{ "8": { "dfa": { "log": { "train_loss": [ 2.080832463722229, 2.048995506820679, 2.0368573234558105, 2.0319985338974, 2.0285650724029543, 2.0264119197082517, 2.0246589180755614, 2.0194223052597047, 2.023226359863281, 2.0190831980895996, 2.0185574532699584, 2.0163141152572632, 2.0120892976379396, 2.0131693548965455, 2.0107243951416014, 2.0126246429443357, 2.0099691207504273, 2.007364001235962, 2.009500975265503, 2.005168029327393, 2.0078421032333376, 2.005056644668579, 2.003016718902588, 2.002305891647339, 2.004946018447876, 2.0044735367202757, 2.0007521640014647, 2.0023752281188965, 2.002063782119751, 1.9980181777954102, 1.999666137161255, 2.0014348561096194, 1.9975938603973389, 1.997772049255371, 1.9968431232452393, 1.9957593490600587, 1.9965335370254516, 1.9956621952819824, 1.9998830209732055, 1.9981224115371705, 1.9981820908355712, 1.9963909757995606, 1.9932935749816894, 1.9949091632843017, 1.9948957403564453, 1.9962669714736938, 1.9956311498641968, 1.993200020980835, 1.9959270191192626, 1.9959917138671874, 1.99261492729187, 1.9928764194107056, 1.9886184701538085, 1.9945403707504272, 1.9950171528625489, 1.9918433042144776, 1.9921051956939697, 1.9939103940582275, 1.9926260836791991, 1.9888539260864257, 1.9864782820129394, 1.9928740283203126, 1.9936177950286864, 1.9908017050933837, 1.9916553665924073, 1.991268727722168, 1.9904707485198974, 1.990502048187256, 1.9922669997406006, 1.990579651412964, 1.989777621154785, 1.9894770226669312, 1.9896350140762329, 1.9901242208480836, 1.9889081002426148, 1.9884048733520507, 1.991174902420044, 1.9889308002090453, 1.9896702439880372, 1.9888165955352783, 1.9881730316925048, 1.9870906281280518, 1.9875952992248536, 1.987178568344116, 1.98706350440979, 1.987743454208374, 1.9882577419281007, 1.98791315574646, 1.9878161795806886, 1.9868540209197998, 1.9870264992523194, 1.9860009889984132, 1.9875505159759521, 1.9869904022979736, 1.9855010274887086, 1.9865515293884277, 1.9860705053710936, 1.9874219760131835, 1.9881664807128907, 1.9873311661148072 ], "train_acc": [ 0.22856, 0.2384, 0.24456, 0.24494, 0.24732, 0.25098, 0.24904, 0.2523, 0.25028, 0.2514, 0.25612, 0.25722, 0.25858, 0.25916, 0.26012, 0.26044, 0.26232, 0.26452, 0.26066, 0.26248, 0.2637, 0.26254, 0.26474, 0.26712, 0.2677, 0.26322, 0.26788, 0.26452, 0.26682, 0.26908, 0.2686, 0.26734, 0.27066, 0.26914, 0.26942, 0.27018, 0.26882, 0.2706, 0.26786, 0.26876, 0.27074, 0.27, 0.27148, 0.27242, 0.27346, 0.27144, 0.27246, 0.27478, 0.2713, 0.27238, 0.2732, 0.27598, 0.27788, 0.27462, 0.27216, 0.27474, 0.27372, 0.27488, 0.2751, 0.27356, 0.27528, 0.2739, 0.2744, 0.27592, 0.27666, 0.27542, 0.2749, 0.27564, 0.27486, 0.27662, 0.27766, 0.27678, 0.27682, 0.27618, 0.27544, 0.2767, 0.27538, 0.27816, 0.27422, 0.27562, 0.2772, 0.27786, 0.27924, 0.2775, 0.2789, 0.27834, 0.27724, 0.27724, 0.27694, 0.27716, 0.27862, 0.27786, 0.27676, 0.2771, 0.278, 0.27708, 0.27678, 0.27732, 0.27858, 0.27724 ], "test_acc": [ 0.25, 0.2553, 0.2484, 0.2586, 0.2481, 0.2619, 0.2688, 0.2577, 