List your company
ResearchFormerly WizWand

Uncertainty Estimation on SciQA

0.8269AUROCEigenscore
0.4257720.5299110.634050.738189May 25, 2025Jul 15, 2025Sep 5, 2025Oct 27, 2025Dec 18, 2025Feb 8, 2026Apr 1, 2026
Updated 7mo ago

Evaluation results

MethodLinks
EigenscoreBackbone=Llama-3.1-8B, Evaluation Setup=TAC2026.040.82690.0981—-3.08
EntropyBackbone=Llama-3.1-8B, Evaluation Setup=TAC2026.040.82430.0638—-27.26
MSPBackbone=Llama-3.1-8B, Evaluation Setup=TAC2026.040.82390.0649—-40.35
PerplexityBackbone=Llama-3.1-8B, Evaluation Setup=TAC2026.040.81570.0991—-34.38
EntropyBackbone=Llama-3.2-3B, Evaluation Setup=TAC2026.040.76280.1556—-19.49
MSPBackbone=Llama-3.2-3B, Evaluation Setup=TAC2026.040.75790.1323—-34.73
SAR-sTarget Model=LLaMA-3-8B-Instruct, Setting=+Corrector2025.050.75720.36180.76—
SARTarget Model=LLaMA-3-8B-Instruct, Setting=+Corrector2025.050.75190.2561.31—
SAR-sTarget Model=LLaMA-3-8B-Instruct, Setting=Vanilla2025.050.74960.3854——
SEBackbone=Llama-3.1-8B, Evaluation Setup=TAC2026.040.74630.0365—-54.78
EnergyBackbone=Llama-3.2-3B, Evaluation Setup=TAC2026.040.74090.1384—-1.71
SARTarget Model=LLaMA-3-8B-Instruct, Setting=Vanilla2025.050.73880.2897——
SAR-tTarget Model=LLaMA-3-8B-Instruct, Setting=+Corrector2025.050.73870.26811.61—
SETarget Model=LLaMA-3-8B-Instruct, Setting=+Corrector2025.050.72930.25231.34—
SAR-tTarget Model=LLaMA-3-8B-Instruct, Setting=Vanilla2025.050.72260.3037——
PerplexityBackbone=Llama-3.2-3B, Evaluation Setup=TAC2026.040.72250.087—-41.77
VCTarget Model=LLaMA-3-8B-Instruct, Setting=+Corrector2025.050.72150.19473.93—
SETarget Model=LLaMA-3-8B-Instruct, Setting=Vanilla2025.050.71590.3054——
LN-PETarget Model=LLaMA-3-8B-Instruct, Setting=+Corrector2025.050.71560.23762.08—
P(True)Target Model=LLaMA-3-8B-Instruct, Setting=+Corrector2025.050.71410.31925.78—
CoE-CBackbone=Llama-3.2-3B, Evaluation Setup=TAC2026.040.70360.0728—-10.33
LN-PETarget Model=LLaMA-3-8B-Instruct, Setting=Vanilla2025.050.69480.2938——
VCBackbone=Llama-3.1-8B, Evaluation Setup=TAC2026.040.68670.042—-53.42
VCTarget Model=LLaMA-3-8B-Instruct, Setting=Vanilla2025.050.68220.3188——
PETarget Model=LLaMA-3-8B-Instruct, Setting=+Corrector2025.050.67980.34071.44—
EigenscoreBackbone=Llama-3.2-3B, Evaluation Setup=TAC2026.040.66940.072—-13.37
CoE-CBackbone=Llama-3.1-8B, Evaluation Setup=TAC2026.040.66790.062—-24.84
PETarget Model=LLaMA-3-8B-Instruct, Setting=Vanilla2025.050.66540.4067——
VCBackbone=Llama-3.2-3B, Evaluation Setup=TAC2026.040.66410.1402—-33.71
SETarget Model=OPT-6.7B, Setting=+Corrector2025.050.66150.42231.63—
P(True)Backbone=Llama-3.1-8B, Evaluation Setup=TAC2026.040.6590.0418—-49.81
P(True)Target Model=LLaMA-3-8B-Instruct, Setting=Vanilla2025.050.65630.3456——
LSTarget Model=LLaMA-3-8B-Instruct, Setting=+Corrector2025.050.65380.181911.71—
EnergyBackbone=Llama-3.1-8B, Evaluation Setup=TAC2026.040.65140.0428—-13.36
SETarget Model=OPT-6.7B, Setting=Vanilla2025.050.64520.5266——
LN-PETarget Model=OPT-6.7B, Setting=+Corrector2025.050.64230.34683.35—
Semantic EntropyBackbone=Llama-3.2-3B, Evaluation Setup=TAC2026.040.64070.0616—-59.87
CoE-RBackbone=Llama-3.1-8B, Evaluation Setup=TAC2026.040.63980.0419—-16.46
SAR-tTarget Model=OPT-6.7B, Setting=+Corrector2025.050.63750.44193.74—
SARTarget Model=OPT-6.7B, Setting=+Corrector2025.050.62720.38992.68—
P(True)Backbone=Llama-3.2-3B, Evaluation Setup=TAC2026.040.610.0233—-58.72
LN-PETarget Model=OPT-6.7B, Setting=Vanilla2025.050.60880.4952——
CoE-RBackbone=Llama-3.2-3B, Evaluation Setup=TAC2026.040.60770.0499—-35.05
SARTarget Model=OPT-6.7B, Setting=Vanilla2025.050.60040.494——
SAR-tTarget Model=OPT-6.7B, Setting=Vanilla2025.050.60.4533——
P(True)Target Model=OPT-6.7B, Setting=+Corrector2025.050.59490.34523.37—
VCTarget Model=OPT-6.7B, Setting=+Corrector2025.050.56550.38994.86—
P(True)Target Model=OPT-6.7B, Setting=Vanilla2025.050.56120.5879——
PETarget Model=OPT-6.7B, Setting=+Corrector2025.050.56020.36925.95—
SAR-sTarget Model=OPT-6.7B, Setting=+Corrector2025.050.54150.34830.95—
LSTarget Model=LLaMA-3-8B-Instruct, Setting=Vanilla2025.050.53670.3864——
SAR-sTarget Model=OPT-6.7B, Setting=Vanilla2025.050.5320.7621——
VCTarget Model=OPT-6.7B, Setting=Vanilla2025.050.51690.6265——
PETarget Model=OPT-6.7B, Setting=Vanilla2025.050.50070.6205——
LSTarget Model=OPT-6.7B, Setting=+Corrector2025.050.4940.32785.29—
LSTarget Model=OPT-6.7B, Setting=Vanilla2025.050.44120.7638——