Stochastic Kernel Mode Decomposition: Information, Uncertainty, and Conditional Regime Inference
Organizations: The Alan Turing Institute, London NW1 2DB, UK · Department of Computing and Mathematical Sciences, California Institute of Technology, Pasadena, CA 91125, USA
Abstract
When do mode energies preserve regime information as measurement quality changes? Stochastic Kernel Mode Decomposition (sKMD) fits joint regime laws on signed KMD mode coordinates and transfers the reference mode posterior to regime probabilities and conditional reconstructions. For centered Gaussian regimes with independently varying known noise variances, fixed coordinate energies retain the full regime posterior precisely when all classes share their cross-covariances and class variances can differ only at isolated vertices of the covariance graph. We extend the characterization to nonzero means and show that freely varying correlated noise forces a fixed linear log-odds decoder to retain the full quadratic feature span even for two distinct centered regimes. At fixed quality, a three-regime example has no sufficient two-energy basis. Its smallest Bayes log-loss gap over bases of condition number at most has sharp order . Exact Bayes energy posteriors supply numerical gap estimates for specified bases. Additional bounds relate feature compression, independently learned mode spans and sensor-noise error to log loss. Two controlled experiments with 64 independent panels each show that signed information can improve prediction over energy summaries, while integration gives small gains over strong signed controls and depends on how the competing point rule is calibrated and how accurately observation noise is measured.
Figures & tables
| Symbol | Meaning | Scope |
|---|---|---|
| , , | regime label in , number of regimes, regime priors | throughout |
| ; , | latent mode coefficients; signed mode coordinates (random, observed value) | § 2 ; § 3 – 5 |
| , , | dictionary (synthesis) matrix, observation operator, observation-noise covariance | § 2 and Part II |
| ; | reference KMD prior covariance; reference posterior mean and covariance | throughout |
| , ; | regime mean and covariance of the mode coefficients; regime precision | throughout |
| , | regime posterior precision and information vector in the transfer formula | Proposition 2.2 |
| Quantity | , orthogonal ( ) | , ( ) |
|---|---|---|
| Bound of Theorem 3.4 | ( ) | — |
| Bound of Theorem 3.5 | ||
| Measured gap | (selected rotation) | |
| Quadrature estimate | (best rotation found) |
| 1 | 1.32201 | 1.43000 | 0.53633 | 0.55883 | 0.39534 | 0.00764 | 1.98665 | 0.98518 | 0.31651 |
|---|---|---|---|---|---|---|---|---|---|
| 2 | 0.35523 | 0.41266 | 0.27079 | 0.01851 | 0.08097 | 0.00418 | 0.42334 | 0.39455 | 0.22079 |
| 3 | 0.92668 | 0.88353 | 1.17183 | 0.12033 | 0.41607 | 0.13382 | 1.53253 | 0.83515 | 0.61437 |
