| { | |
| "paper": "Understanding Behavior Cloning with Action Quantization", | |
| "openreview_id": "9uENnRAcSl", | |
| "arxiv": "2603.20538", | |
| "seed": 20260727, | |
| "scale": 1.0, | |
| "claim_1": { | |
| "verdict": "verified", | |
| "policy": "four-context Bernoulli expert; exact categorical MLE", | |
| "rows": [ | |
| { | |
| "n": 1024, | |
| "repetitions": 3000, | |
| "mean_policy_tv": 0.020053743228177635, | |
| "mean_excess_logloss": 0.0019040958879226993, | |
| "se_policy_tv": 0.00014082953268802653 | |
| }, | |
| { | |
| "n": 2048, | |
| "repetitions": 3000, | |
| "mean_policy_tv": 0.014185037117486466, | |
| "mean_excess_logloss": 0.0009652704325506427, | |
| "se_policy_tv": 0.00010239188525647766 | |
| }, | |
| { | |
| "n": 4096, | |
| "repetitions": 3000, | |
| "mean_policy_tv": 0.010264473648817652, | |
| "mean_excess_logloss": 0.0004994902784918576, | |
| "se_policy_tv": 7.290809107968912e-05 | |
| }, | |
| { | |
| "n": 8192, | |
| "repetitions": 3000, | |
| "mean_policy_tv": 0.007162133524300517, | |
| "mean_excess_logloss": 0.0002435701855625456, | |
| "se_policy_tv": 5.070396414353111e-05 | |
| }, | |
| { | |
| "n": 16384, | |
| "repetitions": 3000, | |
| "mean_policy_tv": 0.005077544541866853, | |
| "mean_excess_logloss": 0.00012165853742260917, | |
| "se_policy_tv": 3.5532769884171806e-05 | |
| }, | |
| { | |
| "n": 32768, | |
| "repetitions": 3000, | |
| "mean_policy_tv": 0.0035740428028987275, | |
| "mean_excess_logloss": 6.0516308897687487e-05, | |
| "se_policy_tv": 2.6082470640695106e-05 | |
| }, | |
| { | |
| "n": 65536, | |
| "repetitions": 3000, | |
| "mean_policy_tv": 0.0025480675672977064, | |
| "mean_excess_logloss": 3.08705928019452e-05, | |
| "se_policy_tv": 1.815565410528769e-05 | |
| }, | |
| { | |
| "n": 131072, | |
| "repetitions": 3000, | |
| "mean_policy_tv": 0.0018010697407487843, | |
| "mean_excess_logloss": 1.5413261137027322e-05, | |
| "se_policy_tv": 1.2974458188285083e-05 | |
| } | |
| ], | |
| "tv_rate": { | |
| "slope": -0.4974457449942507, | |
| "r2": 0.999935441285942 | |
| }, | |
| "tv_rate_bootstrap_95ci": [ | |
| -0.5004936088510592, | |
| -0.49439478253905206 | |
| ], | |
| "excess_logloss_rate": { | |
| "slope": -0.9953732600294025, | |
| "r2": 0.9999408772607475 | |
| }, | |
| "finite_policy_class": { | |
| "rows": [ | |
| { | |
| "policy_class_size": 2, | |
| "mean_uniform_tv": 0.008944511413574219, | |
| "squared_tv_times_n": 0.3276975490152836, | |
| "repetitions": 2048 | |
| }, | |
| { | |
| "policy_class_size": 4, | |
| "mean_uniform_tv": 0.011336088180541992, | |
| "squared_tv_times_n": 0.5263642428908497, | |
| "repetitions": 2048 | |
| }, | |
| { | |
| "policy_class_size": 8, | |
| "mean_uniform_tv": 0.014003157615661621, | |
| "squared_tv_times_n": 0.8031781814643182, | |
| "repetitions": 2048 | |
| }, | |
| { | |
| "policy_class_size": 16, | |
| "mean_uniform_tv": 0.01623070240020752, | |
| "squared_tv_times_n": 1.0790326288552023, | |
| "repetitions": 2048 | |
| }, | |
| { | |
| "policy_class_size": 32, | |
| "mean_uniform_tv": 0.01839888095855713, | |
| "squared_tv_times_n": 1.3865730888792314, | |
| "repetitions": 2048 | |
| }, | |
| { | |
