{"config":{"outdir":"/Data/janis.aiad/nn--landscape/experiments/brainstorm/resnetl3/results/hierarchical3_muon_bbp_latest","seed":314159,"k":3,"p":3,"q":3,"nq":7,"steps":1200,"diag_every":40,"repeats":3,"train_n":1024,"init_scale":0.12,"readout_scale":0.05,"gd_lr":0.02,"muon_lr":0.004,"vector_lr":0.02,"spectral_eps":1e-06,"tie_u_v":false,"d":32,"alpha":5.0,"bbp_checkpoints":4,"block_cap":100.0,"edge_scan_points":56,"mde_iterations":240,"mde_damping":0.35,"mde_tolerance":3e-09,"density_eta":0.05,"density_points":81},"optimizer_semantics":{"gd":"ordinary gradient descent on every parameter","muon_matrix_update":"U diag((sigma/mean(sigma))^a) V^T, normalized to unit RMS","muon_vector_update":"ordinary gradient descent","a_1_7_status":"phase-retrieval transfer control, not an L3 theorem","a_1_3_status":"normalized phase-retrieval branch control, not an L3 theorem"},"training":{"gd":{"final_test_risk_mean":0.5815533632534134,"final_test_risk_std":0.47273587250739846,"final_test_risk_sem":0.2729341832477391,"repeats":3,"final_test_risk_median":0.7568927456127152,"final_test_risk_geometric_mean":0.32050922678639315,"final_test_risk_by_repeat":[0.7568927456127152,0.941568026167945,0.04619931797958035],"total_backtracking_rejects":0,"paired_risk_ratio_vs_gd":[1.0,1.0,1.0],"wins_vs_gd":0,"order_2_half_error_step":120,"order_3_half_error_step":160,"order_4_half_error_step":null,"V_coord_1_align80_step":null,"V_coord_2_align80_step":null,"V_coord_3_align80_step":40},"muon_a0":{"final_test_risk_mean":0.14629840517360146,"final_test_risk_std":0.10817412707548245,"final_test_risk_sem":0.06245436138638258,"repeats":3,"final_test_risk_median":0.08711987543917674,"final_test_risk_geometric_mean":0.12395530031442181,"final_test_risk_by_repeat":[0.27115082127147844,0.08062451881014912,0.08711987543917674],"total_backtracking_rejects":50,"paired_risk_ratio_vs_gd":[0.3582420664528606,0.08562792763713531,1.8857394275318713],"wins_vs_gd":2,"order_2_half_error_step":120,"order_3_half_error_step":120,"order_4_half_error_step":680,"V_coord_1_align80_step":80,"V_coord_2_align80_step":200,"V_coord_3_align80_step":40},"muon_a1_7_transfer":{"final_test_risk_mean":0.36547241965187177,"final_test_risk_std":0.2798257485340855,"final_test_risk_sem":0.1615574712423428,"repeats":3,"final_test_risk_median":0.4113791732431087,"final_test_risk_geometric_mean":0.25561141060753595,"final_test_risk_by_repeat":[0.4113791732431087,0.06553190051261257,0.6195061851998942],"total_backtracking_rejects":49,"paired_risk_ratio_vs_gd":[0.5435105246121649,0.06959868930481695,13.409422742424681],"wins_vs_gd":2,"order_2_half_error_step":80,"order_3_half_error_step":120,"order_4_half_error_step":1160,"V_coord_1_align80_step":160,"V_coord_2_align80_step":240,"V_coord_3_align80_step":40},"muon_a1_3_normalized":{"final_test_risk_mean":0.14571379488316874,"final_test_risk_std":0.13278135410058817,"final_test_risk_sem":0.0766613505333376,"repeats":3,"final_test_risk_median":0.08887456587831988,"final_test_risk_geometric_mean":0.11033785141178372,"final_test_risk_by_repeat":[0.08887456587831988,0.050813238564325874,0.29745358020686047],"total_backtracking_rejects":49,"paired_risk_ratio_vs_gd":[0.1174202902504696,0.05396661436256381,6.438484229103383],"wins_vs_gd":2,"order_2_half_error_step":80,"order_3_half_error_step":80,"order_4_half_error_step":800,"V_coord_1_align80_step":800,"V_coord_2_align80_step":120,"V_coord_3_align80_step":80}},"bbp":[{"optimizer":"gd","step":0,"population_risk":2.573494854014269,"n":145,"d":