| 4 | 0.42748 | 1.19888 | 0.62136 | 0.41533 | 0.09907 | 0.03364 | 1.38651 | 0.63033 | 0.23088 |
| 5 | 1.14832 | 0.75828 | 1.65831 | 0.41444 | 0.41813 | 0.06662 | 1.95537 | 1.14161 | 0.46794 |
| 6 | 0.97385 | 0.96233 | 0.86144 | 0.00072 | 0.06427 | 0.02665 | 1.00636 | 0.96329 | 0.82797 |
| Quantity | Value |
|---|---|
| Local coefficient (600-point quadrature per axis) | 0.360836 |
| Coefficient difference, 400 vs. 600 points per axis | 3.8e-09 |
| Independent-latent example: full Bayes accuracy | 0.621314 |
| Independent-latent example: physical-energy accuracy | 0.610228 |
| Conservative Monte Carlo SE of energy accuracy | 6.8e-05 |
| Physical-energy log-loss deficiency | 0.007520 |
| Library condition | Excess bound | Observed gain | ||
|---|---|---|---|---|
| On grid | 3,000 | 0.0000 | 0.0199 | |
| Off grid | 12,000 | 0.0097 | 0.0286 | |
| Outside library | 3,000 | 3.1653 | 3.2457 |
| Study / units | Result against stated control | Evaluation conditions | Status | Criterion met |
|---|---|---|---|---|
| Binary signed modes / 64 panels (§ 7.2 ) | Integrated NLL , selected signed , same-law point . | The point contrast reflects the point rule’s temperature grid; quality-aware calibration removes of it (§ 7.3 ). | Conf. (primary family) | Illustrative margin only |
| Three-regime continuous quality / 64 panels (§ 7.4 ) | Integrated NLL , selected signed , centered signed sensitivity . | NLL improves by over the selected signed control and by over the retrospective centered control; biased monitoring favors the centered control. | Conf. (primary endpoint); centered: Expl. | None declared |
| Adaptive oscillators / 8 panels (§ 8.1 ) | Energy-interaction logistic means , versus sKMD . | Source contrast inconclusive: , ; shift contrast unresolved: , (bootstrap ). | Expl. | None stated |
| Triggered ringdown / 6 panels (§ 8.2 ) | Source sKMD versus signed OAS . Noise-shift NLL contrast inconclusive. | One -nat event; trigger/dynamics shift also fails. Noisy MSE improves over pooled KMD. | Expl. | No |
| Chirps and fading / 12 panels each (§ 8.3 ) | Fading noise-shift gain over OAS is ; its interval does not certify nats. Chirp contrast inconclusive. | Fading MSE improves over independent-mode KMD, over pooled KMD. | Expl. | No (fading NLL point estimate only) |
| Switching / 12 panels (§ 8.5 ) | Noise-shift NLL gain over point OAS is ; changed hazards reverse the comparison. | MSE gain is over pooled KMD, over point-weighted conditional means. | Expl. | Yes for noise-shift NLL and MSE vs pooled; no elsewhere |
| Control | Difference | Paired 95% interval | Holm |
|---|---|---|---|
| Selected signed | |||
| Same-law point, bounded grid | |||
| Retrospective, exploratory: | |||
| Smooth quality-temperature point | — | ||
| Method | Mean NLL | Integrated minus method (95% CI) |