0.2807, 0.2806, 0.2671, 0.2735, 0.2663, 0.272, 0.2824, 0.2808, 0.2859, 0.2837, 0.282, 0.2895, 0.2788, 0.2787, 0.275, 0.2726, 0.283, 0.2729, 0.2866, 0.2909, 0.2806, 0.2892, 0.2818, 0.2915, 0.288, 0.2928, 0.2895, 0.3015, 0.2906, 0.2869, 0.2936, 0.2929, 0.2996, 0.2852, 0.295, 0.284, 0.2772, 0.302, 0.278, 0.2875, 0.2899, 0.3026, 0.2806, 0.2945, 0.2936, 0.298, 0.2945, 0.2987, 0.2926, 0.2891, 0.2897, 0.2928, 0.2947, 0.2867, 0.2913, 0.2912, 0.2907, 0.2955, 0.2948, 0.2955, 0.2934, 0.2874, 0.2917, 0.2904, 0.2907, 0.2893, 0.2937, 0.2975, 0.2948, 0.2938, 0.2921, 0.2903, 0.2921, 0.2932, 0.2969, 0.2915, 0.294, 0.2947, 0.2891, 0.2975, 0.2948, 0.2956, 0.2919, 0.2948, 0.2936, 0.2931, 0.293, 0.2934, 0.2934, 0.294, 0.2937, 0.2937 ] }, "diagnostics": { "bp_cosine": [ 0.37266281247138977, -0.00045336512266658247, -0.00016533058078493923, -0.00022915060981176794, -9.622798825148493e-05, -0.0005005986895412207 ], "perturbation_rho": [ -0.022135162726044655, 0.0, 0.0, 0.0, 0.0, 0.0 ], "nudging": { "0.001": [ -4.153698682785034e-07, 0.0, 0.0, 0.0, 0.0, 0.0 ], "0.003": [ -1.184176653623581e-06, 0.0, 0.0, 0.0, 0.0, 0.0 ], "0.01": [ -4.217028617858887e-06, 9.313225746154785e-10, 0.0, 0.0, -4.6566128730773926e-09, 0.0 ] }, "hidden_norms_per_layer": [ 50847.30078125, 848742848.0, 4660130304.0, 4956724224.0, 5474357248.0, 5779449344.0, 6897274368.0 ], "bp_grad_norms_per_layer": [ 3.144729134874069e-07, 2.3805521442987754e-10, 2.377963659316862e-10, 2.3782173452779887e-10, 2.382432029435222e-10, 2.381181640753738e-10, 2.3811083660341126e-10 ] }, "drift": { "embed.weight": 319.35155723550326, "embed.bias": 248.60053068216845, "blocks.0.ln.weight": 9.427908278003652, "blocks.0.w1.weight": 267.21029834357404, "blocks.0.w1.bias": 240.44244412845688, "blocks.0.w2.weight": 467.22266822391686, "blocks.1.ln.weight": 9.176869505684051, "blocks.1.w1.weight": 396.8672687385826, "blocks.1.w1.bias": 384.3213865917358, "blocks.1.w2.weight": 392.8080177067534, "blocks.2.ln.weight": 6.748169794145516, "blocks.2.w1.weight": 260.5475963025436, "blocks.2.w1.bias": 233.50388419740455, "blocks.2.w2.weight": 243.9744268156159, "blocks.3.ln.weight": 8.622995843562022, "blocks.3.w1.weight": 331.24292230819924, "blocks.3.w1.bias": 315.5877913098073, "blocks.3.w2.weight": 298.8313159517988, "blocks.4.ln.weight": 8.025346777953455, "blocks.4.w1.weight": 289.7124732778298, "blocks.4.w1.bias": 265.5485861970709, "blocks.4.w2.weight": 262.14080086870655, "blocks.5.ln.weight": 9.46176669645325, "blocks.5.w1.weight": 364.3981229248546, "blocks.5.w1.bias": 349.70853696261963, "blocks.5.w2.weight": 357.5708723169573, "out_ln.weight": 0.5682786122526117, "out_head.weight": 8.418164907479396, "out_head.bias": 0.662517650973914 } }, "fa": { "log": { "train_loss": [ 2.0414198583984375, 1.9522941863250733, 1.912906623916626, 1.8914309909057616, 1.8828500344848633, 1.8722443673706055, 1.867602448654175, 1.856888031578064, 1.8552497367095948, 1.843837871170044, 1.8456551111602784, 1.8392053551864624, 