32,"block_size":6,"alpha_perp":5.0,"block_truncation_fraction":0.0,"truncation_operator_norm":1.2282870108272603e-15,"edge_left_variational":-1.1528043636082597,"edge_right_variational":1.11215336660237,"edge_left_empirical_bulk":-0.9535561908157966,"edge_right_empirical_bulk":0.886263076777374,"edge_left_abs_error":0.1992481727924631,"edge_right_abs_error":0.22589028982499593,"left_edge_derivative":0.001805850118737132,"right_edge_derivative":7.893572791850545e-06,"edge_solver_error":"","outlier_count":2,"informative_outlier_count":2,"max_outlier_teacher_overlap":0.3602374524347155,"max_teacher_overlap":0.5725396562449626,"min_full_eigenvalue":-1.2444907313723694,"max_full_eigenvalue":1.1711697999517232,"raw_edge_left_empirical_bulk":-0.9535561908157975,"raw_edge_right_empirical_bulk":0.886263076777374,"raw_outlier_count":4,"raw_informative_outlier_count":4,"raw_max_outlier_teacher_overlap":0.5193105547481304,"raw_min_full_eigenvalue":-1.244490731372371,"raw_max_full_eigenvalue":1.171169799951723},{"optimizer":"gd","step":400,"population_risk":0.7634824129683356,"n":145,"d":32,"block_size":6,"alpha_perp":5.0,"block_truncation_fraction":0.006896551724137931,"truncation_operator_norm":33.191965986757246,"edge_left_variational":-10.41615543847901,"edge_right_variational":35.06837735318439,"edge_left_empirical_bulk":-9.350007288024276,"edge_right_empirical_bulk":38.45060221020455,"edge_left_abs_error":1.0661481504547332,"edge_right_abs_error":3.3822248570201623,"left_edge_derivative":0.42311378097746727,"right_edge_derivative":1.187139596613434,"edge_solver_error":"","outlier_count":2,"informative_outlier_count":1,"max_outlier_teacher_overlap":0.29187696789184486,"max_teacher_overlap":0.5243264107983826,"min_full_eigenvalue":-12.472923817959776,"max_full_eigenvalue":50.40795783524601,"raw_edge_left_empirical_bulk":-9.336749135300865,"raw_edge_right_empirical_bulk":51.565485761942426,"raw_outlier_count":2,"raw_informative_outlier_count":1,"raw_max_outlier_teacher_overlap":0.33726268219010974,"raw_min_full_eigenvalue":-12.453448468294923,"raw_max_full_eigenvalue":73.19437580995539},{"optimizer":"gd","step":800,"population_risk":0.7541858580018752,"n":145,"d":32,"block_size":6,"alpha_perp":5.0,"block_truncation_fraction":0.0,"truncation_operator_norm":1.1682395085190211e-14,"edge_left_variational":-2.3891394942628628,"edge_right_variational":27.66188074040038,"edge_left_empirical_bulk":-1.2914322550276134,"edge_right_empirical_bulk":14.054567273712355,"edge_left_abs_error":1.0977072392352494,"edge_right_abs_error":13.607313466688026,"left_edge_derivative":0.00015050522152952936,"right_edge_derivative":2.7814705624701217,"edge_solver_error":"","outlier_count":0,"informative_outlier_count":0,"max_outlier_teacher_overlap":0.0,"max_teacher_overlap":0.5117894344741313,"min_full_eigenvalue":-1.7610380230205454,"max_full_eigenvalue":24.039703274218656,"raw_edge_left_empirical_bulk":-1.2914322550276072,"raw_edge_right_empirical_bulk":14.054567273712355,"raw_outlier_count":4,"raw_informative_outlier_count":3,"raw_max_outlier_teacher_overlap":0.5117894344741304,"raw_min_full_eigenvalue":-1.761038023020547,"raw_max_full_eigenvalue":24.039703274218656},{"optimizer":"gd","step":1200,"population_risk":0.7568927456127152,"n":145,"d":32,"block_size":6,"alpha_perp":5.0,"block_truncation_fraction":0.013793103448275862,"truncation_operator_norm":16.515254575611966,"edge_left_variational":-4.205773711893249,"edge_right_variational":34.2480075184185,"edge_left_empirical_bulk":-2.238