|---|---|---|
| Point, bounded grid | 0.65107 | -0.06847 [-0.07159, -0.06535] |
| Global temperature | 0.64579 | -0.06319 [-0.06496, -0.06142] |
| Four quality temperatures | 0.58486 | -0.00227 [-0.00295, -0.00158] |
| Smooth quality temperature | 0.58414 | -0.00155 [-0.00214, -0.00095] |
| Selected signed control | 0.58460 | -0.00200 [-0.00301, -0.00099] |
| Integrated Gaussian | 0.58260 | — |
| Method | 600 fit events | 2400 fit events |
|---|---|---|
| Integrated Gaussian | 1.03934 | 1.03371 |
| Development-selected signed | 1.04273 | 1.03617 |
| Centered signed quadratic (sensitivity) | 1.04059 | 1.03487 |
| Point with continuous global calibration | 1.05992 | 1.05494 |
| Quality calibration of point logits | 1.04399 | 1.03612 |
| Fitted energy polynomial | 1.06278 | 1.05642 |
| Method | Source NLL | Shift NLL | Source Brier | Shift Brier |
|---|---|---|---|---|
| Coupled sKMD | 0.636 | 0.749 | 0.394 | 0.461 |
| Amplitude point QDA | 0.636 | 0.763 | 0.394 | 0.465 |
| Signed quadrature QDA | 1.108 | 1.111 | 0.672 | 0.674 |
| Amplitude quadratic logistic | 0.645 | 0.749 | 0.393 | 0.456 |
| Energy interaction logistic | 0.615 | 0.724 | 0.376 | 0.445 |
| Pooled KMD energies + RF | 0.948 | 1.008 | 0.566 | 0.605 |
| Method | Source | Noise/missingness shift | Trigger/dynamics shift |
|---|---|---|---|
| Coupled sKMD / matched Gaussian | 0.812 | 0.785 | 0.935 |
| Diagonal class covariance + noise | 0.869 | 0.865 | 0.889 |
| Signed OAS-QDA | 0.797 | 0.809 | 0.962 |
| Zero-mean point QDA | 0.882 | 1.334 | 3.166 |
| Signed quadratic logistic | 0.832 | 1.262 | 2.923 |
| Calibrated direct sparse QDA | 0.854 | 1.356 | 3.146 |
| Reconstruction | Source | Noise/missingness shift | Trigger/dynamics shift |
|---|---|---|---|
| Independent KMD | 0.01721 | 0.05748 | 0.15253 |
| Pooled full covariance | 0.01594 | 0.04931 | 0.20717 |
| Coupled class mixture | 0.01524 | 0.04601 | 0.19851 |
| Unregularized least squares | 0.01904 | 0.07572 | 0.13713 |
| Method | Source | Noise/missingness shift | Trigger/dynamics shift |
|---|---|---|---|
| Coupled sKMD / matched Gaussian | 1.021 | 0.808 | 1.138 |
| Signed OAS-QDA | 0.831 | 0.826 | 0.962 |
| Full-V signed quadratic logistic | 0.926 | 2.110 | 3.492 |
| Full-V signed RBF-SVM | 1.051 | 1.178 | 1.192 |
| Full-V energy interaction logistic | 0.946 | 1.940 | 1.917 |
| Full-V energy paths + RF | 0.879 | 0.858 | 0.896 |
| Method | Source | Noise/missingness | Carrier shift |
|---|---|---|---|
| Coupled sKMD / matched Gaussian | 0.760 | 0.768 | 1.218 |
| Signed OAS-QDA | 0.760 | 0.775 | 1.221 |
| Full-V signed quadratic logistic | 0.862 | 2.599 | 7.042 |
| Full-V signed RBF-SVM | 1.049 | 1.199 | 1.212 |
| Full-V energy interaction logistic | 0.895 | 2.898 | 3.910 |
| Full-V energy paths + RF | 0.815 | 0.850 | 1.057 |
| Reconstruction | Source | Noise/missingness | Carrier shift |