1.832319408721924, 1.8315165106582643, 1.8281906174468994, 1.8275285326766968, 1.823442045288086, 1.8223411087799073, 1.8216783404922485, 1.8156820178604125, 1.8176694821166992, 1.8109132452392578, 1.8122728036880493, 1.8086374733734132, 1.8135607794189452, 1.807762541885376, 1.8036900131225586, 1.8017769944000244, 1.8010728991699219, 1.7973168952178955, 1.792228649635315, 1.7974130847549439, 1.7889769022369384, 1.78288419921875, 1.7853075231552125, 1.7797698288726806, 1.7783561060333253, 1.7755371138763427, 1.7767201525115968, 1.7773720627593994, 1.7749587761688232, 1.7682355377960206, 1.7642274634552002, 1.7686031677627563, 1.763687038230896, 1.7665823318862914, 1.7619021569824218, 1.7553145666885377, 1.757571950340271, 1.7612224188995362, 1.7522613665390014, 1.7504500290298461, 1.7475368619537353, 1.7495313736343383, 1.7496113208389281, 1.7413074100494386, 1.7437634180450439, 1.7434646717071534, 1.746056427268982, 1.740490657081604, 1.7333746536254884, 1.7381776495742798, 1.73842948387146, 1.7355398804092408, 1.7360767101287842, 1.7332445461273194, 1.732193847579956, 1.730364631652832, 1.732538067970276, 1.7276911224746705, 1.7270672924804686, 1.7267360440826416, 1.725644203414917, 1.7283870937347412, 1.72088895652771, 1.7216768536376954, 1.723486647377014, 1.7211247494125366, 1.7208690827178954, 1.7219977252578735, 1.7204493698120118, 1.7233893868255614, 1.7198170611190795, 1.7190866617202758, 1.7203575790405274, 1.7177474556350707, 1.7174966396713256, 1.7169412566375732, 1.714385998878479, 1.7148360108184815, 1.7158790673446656, 1.7134223749542237, 1.714623501663208, 1.7133040780639648, 1.715691188583374, 1.7187325539398193, 1.7134061930084228, 1.713903324661255, 1.7177725345230102, 1.7178887582397462 ], "train_acc": [ 0.24766, 0.28764, 0.30724, 0.31554, 0.31774, 0.32862, 0.32412, 0.32998, 0.33022, 0.33776, 0.33938, 0.33914, 0.3422, 0.33992, 0.34394, 0.34276, 0.3413, 0.34586, 0.347, 0.34834, 0.34908, 0.35144, 0.35026, 0.3513, 0.34898, 0.35394, 0.35424, 0.35282, 0.3581, 0.35672, 0.35968, 0.35698, 0.36002, 0.36412, 0.36052, 0.36534, 0.36372, 0.36502, 0.36444, 0.36414, 0.36688, 0.36802, 0.37136, 0.3681, 0.3721, 0.36958, 0.37192, 0.37408, 0.37342, 0.37124, 0.37412, 0.37666, 0.37718, 0.3741, 0.37748, 0.37834, 0.3806, 0.37612, 0.3775, 0.37774, 0.38072, 0.38068, 0.38082, 0.37998, 0.382, 0.38172, 0.38358, 0.3851, 0.38346, 0.38376, 0.38542, 0.38474, 0.38684, 0.38692, 0.3899, 0.38944, 0.38642, 0.38598, 0.38606, 0.38656, 0.38402, 0.38854, 0.38994, 0.38908, 0.3902, 0.38896, 0.39004, 0.39054, 0.3903, 0.38856, 0.38994, 0.38868, 0.38934, 0.39004, 0.39128, 0.39024, 0.39048, 0.3904, 0.38948, 0.39032 ], "test_acc": [ 0.2817, 0.3086, 0.321, 0.3426, 0.3496, 0.3478, 0.3546, 0.3478, 0.3568, 0.3627, 0.3663, 0.3752, 0.3646, 0.3581, 0.3779, 0.3748, 0.3759, 0.3762, 0.3858, 0.3739, 0.3779, 0.3667, 0.3721, 0.3827, 0.369, 0.3761, 0.3719, 0.3794, 0.38, 0.3844, 0.3733, 0.3869, 0.3885, 0.3911, 0.387, 0.3906, 0.3904, 0.395, 0.3849, 0.3987, 0.3901, 0.4009, 