0.26698, 0.26806, 0.26674, 0.26714, 0.26704, 0.26768, 0.26714, 0.2651, 0.26544, 0.26656, 0.2676, 0.26678, 0.26898, 0.26788, 0.26736, 0.2677, 0.26732, 0.26766, 0.26812, 0.2675, 0.26916, 0.26876, 0.26754, 0.26748, 0.26802, 0.27004, 0.27052, 0.2671, 0.27224, 0.26654, 0.26818, 0.26902, 0.2691, 0.26962, 0.2702, 0.26948, 0.2712, 0.27026, 0.26876, 0.26952 ], "test_acc": [ 0.2537, 0.2494, 0.2765, 0.2473, 0.2851, 0.2759, 0.2702, 0.2833, 0.2841, 0.2663, 0.2858, 0.2432, 0.2894, 0.2933, 0.2938, 0.2901, 0.2777, 0.2769, 0.2907, 0.2833, 0.2973, 0.2692, 0.277, 0.2841, 0.2917, 0.2694, 0.2929, 0.2737, 0.2955, 0.2877, 0.2891, 0.302, 0.272, 0.2931, 0.2882, 0.2806, 0.2849, 0.2989, 0.2988, 0.3036, 0.2869, 0.2951, 0.3014, 0.289, 0.2834, 0.2896, 0.2887, 0.2932, 0.2855, 0.2882, 0.2846, 0.3018, 0.2981, 0.3019, 0.2846, 0.2968, 0.2915, 0.2935, 0.2912, 0.2978, 0.2989, 0.2953, 0.2999, 0.292, 0.2923, 0.2856, 0.2956, 0.2993, 0.2906, 0.2931, 0.2925, 0.2948, 0.2968, 0.295, 0.2965, 0.3019, 0.2889, 0.2941, 0.2958, 0.2959, 0.2955, 0.2926, 0.2921, 0.2964, 0.2981, 0.2912, 0.2982, 0.2995, 0.2958, 0.2963, 0.2965, 0.2963, 0.2947, 0.2957, 0.2953, 0.2947, 0.2953, 0.2951, 0.2949, 0.295 ] }, "diagnostics": { "bp_cosine": [ 0.3428037762641907, -3.763916902244091e-05, -0.0007522967061959207, -0.00019637049990706146, -0.0005234384443610907, 0.00033836261718533933 ], "perturbation_rho": [ -0.03871288150548935, 0.0, 0.0, 0.0, 0.0, 0.0 ], "nudging": { "0.001": [ -3.3015385270118713e-07, 0.0, 0.0, 0.0, 0.0, 0.0 ], "0.003": [ -8.167698979377747e-07, 0.0, 0.0, 0.0, 1.862645149230957e-09, 0.0 ], "0.01": [ -2.7455389499664307e-06, 0.0, 0.0, 1.862645149230957e-09, 1.862645149230957e-09, 0.0 ] }, "hidden_norms_per_layer": [ 53848.7734375, 2442806016.0, 5245538816.0, 6828209664.0, 7018600960.0, 9658986496.0, 9768150016.0 ], "bp_grad_norms_per_layer": [ 2.0601439132406085e-07, 2.350288574870518e-10, 2.3180254937749112e-10, 2.322748937633179e-10, 2.3222145040247e-10, 2.3218917066802902e-10, 2.321651759729093e-10 ] }, "drift": { "embed.weight": 347.726745764923, "embed.bias": 320.686611820448, "blocks.0.ln.weight": 9.997068770929598, "blocks.0.w1.weight": 332.0183119830964, "blocks.0.w1.bias": 361.6276454599223, "blocks.0.w2.weight": 491.26444464340926, "blocks.1.ln.weight": 9.760206175642947, "blocks.1.w1.weight": 399.42793179083463, "blocks.1.w1.bias": 382.1218468905457, "blocks.1.w2.weight": 398.4896403314863, "blocks.2.ln.weight": 9.863790992763134, "blocks.2.w1.weight": 403.4218387582533, "blocks.2.w1.bias": 369.67326277166103, "blocks.2.w2.weight": 388.1030551430935, "blocks.3.ln.weight": 7.7321395772148405, "blocks.3.w1.weight": 284.159080804305, "blocks.3.w1.bias": 261.08860693920894, "blocks.3.w2.weight": 276.021700238577, "blocks.4.ln.weight": 10.766653379169998, "blocks.4.w1.weight": 441.4219333141434, "blocks.4.w1.bias": 408.4716885612093, "blocks.4.w2.weight": 432.48056855561003, "blocks.5.ln.weight": 7.252784916283292, "blocks.5.w1.weight": 280.38847615534195, "blocks.5.w1.bias": 255.99327290804064, "blocks.5.w2.weight": 255.63025547094998, "out_ln.weight": 0.644730004333539, "out_head.weight": 9.313913439210426, "out_head.bias": 0.978185244031994 } }, "fa": { "log": { "train_loss": [ 