|---|---|---|---|
| Independent KMD | 0.01336 | 0.04956 | 1.14995 |
| Pooled full covariance | 0.01078 | 0.03720 | 1.23667 |
| Coupled class mixture | 0.01054 | 0.03636 | 1.22386 |
| Unregularized LS | 0.01494 | 0.06877 | 1.01315 |
| Family | Method | Source | Noise | In library | Outside |
|---|---|---|---|---|---|
| Chirp MSE | Joint mixture | 0.0129 | 0.0648 | 0.0843 | 1.4318 |
| Chirp MSE | Pooled mixture | 0.0128 | 0.0657 | 0.0856 | 1.4440 |
| Chirp MSE | Independent mixture | 0.0159 | 0.0772 | 0.0878 | 1.3831 |
| Chirp MSE | Fixed coupled | 0.0103 | 0.0393 | 0.8862 | 1.4478 |
| Chirp MSE | Coupled MAP carrier | 0.0149 | 0.0842 | 0.1109 | 1.4957 |
| Chirp MSE | Joint marginals product | 0.0129 | 0.0649 | 0.0843 | 1.4316 |
| Method | Source | Noise | Hazard | Phase | Low |
|---|---|---|---|---|---|
| Coupled filter | 0.435 | 0.497 | 0.621 | 0.769 | 1.146 |
| Static coupled | 0.789 | 0.871 | 0.796 | 0.988 | 1.145 |
| Point OAS filter | 0.427 | 0.760 | 0.603 | 2.205 | 1.205 |
| Signed quadratic | 0.445 | 5.415 | 0.631 | 6.639 | 4.107 |
| Physical energies | 0.500 | 0.606 | 0.675 | 0.897 | 1.275 |
| Rotated energies | 0.780 | 0.944 | 0.904 | 1.081 | 1.344 |
| Method | Source | Noise | Hazard | Phase | Low |
|---|---|---|---|---|---|
| Coupled filter | 0.0315 | 0.3776 | 0.0319 | 1.8294 | 0.3084 |
| Static coupled | 0.0315 | 0.3954 | 0.0318 | 1.8480 | 0.3084 |
| Point weights | 0.0315 | 0.3852 | 0.0319 | 1.8397 | 0.3084 |
| Full pooled KMD | 0.0316 | 0.4171 | 0.0319 | 1.8707 | 0.3084 |
| Diagonal KMD | 0.0317 | 0.4292 | 0.0320 | 1.9022 | 0.3083 |
| Minimum-norm LS | 0.0468 | 5.7407 | 0.0472 | 1.9263 | 11.0711 |
| Method | Source | Gaussian | BW | MA | EM |
|---|---|---|---|---|---|
| Coupled sKMD | 0.3302 | 0.3375 | 0.3651 | 0.3748 | 0.4744 |
| Diagonal class | 0.3564 | 0.3588 | 0.3862 | 0.3921 | 0.4673 |
| Shared class | 0.3673 | 0.3985 | 0.3965 | 0.4217 | 0.4103 |
| PCA7 integrated | 0.3500 | 0.3388 | 0.3759 | 0.3804 | 0.4545 |
| Point QDA | 0.2799 | 0.2980 | 0.2870 | 0.2957 | 0.3266 |
| Signed quadratic | 0.3064 | 0.3164 | 0.3099 | 0.3261 | 0.3609 |
| Method | Brier | Accuracy | AP | AUROC |
|---|---|---|---|---|
| Coupled sKMD | 0.1860 | 0.8853 | 0.7673 | 0.9311 |
| Diagonal class | 0.1998 | 0.8726 | 0.7341 | 0.9174 |
| Shared class | 0.2389 | 0.8524 | 0.5531 | 0.7964 |
| PCA7 integrated | 0.1892 | 0.8822 | 0.7764 | 0.9168 |
| Point QDA | 0.1668 | 0.8908 | 0.7758 | 0.9280 |
| Signed quadratic | 0.1781 | 0.8925 | 0.7838 | 0.9099 |
| Method | Source | Gaussian | BW | MA | EM |
|---|---|---|---|---|---|
| Coupled sKMD | 0.1032 | 0.1069 | 0.1422 | 0.1345 | 0.1293 |
| Diagonal class | 0.1032 | 0.1070 | 0.1422 | 0.1346 | 0.1294 |
| Shared class | 0.1032 | 0.1069 | 0.1422 | 0.1346 | 0.1293 |
| PCA7 integrated | 0.0232 | 0.0261 | 0.0650 | 0.0576 | 0.0580 |
