Reward Modeling on HelpSteer (test)
0.077MAEILDE
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| ILDEMethod Category=Heuristic Methods2026.05 | 0.077 | 0.066 | 0.275 | — | |
| SelectiveRMMethod Category=Ours2026.05 | 0.087 | 0.063 | 0.308 | — | |
| CCRMethod Category=Statistically Consistent Methods2026.05 | 0.123 | 0.076 | 0.166 | — | |
| CSGNMethod Category=Statistically Consistent Methods2026.05 | 0.156 | 0.07 | 0.222 | — | |
| Robust DivideMixMethod Category=Heuristic Methods2026.05 | 0.162 | 0.067 | 0.263 | — | |
| HOCMethod Category=Statistically Consistent Methods2026.05 | 0.164 | 0.078 | 0.138 | — | |
| SelectMixMethod Category=Heuristic Methods2026.05 | 0.204 | 0.066 | 0.27 | — | |
| F-correctionMethod Category=Statistically Consistent Methods2026.05 | 0.205 | 0.08 | 0.116 | — | |
| ϵ-SoftmaxMethod Category=Heuristic Methods2026.05 | 0.208 | 0.068 | 0.251 | — | |
| LabelWaveMethod Category=Heuristic Methods2026.05 | 0.216 | 0.069 | 0.241 | — | |
| NLSMethod Category=Heuristic Methods2026.05 | 0.227 | 0.073 | 0.195 | — | |
| CNLCUMethod Category=Heuristic Methods2026.05 | 0.228 | 0.074 | 0.178 | — | |
| ROBOTMethod Category=Statistically Consistent Methods2026.05 | 0.23 | 0.073 | 0.191 | — | |
| kMEIDTMMethod Category=Statistically Consistent Methods2026.05 | 0.233 | 0.075 | 0.175 | — | |
| DRLearning Category=Debiased learning methods2026.03 | 0.2428 | — | 0.0902 | 0.4434 | |
| Co-TeachingMethod Category=Heuristic Methods2026.05 | 0.246 | 0.08 | 0.119 | — | |
| CDRMethod Category=Heuristic Methods2026.05 | 0.252 | 0.08 | 0.121 | — | |
| NaiveMethod Category=Statistically Consistent Methods2026.05 | 0.258 | 0.083 | 0.087 | — | |
| SelectMixLearning Category=PU learning methods2026.03 | 0.2624 | — | 0.279 | 0.3947 | |
| NNPULearning Category=PU learning methods2026.03 | 0.2688 | — | 0.2794 | 0.3946 | |
| CausalRM-IPSScenario=1, rho01=0.2, rho10=0.12026.03 | 0.27 | 0.156 | 0.277 | — | |
| SelectMixScenario=2, rho01=0.1, rho10=0.2, Method Category=Denoise-based Methods2026.03 | 0.27 | 0.175 | 0.19 | — | |
| UPULearning Category=PU learning methods2026.03 | 0.2729 | — | 0.2551 | 0.4012 | |
| SelectMixScenario=1, rho01=0.2, rho10=0.1, Method Category=Denoise-based Methods2026.03 | 0.274 | 0.177 | 0.179 | — | |
| F-correctionScenario=2, rho01=0.1, rho10=0.2, Method Category=Denoise-based Methods2026.03 | 0.276 | 0.185 | 0.143 | — | |
| ILDELearning Category=PU learning methods2026.03 | 0.2798 | — | 0.2314 | 0.4075 | |
| Robust DivideMixScenario=2, rho01=0.1, rho10=0.2, Method Category=Denoise-based Methods2026.03 | 0.281 | 0.177 | 0.181 | — | |
| ImplicitRM2026.03 | 0.2856 | — | 0.3114 | 0.2919 | |
| F-correctionScenario=1, rho01=0.2, rho10=0.1, Method Category=Denoise-based Methods2026.03 | 0.286 | 0.189 | 0.124 | — | |
| CausalRM-IPSScenario=2, rho01=0.1, rho10=0.22026.03 | 0.297 | 0.154 | 0.287 | — | |
| IPSScenario=2, rho01=0.1, rho10=0.2, Method Category=Debias-based Methods2026.03 | 0.3 | 0.181 | 0.163 | — | |
| Robust DivideMixScenario=1, rho01=0.2, rho10=0.1, Method Category=Denoise-based Methods2026.03 | 0.302 | 0.178 | 0.175 | — | |
| CoDisScenario=2, rho01=0.1, rho10=0.2, Method Category=Denoise-based Methods2026.03 | 0.303 | 0.182 | 0.155 | — | |
| CausalRM-DRScenario=2, rho01=0.1, rho10=0.22026.03 | 0.304 | 0.151 | 0.302 | — | |