908954365014,"edge_right_empirical_bulk":22.50161814044922,"edge_left_abs_error":1.966864757528235,"edge_right_abs_error":11.746389377969276,"left_edge_derivative":0.0018507353510754053,"right_edge_derivative":3.013796729995608,"edge_solver_error":"","outlier_count":0,"informative_outlier_count":0,"max_outlier_teacher_overlap":0.0,"max_teacher_overlap":0.47883274848584706,"min_full_eigenvalue":-3.255525582989897,"max_full_eigenvalue":30.781168960971403,"raw_edge_left_empirical_bulk":-2.235243952321683,"raw_edge_right_empirical_bulk":32.69668649687731,"raw_outlier_count":2,"raw_informative_outlier_count":2,"raw_max_outlier_teacher_overlap":0.41070308242710113,"raw_min_full_eigenvalue":-3.25420335096927,"raw_max_full_eigenvalue":46.94596575245996},{"optimizer":"muon_a0","step":0,"population_risk":2.573494854014269,"n":145,"d":32,"block_size":6,"alpha_perp":5.0,"block_truncation_fraction":0.0,"truncation_operator_norm":1.2282870108272603e-15,"edge_left_variational":-1.1528043636082597,"edge_right_variational":1.11215336660237,"edge_left_empirical_bulk":-0.9535561908157966,"edge_right_empirical_bulk":0.886263076777374,"edge_left_abs_error":0.1992481727924631,"edge_right_abs_error":0.22589028982499593,"left_edge_derivative":0.001805850118737132,"right_edge_derivative":7.893572791850545e-06,"edge_solver_error":"","outlier_count":2,"informative_outlier_count":2,"max_outlier_teacher_overlap":0.3602374524347155,"max_teacher_overlap":0.5725396562449626,"min_full_eigenvalue":-1.2444907313723694,"max_full_eigenvalue":1.1711697999517232,"raw_edge_left_empirical_bulk":-0.9535561908157975,"raw_edge_right_empirical_bulk":0.886263076777374,"raw_outlier_count":4,"raw_informative_outlier_count":4,"raw_max_outlier_teacher_overlap":0.5193105547481304,"raw_min_full_eigenvalue":-1.244490731372371,"raw_max_full_eigenvalue":1.171169799951723},{"optimizer":"muon_a0","step":400,"population_risk":0.6476046058456904,"n":145,"d":32,"block_size":6,"alpha_perp":5.0,"block_truncation_fraction":0.034482758620689655,"truncation_operator_norm":167.04733012661987,"edge_left_variational":-9.641043465180923,"edge_right_variational":41.179837947787334,"edge_left_empirical_bulk":-5.7081801421320675,"edge_right_empirical_bulk":46.560849080061615,"edge_left_abs_error":3.9328633230488554,"edge_right_abs_error":5.381011132274281,"left_edge_derivative":0.13952917411609178,"right_edge_derivative":2.655973359740322,"edge_solver_error":"","outlier_count":2,"informative_outlier_count":1,"max_outlier_teacher_overlap":0.5388815087637489,"max_teacher_overlap":0.5388815087637489,"min_full_eigenvalue":-8.123753126687832,"max_full_eigenvalue":60.118000596182554,"raw_edge_left_empirical_bulk":-5.660261774701739,"raw_edge_right_empirical_bulk":175.57490987792949,"raw_outlier_count":1,"raw_informative_outlier_count":0,"raw_max_outlier_teacher_overlap":0.21384016592422683,"raw_min_full_eigenvalue":-7.9730945388078895,"raw_max_full_eigenvalue":216.49572476149345},{"optimizer":"muon_a0","step":800,"population_risk":0.28837309779533243,"n":145,"d":32,"block_size":6,"alpha_perp":5.0,"block_truncation_fraction":0.006896551724137931,"truncation_operator_norm":32.698262836142824,"edge_left_variational":-3.658979575525169,"edge_right_variational":35.63817635987969,"edge_left_empirical_bulk":-1.7800117911607096,"edge_right_empirical_bulk":28.406778672436644,"edge_left_abs_error":1.8789677843644592,"edge_right_abs_error":7.231397687443042,"left_edge_derivative":0.023377850629514654,"right_edge_derivative":1.7255514281687283,"