| Point-weighted mixture | 0.1032 | 0.1069 | 0.1422 | 0.1345 | 0.1293 |
| Pooled full | 0.1032 | 0.1069 | 0.1422 | 0.1346 | 0.1294 |
| Method | Source | Gaussian | Wander | Muscle | Electrode |
|---|---|---|---|---|---|
| Localized robust sKMD | 0.071 | 0.039 | 0.068 | 0.083 | 0.071 |
| Robust point + RR | 0.111 | 0.096 | 0.112 | 0.106 | 0.118 |
| Robust PCA43 | 0.102 | 0.063 | 0.098 | 0.103 | 0.150 |
| Gaussian point + RR | 0.119 | 0.121 | 0.128 | 0.124 | 0.137 |
| Signed quadratic logistic | 0.281 | 0.279 | 0.279 | 0.281 | 0.286 |
| Signed RBF | 0.123 | 0.121 | 0.122 | 0.122 | 0.124 |
| Method | Source | Gaussian | Wander | Muscle | Electrode |
|---|---|---|---|---|---|
| Localized robust sKMD | 3.977 | 10.579 | 53.723 | 46.569 | 50.256 |
| Point-weighted conditional means | 3.977 | 10.625 | 53.232 | 46.229 | 49.938 |
| Independent pooled full prior | 3.972 | 9.151 | 49.565 | 44.414 | 47.202 |
| Independent pooled diagonal prior | 3.980 | 13.527 | 55.279 | 48.469 | 51.529 |
| Original kernel KMD | 4.012 | 89.432 | 66.383 | 61.321 | 60.933 |
| Robust PCA43 | 2.663 | 8.119 | 46.596 | 41.999 | 44.423 |
| Bearing NLL | Held-out | |||
|---|---|---|---|---|
| Method | Nominal | Low torque | Low speed | accuracy |
| Coupled sKMD | 1.548 | 1.700 | 10.346 | 5/15 |
| Same-prior point | 1.392 | 1.542 | 8.465 | 5/15 |
| Diagonal uncertainty | 2.520 | 2.417 | 7.850 | 5/15 |
| Signed shrinkage QDA | 1.443 | 1.509 | 9.181 | 6/15 |
| Signed OAS | 2.251 | 2.218 | 5.709 | 3/15 |
| Method | Linked-axis interpolation | Nominal indexing |
|---|---|---|
| Coupled sKMD | 10.346 | 11.722 |
| Same-prior point | 8.465 | 9.591 |
| Diagonal uncertainty | 7.850 | 8.888 |
| Signed shrinkage QDA | 9.181 | 10.511 |
| Signed OAS QDA | 5.709 | 5.836 |
| Calibrated direct sparse QDA | 10.449 | 13.228 |
| Method | Macro NLL | Micro NLL | Macro accuracy |
|---|---|---|---|
| Replicate selector ( ) | 0.655 | 0.889 | 0.759 |
| Same-prior point | 0.655 | 0.889 | 0.759 |
| Point QDA | 0.831 | 1.084 | 0.638 |
| Student | 1.040 | 1.412 | 0.622 |
| Mean-spectrum forest | 0.970 | 1.031 | 0.698 |
| Window-spectrum forest | 0.613 | 0.812 | 0.728 |
| Method | Moderate | Severe | Mean |
|---|---|---|---|
| Integrated sKMD | 0.797 | 0.831 | 0.814 |
| Point prior | 0.774 | 1.026 | 0.900 |
| Full-posterior forest | 1.043 | 1.065 | 1.054 |
| Method | Moderate | Severe | Mean |
|---|---|---|---|
| Integrated sKMD | 0.04212 | 0.08403 | 0.06307 |
| Point prior | 0.04293 | 0.08971 | 0.06632 |
| Full-posterior forest | 0.04861 | 0.09323 | 0.07092 |
| Pooled Gaussian KMD | 0.04946 | 0.09754 | 0.07350 |
| Unlabeled local mixture | 0.04272 | 0.08434 | 0.06353 |
| Unlabeled group mixture | 0.04215 | 0.08094 | 0.06154 |
| Representation | Accuracy | NLL | Brier | |
|---|---|---|---|---|
| 0.50 | Full coefficients | |||
| 0.50 | Fixed energies | |||
| 0.50 | Selected rotation | |||