| "policy_class_size": 64, | |
| "mean_uniform_tv": 0.02026975154876709, | |
| "squared_tv_times_n": 1.682894142868463, | |
| "repetitions": 2048 | |
| }, | |
| { | |
| "policy_class_size": 128, | |
| "mean_uniform_tv": 0.022173762321472168, | |
| "squared_tv_times_n": 2.0139038125635125, | |
| "repetitions": 2048 | |
| }, | |
| { | |
| "policy_class_size": 256, | |
| "mean_uniform_tv": 0.023761272430419922, | |
| "squared_tv_times_n": 2.3125936845317483, | |
| "repetitions": 2048 | |
| } | |
| ], | |
| "linear_fit_against_log_size": { | |
| "slope": 0.41698453741029085, | |
| "intercept": -0.03411278761008589, | |
| "r2": 0.9968640268878375 | |
| } | |
| }, | |
| "quantization": [ | |
| { | |
| "bin_width": 0.25, | |
| "mean_absolute_error": 0.058256349920185066 | |
| }, | |
| { | |
| "bin_width": 0.125, | |
| "mean_absolute_error": 0.029113040960967253 | |
| }, | |
| { | |
| "bin_width": 0.0625, | |
| "mean_absolute_error": 0.014536913994716696 | |
| }, | |
| { | |
| "bin_width": 0.03125, | |
| "mean_absolute_error": 0.007269147606165697 | |
| }, | |
| { | |
| "bin_width": 0.015625, | |
| "mean_absolute_error": 0.00363748111544232 | |
| } | |
| ], | |
| "quantization_rate": { | |
| "slope": 1.000461473267501, | |
| "r2": 0.999999676826127 | |
| }, | |
| "gate": true, | |
| "negative_controls": { | |
| "fixed_error_floor_slope": 0.0, | |
| "parametric_excess_logloss_slope": -0.9953732600294025, | |
| "unquantized_oracle_floor": 0.0 | |
| } | |
| }, | |
| "claim_2": { | |
| "verdict": "verified", | |
| "stable_system": { | |
| "a": 0.9, | |
| "k": 0.35, | |
| "closed_loop": 0.55, | |
| "rows": [ | |
| { | |
| "H": 8, | |
| "cumulative_state_gap": 0.10425944736372471, | |
| "max_action_quantization_error": 0.019999990716951882 | |
| }, | |
| { | |
| "H": 16, | |
| "cumulative_state_gap": 0.21739253800437477, | |
| "max_action_quantization_error": 0.019999990716951882 | |
| }, | |
| { | |
| "H": 32, | |
| "cumulative_state_gap": 0.2919619853667145, | |
| "max_action_quantization_error": 0.019999990716951882 | |
| }, | |
| { | |
| "H": 64, | |
| "cumulative_state_gap": 0.308355322696201, | |
| "max_action_quantization_error": 0.019999990716951882 | |
| }, | |
| { | |
| "H": 128, | |
| "cumulative_state_gap": 0.30893754628499065, | |
| "max_action_quantization_error": 0.019999990716951882 | |
| }, | |
| { | |
| "H": 256, | |
| "cumulative_state_gap": 0.3089382335466882, | |
| "max_action_quantization_error": 0.019999990716951882 | |
| } | |
| ], | |
| "power_rate": { | |
| "slope": 0.26958723281092317, | |
| "r2": 0.6545192550840684 | |
| } | |
| }, | |
| "premise_violating_control": { | |
| "a": 1.23, | |
| "k": 0.15, | |
| "closed_loop": 1.08, | |
| "rows": [ | |
| { | |
| "H": 8, | |
| "cumulative_state_gap": 0.18674792060113443, | |
| "max_action_quantization_error": 0.01999999841360779 | |
| }, | |
| { | |
| "H": 16, | |
| "cumulative_state_gap": 0.6578288575350969, | |
| "max_action_quantization_error": 0.01999999841360779 | |
| }, | |
| { | |
| "H": 32, | |
| "cumulative_state_gap": 3.5866539438399743, | |
| "max_action_quantization_error": 0.019999999434039628 | |
| }, | |
| { | |
| "H": 64, | |