2.0382984645843507, 1.9496157778167724, 1.9119072325897217, 1.8929361431121827, 1.8775380432128905, 1.8741492847442627, 1.8671779189300537, 1.860015816307068, 1.8485039702987671, 1.844084996986389, 1.8363114721679688, 1.8330332998275758, 1.8273526552581787, 1.825867492904663, 1.821722242088318, 1.8222067623519898, 1.8212668908691407, 1.8164310580825807, 1.8167414797592163, 1.8158153560638428, 1.8170397219848633, 1.8139575901031495, 1.814438596458435, 1.807250874671936, 1.806480390663147, 1.805527213973999, 1.7998141473770142, 1.8049808203125, 1.8055741637420655, 1.796914314918518, 1.8009305011749268, 1.797931548538208, 1.7981645835113524, 1.793643822402954, 1.7937120150375365, 1.7967921892929077, 1.7892387997436523, 1.785188671875, 1.786605785293579, 1.787917324485779, 1.7853582260894776, 1.7880721422576904, 1.7864411516571046, 1.7777207873916625, 1.7769269942855834, 1.777174091796875, 1.7795673489761352, 1.773414126586914, 1.7774351581192016, 1.7697969200897217, 1.771698450050354, 1.768428716278076, 1.7676336987304688, 1.765194714126587, 1.7675122838974, 1.7682551845932006, 1.7667323706436158, 1.7661600426864623, 1.7620434980010986, 1.7610514661026, 1.7625030724716186, 1.7606126858520508, 1.760137271118164, 1.7585180844879151, 1.7585363976287842, 1.7592174551010131, 1.755930234375, 1.7548912310791016, 1.7563516518783568, 1.7501784167099, 1.7541064535522461, 1.7512517785263062, 1.7541993662261963, 1.7516761654663087, 1.7516398949813843, 1.749719626121521, 1.7497738579559325, 1.752181930885315, 1.7519074490356445, 1.7518238845443725, 1.7466175811004638, 1.748845227355957, 1.746575064430237, 1.7496823914337158, 1.7468877982330322, 1.7439219277191162, 1.7428745371246337, 1.746016036453247, 1.742904917678833, 1.7454003299713134, 1.7449493420028686, 1.744745844039917, 1.7438948041152955, 1.7453510579681397, 1.7438724200439453, 1.7425698751068115, 1.742772998123169, 1.7421956006622314, 1.7445682699966432, 1.7442712462615966 ], "train_acc": [ 0.25024, 0.28724, 0.30528, 0.31264, 0.32216, 0.32208, 0.32792, 0.33064, 0.3365, 0.3396, 0.34, 0.34242, 0.34468, 0.34422, 0.34752, 0.34912, 0.34524, 0.34978, 0.34856, 0.3501, 0.34806, 0.35048, 0.35214, 0.35416, 0.35338, 0.35482, 0.35432, 0.35244, 0.35388, 0.35836, 0.35712, 0.3568, 0.3591, 0.36118, 0.36208, 0.35824, 0.36296, 0.3625, 0.3616, 0.3628, 0.36246, 0.36298, 0.3605, 0.36632, 0.36574, 0.36678, 0.36478, 0.3649, 0.3662, 0.36734, 0.3665, 0.36722, 0.36714, 0.3702, 0.36908, 0.37058, 0.36998, 0.36858, 0.37244, 0.37104, 0.3722, 0.37344, 0.3726, 0.37172, 0.3741, 0.37406, 0.3731, 0.37294, 0.37284, 0.37628, 0.37682, 0.37742, 0.37544, 0.37656, 0.37632, 0.37814, 0.37808, 0.37642, 0.37748, 0.37462, 0.37968, 0.37766, 0.37704, 0.37846, 0.37784, 0.379, 0.37904, 0.37854, 0.38202, 0.37828, 0.37712, 0.37962, 0.38004, 0.37898, 0.3783, 0.37898, 0.38082, 0.38006, 0.37922, 0.37846 ], "test_acc": [ 0.3027, 0.306, 0.3365, 0.3382, 0.3465, 0.3597, 0.3579, 0.3625, 0.3757, 0.3702, 0.3663, 0.3612, 0.374, 0.378, 0.3778, 0.3778, 0.3807, 0.3691, 0.3833, 0.3714, 0.3788, 0.3629, 0.3794, 0.382, 0.3843, 0.3732, 0.3818, 0.373, 0.3819, 0.3776, 0.3882, 0.3926, 0.385, 0.3853, 0.3879, 0.3927, 0.3894, 0.3945, 0.3903, 