| 0.50 | Three energies | |||
| 0.50 | ||||
| 0.50 |
| Representation | Accuracy gap | NLL gap | Accuracy, | NLL, | |
|---|---|---|---|---|---|
| 0.50 | Fixed energies | ||||
| 0.50 | Selected rotation | ||||
| 0.50 | Three energies | ||||
| 0.50 | |||||
| 0.50 | |||||
| 0.50 |
| sKMD | QDA | OAS | ZQ | Quad | Red | RF | GB | Oracle | |||
|---|---|---|---|---|---|---|---|---|---|---|---|
| 0.50 | 30 | 0 | 1.057 | 1.098 | 1.092 | 1.061 | 1.068 | 1.072 | 1.114 | 1.362 | 1.034 |
| 0.50 | 30 | 1 | 1.058 | 1.100 | 1.092 | 1.061 | 1.110 | 1.108 | 1.094 | 1.276 | 1.033 |
| 0.50 | 30 | 2 | 1.074 | 1.113 | 1.111 | 1.082 | 1.117 | 1.115 | 1.097 | 1.233 | 1.063 |
| 0.50 | 120 | 0 | 1.038 | 1.043 | 1.042 | 1.040 | 1.042 | 1.043 | 1.076 | 1.136 | 1.033 |
| 0.50 | 120 | 1 | 1.041 | 1.046 | 1.045 | 1.042 | 1.049 | 1.049 | 1.074 | 1.093 | 1.035 |
| 0.50 | 120 | 2 | 1.063 | 1.067 | 1.067 | 1.064 | 1.069 | 1.069 | 1.081 | 1.102 | 1.060 |
| sKMD | QDA | OAS | ZQ | Quad | Red | RF | GB | Oracle | |||
|---|---|---|---|---|---|---|---|---|---|---|---|
| 0.50 | 30 | 0 | 1.051 | 1.095 | 1.090 | 1.058 | 1.069 | 1.078 | 1.122 | 1.349 | 1.034 |
| 0.50 | 30 | 1 | 1.053 | 1.098 | 1.090 | 1.059 | 1.110 | 1.104 | 1.103 | 1.304 | 1.033 |
| 0.50 | 30 | 2 | 1.073 | 1.113 | 1.110 | 1.081 | 1.119 | 1.117 | 1.102 | 1.235 | 1.063 |
| 0.50 | 120 | 0 | 1.038 | 1.043 | 1.042 | 1.039 | 1.042 | 1.044 | 1.094 | 1.141 | 1.033 |
| 0.50 | 120 | 1 | 1.041 | 1.046 | 1.045 | 1.042 | 1.049 | 1.049 | 1.089 | 1.093 | 1.035 |
| 0.50 | 120 | 2 | 1.063 | 1.067 | 1.067 | 1.064 | 1.069 | 1.069 | 1.092 | 1.105 | 1.060 |
| sKMD NLL | Diag NLL | Point NLL | KMD MSE | Pool MSE | sKMD MSE | Gain | |||
|---|---|---|---|---|---|---|---|---|---|
| 0.50 | 30 | 0 | 1.0567 | 1.0859 | 1.0583 | 1.208 | 1.207 | 1.207 | 0.07 |
| 0.50 | 30 | 1 | 1.0584 | 1.0857 | 1.0616 | 2.454 | 2.450 | 2.450 | 0.20 |
| 0.50 | 30 | 2 | 1.0742 | 1.0889 | 1.0802 | 10.965 | 10.898 | 10.885 | 0.73 |
| 0.50 | 120 | 0 | 1.0384 | 1.0652 | 1.0386 | 1.206 | 1.204 | 1.203 | 0.21 |
| 0.50 | 120 | 1 | 1.0409 | 1.0670 | 1.0413 | 2.433 | 2.429 | 2.426 | 0.26 |
| 0.50 | 120 | 2 | 1.0634 | 1.0787 | 1.0655 | 11.047 | 10.951 | 10.924 | 1.12 |
| sKMD NLL | Diag NLL | Point NLL | KMD MSE | Pool MSE | sKMD MSE | Gain | |||
|---|---|---|---|---|---|---|---|---|---|
| 0.50 | 30 | 0 | 1.0506 | 1.0865 | 1.0512 | 2.089 | 2.089 | 2.089 | 0.03 |
| 0.50 | 30 | 1 | 1.0529 | 1.0902 | 1.0541 | 3.357 | 3.357 | 3.356 | 0.03 |
| 0.50 | 30 | 2 | 1.0734 | 1.0930 | 1.0771 | 11.656 | 11.656 | 11.645 | 0.09 |
| 0.50 | 120 | 0 | 1.0382 | 1.0824 | 1.0383 | 1.432 | 1.432 | 1.431 | 0.05 |
| 0.50 | 120 | 1 | 1.0414 | 1.0840 | 1.0418 | 2.667 | 2.667 | 2.664 | 0.09 |
| 0.50 | 120 | 2 | 1.0633 | 1.0888 | 1.0641 | 11.163 | 11.163 | 11.138 | 0.22 |
Appendix figures & tables4 assets
Supplementary material from the paper’s appendix.