| "cumulative_state_gap": 51.87888413395724, | |
| "max_action_quantization_error": 0.019999999434039628 | |
| }, | |
| { | |
| "H": 128, | |
| "cumulative_state_gap": 7368.9180862935245, | |
| "max_action_quantization_error": 0.019999999434039628 | |
| }, | |
| { | |
| "H": 256, | |
| "cumulative_state_gap": 139899822.37753516, | |
| "max_action_quantization_error": 0.019999999896175878 | |
| } | |
| ], | |
| "semilog_slope": 0.0976068978271001, | |
| "expected_log_multiplier": 0.0769610411361284 | |
| }, | |
| "exact_rational_certificate": { | |
| "epsilon": "1/1000", | |
| "stable_alpha": "1/2", | |
| "unstable_alpha": "2", | |
| "stable_rows": [ | |
| { | |
| "H": 8, | |
| "state_gap_numerator": 51, | |
| "state_gap_denominator": 25600, | |
| "cumulative_gap_numerator": 1793, | |
| "cumulative_gap_denominator": 128000, | |
| "cumulative_gap_float": 0.0140078125 | |
| }, | |
| { | |
| "H": 16, | |
| "state_gap_numerator": 13107, | |
| "state_gap_denominator": 6553600, | |
| "cumulative_gap_numerator": 983041, | |
| "cumulative_gap_denominator": 32768000, | |
| "cumulative_gap_float": 0.030000030517578127 | |
| }, | |
| { | |
| "H": 32, | |
| "state_gap_numerator": 858993459, | |
| "state_gap_denominator": 429496729600, | |
| "cumulative_gap_numerator": 133143986177, | |
| "cumulative_gap_denominator": 2147483648000, | |
| "cumulative_gap_float": 0.06200000000046566 | |
| }, | |
| { | |
| "H": 64, | |
| "state_gap_numerator": 3689348814741910323, | |
| "state_gap_denominator": 1844674407370955161600, | |
| "cumulative_gap_numerator": 1162144876643701751809, | |
| "cumulative_gap_denominator": 9223372036854775808000, | |
| "cumulative_gap_float": 0.126 | |
| }, | |
| { | |
| "H": 128, | |
| "state_gap_numerator": 68056473384187692692674921486353642291, | |
| "state_gap_denominator": 34028236692093846346337460743176821145600, | |
| "cumulative_gap_numerator": 43215860598959184859848575143834562854913, | |
| "cumulative_gap_denominator": 170141183460469231731687303715884105728000, | |
| "cumulative_gap_float": 0.254 | |
| }, | |
| { | |
| "H": 256, | |
| "state_gap_numerator": 23158417847463239084714197001737581570653996933128112807891516801582625927987, | |
| "state_gap_denominator": 11579208923731619542357098500868790785326998466564056403945758400791312963993600, | |
| "cumulative_gap_numerator": 29526982755515629833010601177215416502583846089738343830061683922017848058183681, | |
| "cumulative_gap_denominator": 57896044618658097711785492504343953926634992332820282019728792003956564819968000, | |
| "cumulative_gap_float": 0.51 | |
| }, | |
| { | |
| "H": 512, | |
| "state_gap_numerator": 2681561585988519419914804999641169225495873164118478675544712288744352806014709395360374859633380685538006371637297210170750776562389313989286729801216819, | |
| "state_gap_denominator": 1340780792994259709957402499820584612747936582059239337772356144372176403007354697680187429816690342769003185818648605085375388281194656994643364900608409600, | |
| "cumulative_gap_numerator": 6851389852200667117882326774083187371141955934322713016016739897741821419367582505145757766363287651549606279533294371986268234116904697242627594642108973057, | |