0.3924, 0.3896, 0.3941, 0.3919, 0.3888, 0.3863, 0.3927, 0.3922, 0.3903, 0.3939, 0.3828, 0.3963, 0.3935, 0.3958, 0.3971, 0.4048, 0.4003, 0.3934, 0.3945, 0.395, 0.3979, 0.3908, 0.3969, 0.4012, 0.3998, 0.3987, 0.3965, 0.4025, 0.3984, 0.4038, 0.3982, 0.4015, 0.4027, 0.3997, 0.4022, 0.4045, 0.4031, 0.4028, 0.4041, 0.3998, 0.4012, 0.3998, 0.3979, 0.4026, 0.401, 0.4006, 0.4048, 0.4018, 0.4009, 0.4025, 0.4024, 0.4009, 0.4036, 0.4025, 0.4022, 0.4026, 0.401, 0.4018, 0.401, 0.4015, 0.4016 ] }, "diagnostics": { "bp_cosine": [ 0.02052094228565693, 0.07671980559825897, -0.10412630438804626, -0.03588568791747093, -0.07592153549194336, 0.9841817617416382 ], "perturbation_rho": [ -0.0499441958963871, 0.03510870411992073, 0.012152664363384247, -0.01593763567507267, -0.030490349978208542, 0.01056537963449955 ], "nudging": { "0.001": [ -1.460895873606205e-06, -1.8533319234848022e-07, 8.102506399154663e-08, 3.9814040064811707e-08, 7.660128176212311e-08, -1.0418007150292397e-06 ], "0.003": [ -4.51505184173584e-06, -7.73230567574501e-07, 4.012836143374443e-07, 1.9150320440530777e-07, 2.3317988961935043e-07, -3.8853613659739494e-06 ], "0.01": [ -1.504761166870594e-05, -2.689310349524021e-06, 1.5775440260767937e-06, 6.139744073152542e-07, 9.952345862984657e-07, -1.3828510418534279e-05 ] }, "hidden_norms_per_layer": [ 7937.806640625, 157942.125, 845318.25, 1734418.625, 1908211.875, 2166763.75, 1424229.125 ], "bp_grad_norms_per_layer": [ 2.3930177121656016e-05, 1.1552400565051357e-06, 6.149262503640784e-07, 6.002979375807627e-07, 6.000758503432735e-07, 5.996230356686283e-07, 5.751607545789739e-07 ] }, "drift": { "embed.weight": 54.05447299340841, "embed.bias": 16.031038115248215, "blocks.0.ln.weight": 1.2416108593211508, "blocks.0.w1.weight": 17.922589218376636, "blocks.0.w1.bias": 14.956067338203475, "blocks.0.w2.weight": 64.67351730215137, "blocks.1.ln.weight": 1.1286859632363149, "blocks.1.w1.weight": 23.71804846947617, "blocks.1.w1.bias": 17.43068124645083, "blocks.1.w2.weight": 41.38902946126262, "blocks.2.ln.weight": 0.7406672335136909, "blocks.2.w1.weight": 24.88918465627309, "blocks.2.w1.bias": 25.668393301356875, "blocks.2.w2.weight": 26.33124026065157, "blocks.3.ln.weight": 0.5408212886770045, "blocks.3.w1.weight": 19.746562187924944, "blocks.3.w1.bias": 21.605426127504696, "blocks.3.w2.weight": 26.408093498021923, "blocks.4.ln.weight": 0.7373842468395975, "blocks.4.w1.weight": 21.153183487980545, "blocks.4.w1.bias": 21.145538252519305, "blocks.4.w2.weight": 45.59224517841596, "blocks.5.ln.weight": 0.8316077429831186, "blocks.5.w1.weight": 20.11472175424346, "blocks.5.w1.bias": 17.387776840886495, "blocks.5.w2.weight": 81.52952222436008, "out_ln.weight": 0.38061390763416836, "out_head.weight": 7.016437743402715, "out_head.bias": 0.6179754221306049 } } }, "config": { "dataset": "cifar10", "d_hidden": 512, "num_blocks": 6, "batch_size": 128, "epochs": 100, "lr": 0.001, "lr_fb": 0.001, "wd": 0.01, "lam": 0.1, "K": 4, "sigma_bridge": 0.05, "ema_momentum": 0.995, "term_grad_weight": 1.0, "seeds": [ 2 ], "gpu": 0, "output_dir": "results/fa_dfa_d512_L6_seed2", "methods": [ "fa", "dfa" ], "random_targets": false, "penalty_lam": 0.0, "num_classes": 10 } }