edge_solver_error":"","outlier_count":1,"informative_outlier_count":1,"max_outlier_teacher_overlap":0.4351356415110766,"max_teacher_overlap":0.4351356415110766,"min_full_eigenvalue":-2.3436048209027804,"max_full_eigenvalue":36.41844184006008,"raw_edge_left_empirical_bulk":-1.7667978032552076,"raw_edge_right_empirical_bulk":49.55718014422597,"raw_outlier_count":1,"raw_informative_outlier_count":1,"raw_max_outlier_teacher_overlap":0.2894566173491217,"raw_min_full_eigenvalue":-2.3316930549982944,"raw_max_full_eigenvalue":63.53164365327493},{"optimizer":"muon_a0","step":1200,"population_risk":0.27115082127147844,"n":145,"d":32,"block_size":6,"alpha_perp":5.0,"block_truncation_fraction":0.027586206896551724,"truncation_operator_norm":54.48360881217696,"edge_left_variational":-22.258369048988172,"edge_right_variational":36.97132288633089,"edge_left_empirical_bulk":-24.298501086198506,"edge_right_empirical_bulk":34.870446106504986,"edge_left_abs_error":2.040132037210334,"edge_right_abs_error":2.100876779825903,"left_edge_derivative":0.23459846940022544,"right_edge_derivative":0.3939669259093838,"edge_solver_error":"","outlier_count":2,"informative_outlier_count":1,"max_outlier_teacher_overlap":0.3276228895010032,"max_teacher_overlap":0.3745476551941825,"min_full_eigenvalue":-30.527012073526098,"max_full_eigenvalue":42.64859846453434,"raw_edge_left_empirical_bulk":-24.190088790565465,"raw_edge_right_empirical_bulk":63.6613338924363,"raw_outlier_count":3,"raw_informative_outlier_count":2,"raw_max_outlier_teacher_overlap":0.3909758089908552,"raw_min_full_eigenvalue":-30.277741609112944,"raw_max_full_eigenvalue":87.64948721787209},{"optimizer":"muon_a1_7_transfer","step":0,"population_risk":2.573494854014269,"n":145,"d":32,"block_size":6,"alpha_perp":5.0,"block_truncation_fraction":0.0,"truncation_operator_norm":1.2282870108272603e-15,"edge_left_variational":-1.1528043636082597,"edge_right_variational":1.11215336660237,"edge_left_empirical_bulk":-0.9535561908157966,"edge_right_empirical_bulk":0.886263076777374,"edge_left_abs_error":0.1992481727924631,"edge_right_abs_error":0.22589028982499593,"left_edge_derivative":0.001805850118737132,"right_edge_derivative":7.893572791850545e-06,"edge_solver_error":"","outlier_count":2,"informative_outlier_count":2,"max_outlier_teacher_overlap":0.3602374524347155,"max_teacher_overlap":0.5725396562449626,"min_full_eigenvalue":-1.2444907313723694,"max_full_eigenvalue":1.1711697999517232,"raw_edge_left_empirical_bulk":-0.9535561908157975,"raw_edge_right_empirical_bulk":0.886263076777374,"raw_outlier_count":4,"raw_informative_outlier_count":4,"raw_max_outlier_teacher_overlap":0.5193105547481304,"raw_min_full_eigenvalue":-1.244490731372371,"raw_max_full_eigenvalue":1.171169799951723},{"optimizer":"muon_a1_7_transfer","step":400,"population_risk":0.6414389472957271,"n":145,"d":32,"block_size":6,"alpha_perp":5.0,"block_truncation_fraction":0.034482758620689655,"truncation_operator_norm":137.74735404165884,"edge_left_variational":-12.682555442068377,"edge_right_variational":42.94093353789053,"edge_left_empirical_bulk":-7.358004256236539,"edge_right_empirical_bulk":46.76415690365202,"edge_left_abs_error":5.324551185831838,"edge_right_abs_error":3.8232233657614927,"left_edge_derivative":0.9085408171697509,"right_edge_derivative":2.915579934496191,"edge_solver_error":"","outlier_count":2,"informative_outlier_count":1,"max_outlier_teacher_overlap":0.5409206386048074,"max_teacher_overlap":0.5409206386048074,"min_full_eigenvalue":-10.501495537219