# Bias Mitigation for Stereotype Queries on UTKFACE Race Source: [Bias Mitigation for Stereotype Queries on UTKFACE Race]() Collection index: [llms.txt]() Data updated: 2026-03-20T10:21:39.784Z Evaluation of bias mitigation techniques targeting stereotype queries specifically on the race attribute subset of the UTKFace dataset, measured using metrics like KL Divergence and MaxSkew. A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage. | Scope | Recorded value | | --- | --- | | Task | Bias Mitigation for Stereotype Queries | | Dataset (including any stated split) | UTKFACE Race | | Metric | KL Divergence | | Unit | SCALAR | | Direction | Lower is better | | Results loaded | 33 of 33 | | Coverage | All results currently returned for this benchmark | Evaluation settings may differ between results. The original papers and recorded configurations are needed to establish comparability. A shared metric name alone does not establish an identical protocol. ## Metrics - [KL Divergence](): Lower is better; unit: SCALAR. - [MaxSkew](): Lower is better; unit: SCALAR. ## Recorded results | Method | KL Divergence | Result date | Report paper | | --- | --- | --- | --- | | ZSDEBIAS | 0.027 | Not recorded | 69bcd4ab0636e5bf4afc2bf4 | | BENDSEMbi | 0.035 | Not recorded | 69bcd4ab0636e5bf4afc2bf4 | | SEMi | 0.039 | Not recorded | 69bcd4ab0636e5bf4afc2bf4 | | BEND-VLM | 0.041 | Not recorded | 68fa759e1e209131cea68fe9 | | BENDSEMbi | 0.042 | Not recorded | 69bcd4ab0636e5bf4afc2bf4 | | BEND-VLM | 0.047 | Not recorded | 68fa759e1e209131cea68fe9 | | SEMb | 0.084 | Not recorded | 69bcd4ab0636e5bf4afc2bf4 | | SEMbi | 0.085 | Not recorded | 69bcd4ab0636e5bf4afc2bf4 | | ZSDEBIAS | 0.091 | Not recorded | 69bcd4ab0636e5bf4afc2bf4 | | BENDVLM | 0.101 | Not recorded | 69bcd4ab0636e5bf4afc2bf4 | | Baseline CLIP | 0.107 | Not recorded | 68fa759e1e209131cea68fe9 | | SEMb | 0.107 | Not recorded | 69bcd4ab0636e5bf4afc2bf4 | | SEMbi | 0.107 | Not recorded | 69bcd4ab0636e5bf4afc2bf4 | | SEMi | 0.11 | Not recorded | 69bcd4ab0636e5bf4afc2bf4 | | Baseline CLIP | 0.114 | Not recorded | 68fa759e1e209131cea68fe9 | | BENDVLM | 0.126 | Not recorded | 69bcd4ab0636e5bf4afc2bf4 | | BASE CLIP | 0.127 | Not recorded | 69bcd4ab0636e5bf4afc2bf4 | | PRISM | 0.133 | Not recorded | 69bcd4ab0636e5bf4afc2bf4 | | PRISM | 0.134 | Not recorded | 69bcd4ab0636e5bf4afc2bf4 | | BASE CLIP | 0.152 | Not recorded | 69bcd4ab0636e5bf4afc2bf4 | | ROBOSHOT | 0.152 | Not recorded | 69bcd4ab0636e5bf4afc2bf4 | | DebiasCLIP | 0.158 | Not recorded | 68fa759e1e209131cea68fe9 | | Orth-Proj. | 0.182 | Not recorded | 68fa759e1e209131cea68fe9 | | Orth-Cal. | 0.196 | Not recorded | 68fa759e1e209131cea68fe9 | | ROBOSHOT | 0.206 | Not recorded | 69bcd4ab0636e5bf4afc2bf4 | | Orth-Cal. | 0.251 | Not recorded | 68fa759e1e209131cea68fe9 | | Orth-Proj. | 0.259 | Not recorded | 68fa759e1e209131cea68fe9 | | ORTH-CALI | 0.302 | Not recorded | 69bcd4ab0636e5bf4afc2bf4 | | ORTH-CALI | 0.307 | Not recorded | 69bcd4ab0636e5bf4afc2bf4 | | PRISM-MINI | 0.315 | Not recorded | 69bcd4ab0636e5bf4afc2bf4 | | ORTH-PROJ | 0.322 | Not recorded | 69bcd4ab0636e5bf4afc2bf4 | | PRISM-MINI | 0.336 | Not recorded | 69bcd4ab0636e5bf4afc2bf4 | | ORTH-PROJ | 0.34 | Not recorded | 69bcd4ab0636e5bf4afc2bf4 | ### ZSDEBIAS KL Divergence: 0.027. Result ID: 69bd1f9647f34ce924a7e1a7. [Reporting paper]() Recorded evaluation configuration: {"Backbone":"ResNet-50","Zero-shot":"true","Prompting Category":"Bias prompts only"}. This record does not assert independent reproduction. ### BENDSEMbi KL Divergence: 0.035. Result ID: 69bd1f9647f34ce924a7e1b2. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Backbone":"ResNet-101","Zero-shot":"false","Prompting Category":"Bias prompts + input-specific prompts + labeled images"}. This record does not assert independent reproduction. ### SEMi KL Divergence: 0.039. Result ID: 69bd1f9647f34ce924a7e19f. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Backbone":"ResNet-50","Zero-shot":"true","Prompting Category":"Bias-agnostic + input-specific prompts"}. This record does not assert independent reproduction. ### BEND-VLM KL Divergence: 0.041. Result ID: 6993f3423a6c31484919b8f6. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Backbone":"CLIP-VIT-B-P16"}. This record does not assert independent reproduction. ### BENDSEMbi KL Divergence: 0.042. Result ID: 69bd1f9647f34ce924a7e1b1. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Backbone":"ResNet-50","Zero-shot":"false","Prompting Category":"Bias prompts + input-specific prompts + labeled images"}. This record does not assert independent reproduction. ### BEND-VLM KL Divergence: 0.047. Result ID: 6993f3423a6c31484919b8fa. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Backbone":"CLIP-VIT-L-P14"}. This record does not assert independent reproduction. ### SEMb KL Divergence: 0.084. Result ID: 69bd1f9647f34ce924a7e1a6. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Backbone":"ResNet-101","Zero-shot":"true","Prompting Category":"Bias prompts only"}. This record does not assert independent reproduction. ### SEMbi KL Divergence: 0.085. Result ID: 69bd1f9647f34ce924a7e1ac. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Backbone":"ResNet-101","Zero-shot":"true","Prompting Category":"Bias prompts + input-specific prompts"}. This record does not assert independent reproduction. ### ZSDEBIAS KL Divergence: 0.091. Result ID: 69bd1f9647f34ce924a7e1a8. [Reporting paper]() Recorded evaluation configuration: {"Backbone":"ResNet-101","Zero-shot":"true","Prompting Category":"Bias prompts only"}. This record does not assert independent reproduction. ### BENDVLM KL Divergence: 0.101. Result ID: 69bd1f9647f34ce924a7e1af. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Backbone":"ResNet-50","Zero-shot":"false","Prompting Category":"Bias prompts + input-specific prompts + labeled images"}. This record does not assert independent reproduction. ### Baseline CLIP KL Divergence: 0.107. Result ID: 6993f3423a6c31484919b8f7. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Backbone":"CLIP-VIT-L-P14"}. This record does not assert independent reproduction. ### SEMb KL Divergence: 0.107. Result ID: 69bd1f9647f34ce924a7e1a5. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Backbone":"ResNet-50","Zero-shot":"true","Prompting Category":"Bias prompts only"}. This record does not assert independent reproduction. ### SEMbi KL Divergence: 0.107. Result ID: 69bd1f9647f34ce924a7e1ab. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Backbone":"ResNet-50","Zero-shot":"true","Prompting Category":"Bias prompts + input-specific prompts"}. This record does not assert independent reproduction. ### SEMi KL Divergence: 0.11. Result ID: 69bd1f9647f34ce924a7e1a0. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Backbone":"ResNet-101","Zero-shot":"true","Prompting Category":"Bias-agnostic + input-specific prompts"}. This record does not assert independent reproduction. ### Baseline CLIP KL Divergence: 0.114. Result ID: 6993f3423a6c31484919b8f2. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Backbone":"CLIP-VIT-B-P16"}. This record does not assert independent reproduction. ### BENDVLM KL Divergence: 0.126. Result ID: 69bd1f9647f34ce924a7e1b0. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Backbone":"ResNet-101","Zero-shot":"false","Prompting Category":"Bias prompts + input-specific prompts + labeled images"}. This record does not assert independent reproduction. ### BASE CLIP KL Divergence: 0.127. Result ID: 69bd1f9647f34ce924a7e19b. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Backbone":"ResNet-50","Zero-shot":"true"}. This record does not assert independent reproduction. ### PRISM KL Divergence: 0.133. Result ID: 69bd1f9647f34ce924a7e1ae. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Backbone":"ResNet-101","Zero-shot":"true","Prompting Category":"Bias prompts + input-specific prompts"}. This record does not assert independent reproduction. ### PRISM KL Divergence: 0.134. Result ID: 69bd1f9647f34ce924a7e1ad. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Backbone":"ResNet-50","Zero-shot":"true","Prompting Category":"Bias prompts + input-specific prompts"}. This record does not assert independent reproduction. ### BASE CLIP KL Divergence: 0.152. Result ID: 69bd1f9647f34ce924a7e19c. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Backbone":"ResNet-101","Zero-shot":"true"}. This record does not assert independent reproduction. ### ROBOSHOT KL Divergence: 0.152. Result ID: 69bd1f9647f34ce924a7e19d. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Backbone":"ResNet-50","Zero-shot":"true","Prompting Category":"Bias-agnostic + input-specific prompts"}. This record does not assert independent reproduction. ### DebiasCLIP KL Divergence: 0.158. Result ID: 6993f3423a6c31484919b8f5. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Backbone":"CLIP-VIT-B-P16"}. This record does not assert independent reproduction. ### Orth-Proj. KL Divergence: 0.182. Result ID: 6993f3423a6c31484919b8f8. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Backbone":"CLIP-VIT-L-P14"}. This record does not assert independent reproduction. ### Orth-Cal. KL Divergence: 0.196. Result ID: 6993f3423a6c31484919b8f9. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Backbone":"CLIP-VIT-L-P14"}. This record does not assert independent reproduction. ### ROBOSHOT KL Divergence: 0.206. Result ID: 69bd1f9647f34ce924a7e19e. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Backbone":"ResNet-101","Zero-shot":"true","Prompting Category":"Bias-agnostic + input-specific prompts"}. This record does not assert independent reproduction. ### Orth-Cal. KL Divergence: 0.251. Result ID: 6993f3423a6c31484919b8f4. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Backbone":"CLIP-VIT-B-P16"}. This record does not assert independent reproduction. ### Orth-Proj. KL Divergence: 0.259. Result ID: 6993f3423a6c31484919b8f3. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Backbone":"CLIP-VIT-B-P16"}. This record does not assert independent reproduction. ### ORTH-CALI KL Divergence: 0.302. Result ID: 69bd1f9647f34ce924a7e1aa. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Backbone":"ResNet-101","Zero-shot":"true","Prompting Category":"Bias prompts + input-specific prompts"}. This record does not assert independent reproduction. ### ORTH-CALI KL Divergence: 0.307. Result ID: 69bd1f9647f34ce924a7e1a9. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Backbone":"ResNet-50","Zero-shot":"true","Prompting Category":"Bias prompts + input-specific prompts"}. This record does not assert independent reproduction. ### PRISM-MINI KL Divergence: 0.315. Result ID: 69bd1f9647f34ce924a7e1a4. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Backbone":"ResNet-101","Zero-shot":"true","Prompting Category":"Bias prompts only"}. This record does not assert independent reproduction. ### ORTH-PROJ KL Divergence: 0.322. Result ID: 69bd1f9647f34ce924a7e1a2. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Backbone":"ResNet-101","Zero-shot":"true","Prompting Category":"Bias prompts only"}. This record does not assert independent reproduction. ### PRISM-MINI KL Divergence: 0.336. Result ID: 69bd1f9647f34ce924a7e1a3. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Backbone":"ResNet-50","Zero-shot":"true","Prompting Category":"Bias prompts only"}. This record does not assert independent reproduction. ### ORTH-PROJ KL Divergence: 0.34. Result ID: 69bd1f9647f34ce924a7e1a1. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Backbone":"ResNet-50","Zero-shot":"true","Prompting Category":"Bias prompts only"}. This record does not assert independent reproduction. Page 1 of 1. This is an index page, not the full collection.