| "cumulative_gap_denominator": 6703903964971298549787012499102923063739682910296196688861780721860882015036773488400937149083451713845015929093243025426876941405973284973216824503042048000, | |
| "cumulative_gap_float": 1.022 | |
| }, | |
| { | |
| "H": 1024, | |
| "state_gap_numerator": 35953862697246318154586103815780494672359539578846131454686016231546535161100192626541695464481507204224022775974278671531757953762883324498569486127894824875553578684973097055260443920249218823890616590417001153767630136468492576294782622108165447432670102136917259647989449187695943260967071265924844827443, | |
| "state_gap_denominator": 17976931348623159077293051907890247336179769789423065727343008115773267580550096313270847732240753602112011387987139335765878976881441662249284743063947412437776789342486548527630221960124609411945308295208500576883815068234246288147391311054082723716335051068458629823994724593847971630483535632962422413721600, | |
| "cumulative_gap_numerator": 183904007696414917360707921017717230249119044945797962390718973024360527349027485284760772300822909349605876499108435404884941933497148204810182921544182029238456554973637391437657170652074754284200503859982960901521428148036339527747813112083266263618107572430331783099466032595064749779846569525205581292371969, | |
| "cumulative_gap_denominator": 89884656743115795386465259539451236680898848947115328636715040578866337902750481566354238661203768010560056939935696678829394884407208311246423715319737062188883946712432742638151109800623047059726541476042502884419075341171231440736956555270413618581675255342293149119973622969239858152417678164812112068608000, | |
| "cumulative_gap_float": 2.046 | |
| } | |
| ], | |
| "unstable_rows": [ | |
| { | |
| "H": 8, | |
| "state_gap_numerator": 51, | |
| "state_gap_denominator": 200, | |
| "cumulative_gap_numerator": 251, | |
| "cumulative_gap_denominator": 500, | |
| "cumulative_gap_float": 0.502 | |
| }, | |
| { | |
| "H": 16, | |
| "state_gap_numerator": 13107, | |
| "state_gap_denominator": 200, | |
| "cumulative_gap_numerator": 65527, | |
| "cumulative_gap_denominator": 500, | |
| "cumulative_gap_float": 131.054 | |
| }, | |
| { | |
| "H": 32, | |
| "state_gap_numerator": 858993459, | |
| "state_gap_denominator": 200, | |
| "cumulative_gap_numerator": 4294967279, | |
| "cumulative_gap_denominator": 500, | |
| "cumulative_gap_float": 8589934.558 | |
| }, | |
| { | |
| "H": 64, | |
| "state_gap_numerator": 3689348814741910323, | |
| "state_gap_denominator": 200, | |
| "cumulative_gap_numerator": 18446744073709551583, | |
| "cumulative_gap_denominator": 500, | |
| "cumulative_gap_float": 3.68934881474191e+16 | |
| }, | |
| { | |
| "H": 128, | |
| "state_gap_numerator": 68056473384187692692674921486353642291, | |
| "state_gap_denominator": 200, | |
| "cumulative_gap_numerator": 340282366920938463463374607431768211391, | |
| "cumulative_gap_denominator": 500, | |
| "cumulative_gap_float": 6.80564733841877e+35 | |
| }, | |
| { | |
| "H": 256, | |
| "state_gap_numerator": 23158417847463239084714197001737581570653996933128112807891516801582625927987, | |
| "state_gap_denominator": 200, | |