0.3893, 0.385, 0.3918, 0.3995, 0.392, 0.3934, 0.3954, 0.4044, 0.3976, 0.3997, 0.3967, 0.402, 0.404, 0.3987, 0.3997, 0.3959, 0.3955, 0.3962, 0.4028, 0.4067, 0.3984, 0.4075, 0.4072, 0.4011, 0.4028, 0.4082, 0.4069, 0.4041, 0.4058, 0.4022, 0.4061, 0.4029, 0.4059, 0.4071, 0.4086, 0.4069, 0.4058, 0.4081, 0.4083, 0.4114, 0.4078, 0.4083, 0.4074, 0.4097, 0.4064, 0.4087, 0.4057, 0.4068, 0.4067, 0.407, 0.4072, 0.4089, 0.4094, 0.4092, 0.4088, 0.4081, 0.4078, 0.408 ] }, "diagnostics": { "bp_cosine": [ 0.016317440196871758, 0.08700302243232727, -0.07087832689285278, -0.04807429015636444, -0.011868046596646309, 0.9980045557022095 ], "perturbation_rho": [ 0.021565234288573265, 0.02034841850399971, 0.026769006624817848, -0.05038269981741905, 0.007762039080262184, 0.04057261347770691 ], "nudging": { "0.001": [ -1.8766731955111027e-06, -6.295740604400635e-07, 1.050066202878952e-07, 5.390029400587082e-08, -8.824281394481659e-08, -1.8120626918971539e-06 ], "0.003": [ -6.020389264449477e-06, -1.8251012079417706e-06, 4.4528860598802567e-07, 2.2060703486204147e-07, 2.1478626877069473e-08, -6.290327291935682e-06 ], "0.01": [ -2.0035397028550506e-05, -6.116693839430809e-06, 1.5703844837844372e-06, 9.065261110663414e-07, 1.710723154246807e-07, -2.1861022105440497e-05 ] }, "hidden_norms_per_layer": [ 7090.23486328125, 97328.1015625, 1195400.0, 1509493.75, 1603199.125, 1626782.875, 733314.5 ], "bp_grad_norms_per_layer": [ 2.7028392651118338e-05, 1.845057795435423e-06, 8.682639531798486e-07, 8.779788345236739e-07, 8.714667956155608e-07, 8.789162393441075e-07, 8.642545594739204e-07 ] }, "drift": { "embed.weight": 49.97107099246965, "embed.bias": 16.073336112282973, "blocks.0.ln.weight": 1.1864794930711844, "blocks.0.w1.weight": 16.772962175548443, "blocks.0.w1.bias": 11.717179109423967, "blocks.0.w2.weight": 54.31408520827723, "blocks.1.ln.weight": 1.1429608481901679, "blocks.1.w1.weight": 25.1727265395903, "blocks.1.w1.bias": 23.582538699977512, "blocks.1.w2.weight": 44.89348831927897, "blocks.2.ln.weight": 0.6307651937494874, "blocks.2.w1.weight": 20.089313457148293, "blocks.2.w1.bias": 20.43961041250799, "blocks.2.w2.weight": 36.04199442331163, "blocks.3.ln.weight": 0.5625176858716091, "blocks.3.w1.weight": 17.29253500275298, "blocks.3.w1.bias": 18.353681256446446, "blocks.3.w2.weight": 45.166175379509205, "blocks.4.ln.weight": 0.5780873641807247, "blocks.4.w1.weight": 15.563905559896059, "blocks.4.w1.bias": 13.46488029874075, "blocks.4.w2.weight": 59.00645861860142, "blocks.5.ln.weight": 0.7521925423173332, "blocks.5.w1.weight": 19.685866374226443, "blocks.5.w1.bias": 20.12611595031584, "blocks.5.w2.weight": 41.84701231716978, "out_ln.weight": 0.3374467885491731, "out_head.weight": 5.972151190658582, "out_head.bias": 0.6096349208966545 } } }, "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": [ 8 ], "gpu": 0, "output_dir": "results/fa_dfa_d512_L6_seed8", "methods": [ "fa", "dfa" ], "random_targets": false, "penalty_lam": 0.0, "num_classes": 10 } }