004,"max_full_eigenvalue":61.712686969311605,"raw_edge_left_empirical_bulk":-7.270989325311044,"raw_edge_right_empirical_bulk":160.51335011962988,"raw_outlier_count":2,"raw_informative_outlier_count":0,"raw_max_outlier_teacher_overlap":0.20391567444620462,"raw_min_full_eigenvalue":-10.21116000278947,"raw_max_full_eigenvalue":186.9272441430432},{"optimizer":"muon_a1_7_transfer","step":800,"population_risk":0.5237564910403706,"n":145,"d":32,"block_size":6,"alpha_perp":5.0,"block_truncation_fraction":0.027586206896551724,"truncation_operator_norm":79.98189468660634,"edge_left_variational":-5.282080992920072,"edge_right_variational":43.313050648947055,"edge_left_empirical_bulk":-3.5648930282847147,"edge_right_empirical_bulk":28.76658076522706,"edge_left_abs_error":1.7171879646353574,"edge_right_abs_error":14.546469883719993,"left_edge_derivative":0.0028937635125961705,"right_edge_derivative":0.05668329859401898,"edge_solver_error":"","outlier_count":1,"informative_outlier_count":1,"max_outlier_teacher_overlap":0.525729906752915,"max_teacher_overlap":0.525729906752915,"min_full_eigenvalue":-4.743293121731041,"max_full_eigenvalue":44.58761211565991,"raw_edge_left_empirical_bulk":-3.5353727092956504,"raw_edge_right_empirical_bulk":87.4286922575503,"raw_outlier_count":1,"raw_informative_outlier_count":1,"raw_max_outlier_teacher_overlap":0.3039840301496974,"raw_min_full_eigenvalue":-4.69938451452596,"raw_max_full_eigenvalue":116.78581717914348},{"optimizer":"muon_a1_7_transfer","step":1200,"population_risk":0.4113791732431087,"n":145,"d":32,"block_size":6,"alpha_perp":5.0,"block_truncation_fraction":0.06206896551724138,"truncation_operator_norm":78.18787857625932,"edge_left_variational":-19.529984989633427,"edge_right_variational":46.47885445256679,"edge_left_empirical_bulk":-22.062987780926168,"edge_right_empirical_bulk":38.760939981685596,"edge_left_abs_error":2.5330027912927413,"edge_right_abs_error":7.7179144708811975,"left_edge_derivative":5.238357925918535e-05,"right_edge_derivative":0.3734698473019003,"edge_solver_error":"","outlier_count":2,"informative_outlier_count":1,"max_outlier_teacher_overlap":0.5002408047117991,"max_teacher_overlap":0.5002408047117991,"min_full_eigenvalue":-27.08749072007812,"max_full_eigenvalue":51.761132627449975,"raw_edge_left_empirical_bulk":-21.81482750773577,"raw_edge_right_empirical_bulk":86.58293810816707,"raw_outlier_count":2,"raw_informative_outlier_count":1,"raw_max_outlier_teacher_overlap":0.3615413769819549,"raw_min_full_eigenvalue":-26.761001442767512,"raw_max_full_eigenvalue":121.14272778613342},{"optimizer":"muon_a1_3_normalized","step":0,"population_risk":2.573494854014269,"n":145,"d":32,"block_size":6,"alpha_perp":5.0,"block_truncation_fraction":0.0,"truncation_operator_norm":1.2282870108272603e-15,"edge_left_variational":-1.1528043636082597,"edge_right_variational":1.11215336660237,"edge_left_empirical_bulk":-0.9535561908157966,"edge_right_empirical_bulk":0.886263076777374,"edge_left_abs_error":0.1992481727924631,"edge_right_abs_error":0.22589028982499593,"left_edge_derivative":0.001805850118737132,"right_edge_derivative":7.893572791850545e-06,"edge_solver_error":"","outlier_count":2,"informative_outlier_count":2,"max_outlier_teacher_overlap":0.3602374524347155,"max_teacher_overlap":0.5725396562449626,"min_full_eigenvalue":-1.2444907313723694,"max_full_eigenvalue":1.1711697999517232,"raw_edge_left_empirical_bulk":-0.9535561908157975,"raw_edge_right_empirical_bulk":0.886263076777374,"raw_outlier_count":4,"raw_informative_outlier_count":4,"raw_max_outlier_teacher_overlap":0.5193105547481304,"raw_min_full_eigenvalue":-1.244490731372371,"raw_max_full_eigenvalue":1.171169799951723},{"optimizer":"muon_a1_3_normalized","step":400,"population_risk":0.6561049903892726,"n":145,"d":32,"block_size":6,"alpha_perp":5.0,"block_truncation_fraction":0.041379310344827586,"truncation_operator_norm":55.714974268665,"edge_left_variational":-3.715438959090508,"edge_right_variational":46.41501978668872,"edge_left_empirical_bulk":-2.902275925408889,"edge_right_empirical_bulk":47.39551837756646,"edge_left_abs_error":0.8131630336816191,"edge_right_abs_error":0.9804985908777439,"left_edge_derivative":0.016697244366903174,"right_edge_derivative":0.0051212772125667425,"edge_solver_error":"","outlier_count":2,"informative_outlier_count":1,"max_outlier_teacher_overlap":0.565699847092403,"max_teacher_overlap":0.565699847092403,"min_full_eigenvalue":-3.7803499418940945,"max_full_eigenvalue":70.43949218247336,"raw_edge_left_empirical_bulk":-2.8968290361012805,"raw_edge_right_empirical_bulk":92.2973056983623,"raw_outlier_count":1,"raw_informative_outlier_count":1,"raw_max_outlier_teacher_overlap":0.35328523530896117,"raw_min_full_eigenvalue":-3.775187283070141,"raw_max_full_eigenvalue":111.79055536678594},{"optimizer":"muon_a1_3_normalized","step":800,"population_risk":0.3617530281046313,"n":145,"d":32,"block_size":6,"alpha_perp":5.0,"block_truncation_fraction":0.041379310344827586,"truncation_operator_norm":47.08895718348959,"edge_left_variational":-3.4313721065394103,"edge_right_variational":41.5036329021906,"edge_left_empirical_bulk":-1.8593611120653832,"edge_right_empirical_bulk":42.490449187040326,"edge_left_abs_error":1.572010994474027,"edge_right_abs_error":0.9868162848497235,"left_edge_derivative":0.022765294750778153,"right_edge_derivative":3.7922053244983545,"edge_solver_error":"","outlier_count":1,"informative_outlier_count":1,"max_outlier_teacher_overlap":0.30248734196140586,"max_teacher_overlap":0.4458710358359327,"min_full_eigenvalue":-2.9638866140527087,"max_full_eigenvalue":44.97415302776229,"raw_edge_left_empirical_bulk":-1.8565788682786926,"raw_edge_right_empirical_bulk":64.05989100874581,"raw_outlier_count":2,"raw_informative_outlier_count":2,"raw_max_outlier_teacher_overlap":0.34847637243737867,"raw_min_full_eigenvalue":-2.9585317054274616,"raw_max_full_eigenvalue":84.65863925202927},{"optimizer":"muon_a1_3_normalized","step":1200,"population_risk":0.08887456587831988,"n":145,"d":32,"block_size":6,"alpha_perp":5.0,"block_truncation_fraction":0.034482758620689655,"truncation_operator_norm":26.500018058464995,"edge_left_variational":-2.828268179162606,"edge_right_variational":34.01668393797921,"edge_left_empirical_bulk":-2.771115068702436,"edge_right_empirical_bulk":24.878944438187034,"edge_left_abs_error":0.05715311046016991,"edge_right_abs_error":9.137739499792175,"left_edge_derivative":0.02631627684517923,"right_edge_derivative":3.541355410902265,"edge_solver_error":"","outlier_count":1,"informative_outlier_count":1,"max_outlier_teacher_overlap":0.27589269269078354,"max_teacher_overlap":0.5253301838020098,"min_full_eigenvalue":-4.048508727030897,"max_full_eigenvalue":32.00280450894807,"raw_edge_left_empirical_bulk":-2.7685599886928633,"raw_edge_right_empirical_bulk":39.19047008869062,"raw_outlier_count":3,"raw_informative_outlier_count":3,"raw_max_outlier_teacher_overlap":0.3863072460359194,"raw_min_full_eigenvalue":-4.047151161610681,"raw_max_full_eigenvalue":55.14596622543704}]}