| "cumulative_gap_numerator": 115792089237316195423570985008687907853269984665640564039457584007913129639807, | |
| "cumulative_gap_denominator": 500, | |
| "cumulative_gap_float": 2.315841784746324e+74 | |
| }, | |
| { | |
| "H": 512, | |
| "state_gap_numerator": 2681561585988519419914804999641169225495873164118478675544712288744352806014709395360374859633380685538006371637297210170750776562389313989286729801216819, | |
| "state_gap_denominator": 200, | |
| "cumulative_gap_numerator": 13407807929942597099574024998205846127479365820592393377723561443721764030073546976801874298166903427690031858186486050853753882811946569946433649006083839, | |
| "cumulative_gap_denominator": 500, | |
| "cumulative_gap_float": 2.6815615859885195e+151 | |
| }, | |
| { | |
| "H": 1024, | |
| "state_gap_numerator": 35953862697246318154586103815780494672359539578846131454686016231546535161100192626541695464481507204224022775974278671531757953762883324498569486127894824875553578684973097055260443920249218823890616590417001153767630136468492576294782622108165447432670102136917259647989449187695943260967071265924844827443, | |
| "state_gap_denominator": 200, | |
| "cumulative_gap_numerator": 179769313486231590772930519078902473361797697894230657273430081157732675805500963132708477322407536021120113879871393357658789768814416622492847430639474124377767893424865485276302219601246094119453082952085005768838150682342462881473913110540827237163350510684586298239947245938479716304835356329624224136703, | |
| "cumulative_gap_denominator": 500, | |
| "cumulative_gap_float": 3.595386269724632e+305 | |
| } | |
| ], | |
| "stable_power_rate": { | |
| "slope": 1.0228134265418711, | |
| "r2": 0.9997986173385389 | |
| }, | |
| "unstable_semilog_slope": 0.6931806827283984, | |
| "expected_unstable_semilog_slope": 0.6931471805599453 | |
| }, | |
| "rtvc_certificate": { | |
| "pairs": 11000, | |
| "bin_width": 0.04, | |
| "epsilon_q": 0.02, | |
| "epsilon_prime": 0.06, | |
| "delta0": 0.019999999999999997, | |
| "violations": 0 | |
| }, | |
| "gate": true, | |
| "scope": "Deterministic globally incrementally stable special case; the unstable control violates P-IISS." | |
| }, | |
| "claim_3": { | |
| "verdict": "verified", | |
| "construction": "Appendix D.2 deterministic scalar construction, A=0.2, B=0.3, d=0.6, expert arctan", | |
| "rows": [ | |
| { | |
| "epsilon_q": 0.08, | |
| "H": 64, | |
| "draws": 300000, | |
| "expert_distribution_error_per_H": 0.2865770810705352, | |
| "deployment_regret_per_H": 0.46551424401603325, | |
| "large_error_visit_fraction": 0.4794231770833333 | |
| }, | |
| { | |
| "epsilon_q": 0.04, | |
| "H": 96, | |
| "draws": 300000, | |
| "expert_distribution_error_per_H": 0.14683014036768757, | |
| "deployment_regret_per_H": 0.5872326641102398, | |
| "large_error_visit_fraction": 0.9458480208333333 | |
| }, | |
| { | |
| "epsilon_q": 0.02, | |
| "H": 144, | |
| "draws": 300000, | |
| "expert_distribution_error_per_H": 0.075415110039256, | |
| "deployment_regret_per_H": 0.6539699962570422, | |
| "large_error_visit_fraction": 0.9569177083333333 | |
| }, | |
| { | |
| "epsilon_q": 0.01, | |
| "H": 224, | |
| "draws": 300000, | |
| "expert_distribution_error_per_H": 0.03877583198711406, | |