| Co-TeachingScenario=2, rho01=0.1, rho10=0.2, Method Category=Denoise-based Methods2026.03 | 0.306 | 0.185 | 0.144 | — | |
| NaiveScenario=2, rho01=0.1, rho10=0.2, Method Category=Debias-based Methods2026.03 | 0.308 | 0.186 | 0.138 | — | |
| MTDRScenario=2, rho01=0.1, rho10=0.2, Method Category=Debias-based Methods2026.03 | 0.309 | 0.176 | 0.186 | — | |
| SDRScenario=2, rho01=0.1, rho10=0.2, Method Category=Debias-based Methods2026.03 | 0.318 | 0.174 | 0.193 | — | |
| CausalRM-DRScenario=1, rho01=0.2, rho10=0.12026.03 | 0.32 | 0.155 | 0.283 | — | |
| ILDEScenario=2, rho01=0.1, rho10=0.2, Method Category=Denoise-based Methods2026.03 | 0.32 | 0.175 | 0.191 | — | |
| NaiveScenario=1, rho01=0.2, rho10=0.1, Method Category=Debias-based Methods2026.03 | 0.327 | 0.197 | 0.09 | — | |
| MTIPSScenario=2, rho01=0.1, rho10=0.2, Method Category=Debias-based Methods2026.03 | 0.327 | 0.18 | 0.169 | — | |
| LabelWaveScenario=2, rho01=0.1, rho10=0.2, Method Category=Denoise-based Methods2026.03 | 0.329 | 0.181 | 0.161 | — | |
| CoDisScenario=1, rho01=0.2, rho10=0.1, Method Category=Denoise-based Methods2026.03 | 0.331 | 0.184 | 0.149 | — | |
| CUBPRLearning Category=PU learning methods2026.03 | 0.334 | — | 0.212 | 0.4126 | |
| IPSScenario=1, rho01=0.2, rho10=0.1, Method Category=Debias-based Methods2026.03 | 0.335 | 0.187 | 0.136 | — | |
| UPLLearning Category=PU learning methods2026.03 | 0.335 | — | 0.2251 | 0.4092 | |
| MTDRScenario=1, rho01=0.2, rho10=0.1, Method Category=Debias-based Methods2026.03 | 0.337 | 0.176 | 0.184 | — | |
| Co-TeachingScenario=1, rho01=0.2, rho10=0.1, Method Category=Denoise-based Methods2026.03 | 0.337 | 0.188 | 0.13 | — | |
| MTIPSScenario=1, rho01=0.2, rho10=0.1, Method Category=Debias-based Methods2026.03 | 0.344 | 0.182 | 0.157 | — | |
| LAGAMLearning Category=PU learning methods2026.03 | 0.3461 | — | 0.2321 | 0.4074 | |
| CVIBScenario=2, rho01=0.1, rho10=0.2, Method Category=Debias-based Methods2026.03 | 0.348 | 0.179 | 0.17 | — | |
| DRScenario=2, rho01=0.1, rho10=0.2, Method Category=Debias-based Methods2026.03 | 0.348 | 0.177 | 0.179 | — | |
| SDRScenario=1, rho01=0.2, rho10=0.1, Method Category=Debias-based Methods2026.03 | 0.352 | 0.175 | 0.192 | — | |
| DRScenario=1, rho01=0.2, rho10=0.1, Method Category=Debias-based Methods2026.03 | 0.353 | 0.179 | 0.171 | — | |
| CVIBScenario=1, rho01=0.2, rho10=0.1, Method Category=Debias-based Methods2026.03 | 0.362 | 0.18 | 0.167 | — | |
| ILDEScenario=1, rho01=0.2, rho10=0.1, Method Category=Denoise-based Methods2026.03 | 0.362 | 0.177 | 0.183 | — | |
| IPSLearning Category=Debiased learning methods2026.03 | 0.3785 | — | 0.0541 | 0.4498 | |
| NaiveLearning Category=Debiased learning methods2026.03 | 0.3907 | — | 0.1179 | 0.4366 | |
| UBPRLearning Category=PU learning methods2026.03 | 0.3944 | — | 0.1044 | 0.4399 | |
| SDRLearning Category=Debiased learning methods2026.03 | 0.3981 | — | 0.1794 | 0.4211 | |
| MTIPSLearning Category=Debiased learning methods2026.03 | 0.3999 | — | 0.1772 | 0.4217 | |
| MTDRLearning Category=Debiased learning methods2026.03 | 0.4057 | — | 0.1291 | 0.4338 | |
| LabelWaveScenario=1, rho01=0.2, rho10=0.1, Method Category=Denoise-based Methods2026.03 | 0.407 | 0.182 | 0.158 | — | |
| BPRLearning Category=PU learning methods2026.03 | 0.4314 | — | 0.0859 | 0.4444 |