Appendix
| Method | Heterogeneous | Low | Very high | Student | Null |
|---|---|---|---|---|---|
| Integrated Gaussian | 0.5826 | 0.5023 | 0.6912 | 0.6060 | 0.6937 |
| Selected signed | 0.5846 | 0.5011 | 0.9609 | 0.6081 | 0.6939 |
| Same-law point | 0.6511 | 0.5833 | 1.7004 | 0.6642 | 0.6954 |
| Signed OAS-QDA | 0.6454 | 0.6034 | 1.3576 | 0.6557 | 0.6936 |
| Signed quality interaction | 0.5846 | 0.5011 | 0.9609 | 0.6081 | 0.6936 |
| Energy interaction | 0.6937 | 0.6937 | 0.7847 | 0.7236 | 0.6937 |
| Method | Heterogeneous | Low | Very high | Student | Null |
|---|---|---|---|---|---|
| Integrated Gaussian | 0.29613 | 0.00249 | 0.90226 | 0.29229 | 0.29878 |
| Integrated scale mixture | 0.29996 | 0.00250 | 0.90502 | 0.28300 | 0.30056 |
| Point-weighted Gaussian | 0.29893 | 0.00250 | 0.92177 | 0.29533 | 0.29870 |
| Selected-signed weights | 0.29645 | 0.00249 | 0.91520 | 0.29297 | 0.29875 |
| Pooled full KMD | 0.29937 | 0.00250 | 0.90249 | 0.30071 | 0.29879 |
| Independent-mode KMD | 0.29916 | 0.00250 | 0.90207 | 0.30049 | 0.29867 |
| Fit events | Method | Continuous | High noise | Biased monitor |
|---|---|---|---|---|
| 600 | Integrated Gaussian | 1.03934 | 1.09599 | 1.04935 |
| 600 | Selected signed | 1.04273 | 1.22756 | 1.04872 |
| 600 | Continuous global point | 1.05992 | 1.58551 | 1.05998 |
| 600 | Quality-logit calibration | 1.04399 | 1.22794 | 1.05049 |
| 600 | Signed quality quadratic | 1.04368 | 1.21308 | 1.04820 |
| 600 | Energy quality polynomial | 1.06278 | 1.46583 | 1.06759 |
| Fit events | Reconstruction | Continuous | High noise | Biased monitor |
|---|---|---|---|---|
| 600 | Integrated regime | 0.44660 | 1.26067 | 0.48992 |
| 600 | Selected-signed weights | 0.44675 | 1.27209 | 0.48918 |
| 600 | Pooled full KMD | 0.44867 | 1.26087 | 0.48874 |
| 600 | Independent KMD | 0.45448 | 1.27021 | 0.49598 |
| 2400 | Integrated regime | 0.44467 | 1.25527 | 0.49124 |
| 2400 | Selected-signed weights | 0.44485 | 1.26189 | 0.49071 |
Conditional Distribution Estimation
Conditional distribution estimation models the full distribution of an outcome given covariates, enabling uncertainty-aware prediction and decisions.
See all 90 papers