| "deployment_regret_per_H": 0.6914201307788694, | |
| "large_error_visit_fraction": 0.9678262946428572 | |
| } | |
| ], | |
| "in_distribution_error_rate": { | |
| "slope": 0.9618303412708149, | |
| "r2": 0.9999985613126946 | |
| }, | |
| "deployment_error_rate": { | |
| "slope": -0.18675041159432026, | |
| "r2": 0.9122358671928921 | |
| }, | |
| "amplification": [ | |
| 1.6243945338443038, | |
| 3.999401367047049, | |
| 8.671604349799791, | |
| 17.83121329308012 | |
| ], | |
| "gate": true, | |
| "negative_control": "Claim 4 runs the same construction with auxiliary-state actions." | |
| }, | |
| "claim_4": { | |
| "verdict": "registered_claim_falsified_corrected_theorem_verified", | |
| "source_correction": "Theorem 7 is H*[sqrt((log(|Pi|/delta)+log(|M|/delta))/n)+epsilon_q], and assumes realizability of both q#pi* in Pi and T\u2218rho in M. The registered claim omits log|M| and transition-model realizability.", | |
| "algorithm": "Algorithm 1 exact-realizable auxiliary rollout on Appendix D.2 construction", | |
| "rows": [ | |
| { | |
| "epsilon_q": 0.08, | |
| "H": 64, | |
| "draws": 300000, | |
| "expert_distribution_error_per_H": 0.2865199387279938, | |
| "deployment_regret_per_H": 0.18374868238076336, | |
| "large_error_visit_fraction": 0.02047895833333333 | |
| }, | |
| { | |
| "epsilon_q": 0.04, | |
| "H": 96, | |
| "draws": 300000, | |
| "expert_distribution_error_per_H": 0.146788750047013, | |
| "deployment_regret_per_H": 0.09490071568936688, | |
| "large_error_visit_fraction": 0.008620555555555555 | |
| }, | |
| { | |
| "epsilon_q": 0.02, | |
| "H": 144, | |
| "draws": 300000, | |
| "expert_distribution_error_per_H": 0.07538600164748555, | |
| "deployment_regret_per_H": 0.049353863408457925, | |
| "large_error_visit_fraction": 0.007158009259259259 | |
| }, | |
| { | |
| "epsilon_q": 0.01, | |
| "H": 224, | |
| "draws": 300000, | |
| "expert_distribution_error_per_H": 0.03875635878911714, | |
| "deployment_regret_per_H": 0.02579398490042544, | |
| "large_error_visit_fraction": 0.005360535714285714 | |
| } | |
| ], | |
| "online_rows_same_seed": [ | |
| { | |
| "epsilon_q": 0.08, | |
| "H": 64, | |
| "draws": 300000, | |
| "expert_distribution_error_per_H": 0.2865199387279938, | |
| "deployment_regret_per_H": 0.46546787982748505, | |
| "large_error_visit_fraction": 0.47939359375 | |
| }, | |
| { | |
| "epsilon_q": 0.04, | |
| "H": 96, | |
| "draws": 300000, | |
| "expert_distribution_error_per_H": 0.146788750047013, | |
| "deployment_regret_per_H": 0.5871712285124917, | |
| "large_error_visit_fraction": 0.9457463541666666 | |
| }, | |
| { | |
| "epsilon_q": 0.02, | |
| "H": 144, | |
| "draws": 300000, | |
| "expert_distribution_error_per_H": 0.07538600164748555, | |
| "deployment_regret_per_H": 0.6539321985619908, | |
| "large_error_visit_fraction": 0.9568546296296296 | |
| }, | |
| { | |
| "epsilon_q": 0.01, | |
| "H": 224, | |
| "draws": 300000, | |
| "expert_distribution_error_per_H": 0.03875635878911714, | |
| "deployment_regret_per_H": 0.691389839642947, | |
| "large_error_visit_fraction": 0.9677867113095238 | |
| } | |
| ], | |
| "auxiliary_error_rate": { | |
| "slope": 0.9441137928467624, | |
| "r2": 0.9999834777303975 | |
| }, | |
| "expert_error_rate": { | |
| "slope": 0.9619764641320817, | |
| "r2": 0.9999985889480645 | |
| }, | |
| "improvement_online_over_auxiliary": [ | |
| 2.5331766943664076, | |
| 6.187216020945995, | |
| 13.24986846824892, | |
| 26.804305046776363 | |
| ], | |
| "gate": true, | |
| "destructive_control": "Actions conditioned on the real OOD state reproduce Theorem 6's constant-error trap." | |
| }, | |
| "claim_5": { | |
| "verdict": "verified", | |
| "deterministic": { | |
| "rows": [ | |
| { | |
| "n": 16, | |
| "delta": 0.020833333333333332, | |
| "no_rare_mc": 0.7112533333333333, | |
| "no_rare_exact": 0.7140126875194704, | |
| "minimax_regret_per_H": 0.0074088888888888885 | |
| }, | |
| { | |
| "n": 32, | |
| "delta": 0.010416666666666666, | |
| "no_rare_mc": 0.7158233333333334, | |
| "no_rare_exact": 0.7152797201836795, | |
| "minimax_regret_per_H": 0.003728246527777778 | |
| }, | |
| { | |
| "n": 64, | |
| "delta": 0.005208333333333333, | |
| "no_rare_mc": 0.7157666666666667, | |
| "no_rare_exact": 0.7159074252003099, | |
| "minimax_regret_per_H": 0.0018639756944444444 | |
| }, | |
| { | |
| "n": 128, | |
| "delta": 0.0026041666666666665, | |
| "no_rare_mc": 0.7154066666666666, | |
| "no_rare_exact": 0.716219842818941, | |
| "minimax_regret_per_H": 0.0009315190972222221 | |
| }, | |
| { | |
| "n": 256, | |
| "delta": 0.0013020833333333333, | |
| "no_rare_mc": 0.71569, | |
| "no_rare_exact": 0.716375695116003, | |
| "minimax_regret_per_H": 0.0004659440104166667 | |
| }, | |
| { | |
| "n": 512, | |
| "delta": 0.0006510416666666666, | |
| "no_rare_mc": 0.7176066666666666, | |
| "no_rare_exact": 0.7164535324107901, | |
| "minimax_regret_per_H": 0.00023359592013888887 | |
| }, | |
| { | |
| "n": 1024, | |
| "delta": 0.0003255208333333333, | |
| "no_rare_mc": 0.7176466666666667, | |
| "no_rare_exact": 0.7164924288788538, | |
| "minimax_regret_per_H": 0.0001168044704861111 | |
| }, | |
| { | |
| "n": 2048, | |
| "delta": 0.00016276041666666666, | |
| "no_rare_mc": 0.7162933333333333, | |
| "no_rare_exact": 0.7165118715724306, | |
| "minimax_regret_per_H": 5.8292100694444445e-05 | |
| } | |
| ], | |
| "rate": { | |
| "slope": -0.9987935382476681, | |
| "r2": 0.9999987671970029 | |
| } | |
| }, | |
| "stochastic": { | |
| "rows": [ | |
| { | |
| "n": 16, | |
| "separation": 0.075, | |
| "testing_error": 0.2777866666666667, | |
| "minimax_regret_per_H": 0.041668000000000004 | |
| }, | |
| { | |
| "n": 32, | |
| "separation": 0.05303300858899106, | |
| "testing_error": 0.27522, | |
| "minimax_regret_per_H": 0.029191489247724238 | |
| }, | |
| { | |
| "n": 64, | |
| "separation": 0.0375, | |
| "testing_error": 0.2770666666666667, | |
| "minimax_regret_per_H": 0.02078 | |
| }, | |
| { | |
| "n": 128, | |
| "separation": 0.02651650429449553, | |
| "testing_error": 0.27476, | |
| "minimax_regret_per_H": 0.014571349439911183 | |
| }, | |
| { | |
| "n": 256, | |
| "separation": 0.01875, | |
| "testing_error": 0.27546000000000004, | |
| "minimax_regret_per_H": 0.01032975 | |
| }, | |
| { | |
| "n": 512, | |
| "separation": 0.013258252147247764, | |
| "testing_error": 0.27627999999999997, | |
| "minimax_regret_per_H": 0.007325979806483224 | |
| }, | |
| { | |
| "n": 1024, | |
| "separation": 0.009375, | |
| "testing_error": 0.27457333333333334, | |
| "minimax_regret_per_H": 0.00514825 | |
| }, | |
| { | |
| "n": 2048, | |
| "separation": 0.006629126073623882, | |
| "testing_error": 0.27358666666666664, | |
| "minimax_regret_per_H": 0.0036272810107916913 | |
| } | |
| ], | |
| "rate": { | |
| "slope": -0.5021364313210787, | |
| "r2": 0.9999835397879303 | |
| } | |
| }, | |
| "uniform_quantizer": [ | |
| { | |
| "K": 4, | |
| "max_cell_diameter_epsilon_q": 0.25 | |
| }, | |
| { | |
| "K": 8, | |
| "max_cell_diameter_epsilon_q": 0.125 | |
| }, | |
| { | |
| "K": 16, | |
| "max_cell_diameter_epsilon_q": 0.0625 | |
| }, | |
| { | |
| "K": 32, | |
| "max_cell_diameter_epsilon_q": 0.03125 | |
| }, | |
| { | |
| "K": 64, | |
| "max_cell_diameter_epsilon_q": 0.015625 | |
| } | |
| ], | |
| "gate": true, | |
| "negative_controls": "Removing the rare-state pair sets the deterministic testing term to zero; fixing stochastic separation instead of n^-1/2 removes the claimed rate." | |
| }, | |
| "claim_6": { | |
| "verdict": "registered_empirical_claim_falsified_independent_mechanism_supported", | |
| "source_correction": "The paper reports no experiment comparing binning and learned quantizers or deterministic-expert RTVC violation frequency. Proposition 5 is theoretical; the nearby practice statement cites external Pertsch et al. (2025).", | |
| "independent_diagnostic": { | |
| "training_samples": 300000, | |
| "K": 16, | |
| "uniform_train_mae": 0.030816012479267792, | |
| "learned_train_mae": 0.008234504329978, | |
| "uniform_global_max_error": 0.0625, | |
| "learned_global_max_error": 0.43053728525219837, | |
| "epsilon_prime": 0.1875, | |
| "tested_state_radius": 0.0625, | |
| "uniform_rtvc_violations": 0, | |
| "learned_rtvc_violations": 625, | |
| "uniform_max_output_change": 0.125, | |
| "learned_max_output_change": 0.8610763955740226, | |
| "learned_centers": [ | |
| -0.6687394515046046, | |
| -0.623468769649742, | |
| -0.5910187759015997, | |
| -0.5627725784936948, | |
| -0.5357621103923741, | |
| -0.5075413850673305, | |
| -0.47536141379959346, | |
| -0.43123911032182416, | |
| 0.42983728525219844, | |
| 0.47419526064576834, | |
| 0.5064991412100074, | |
| 0.5347021312019845, | |
| 0.561846239084266, | |
| 0.5899897496447543, | |
| 0.6222380951251071, | |
| 0.6664194369345043 | |
| ] | |
| }, | |
| "gate": true, | |
| "destructive_control": "Uniformly distributed training actions make Lloyd cells approximately uniform and remove the tail-cell pathology." | |
| }, | |
| "environment": { | |
| "python": "3.12.13", | |
| "numpy": "1.26.4", | |
| "platform": "Linux-6.18.33.2-microsoft-standard-WSL2-x86_64-with-glibc2.39", | |
| "processor": "x86_64" | |
| }, | |
| "wall_seconds": 39.67354440689087, | |
| "all_gates_pass": true | |
| } | |
Xet Storage Details
- Size:
- 26.3 kB
- Xet hash:
- 3979c77633da443e8d7f5540063e9da1c85aa727db778e53375efb9d1cd530e0
·
Xet efficiently stores files, intelligently splitting them into unique chunks and accelerating uploads and downloads. More info.