List your company

Commonsense Question Answering on CSQA (Acc, Uintra, Uinter, Usys)

98.9AccuracyClean (no attack)
12.78835.14457.579.856Feb 6, 2026Mar 3, 2026Mar 28, 2026Apr 22, 2026May 17, 2026Jun 11, 2026Jul 7, 2026

Evaluation results

MethodLinks
Clean (no attack)Defense=—2026.0798.9———
AcMASDefense=Correction2026.0798.6———
UDPO# compromised agents=12026.0285.10.0980.0720.108
G-SafeguardDefense=Isolation2026.0778.9———
UDPO# compromised agents=22026.0276.40.1480.1120.162
UDPO# compromised agents=32026.0273.90.2150.1680.235
BlindGuardDefense=Isolation2026.0773.5———
RMAAC# compromised agents=12026.0269.50.2050.1620.275
LoraHubSetting=single-task, Number of runs=52026.0666.1———
EMMSetting=single-task, Number of runs=52026.0665.9———
Expert FusionSetting=single-task, Number of runs=52026.0665———
Best Single ExpertSetting=single-task, Number of runs=52026.0664.3———
MERGEvolveSetting=single-task, Number of runs=52026.0664.3———
Pack of LLMsSetting=single-task, Number of runs=52026.0663.3———
Standard MAD# compromised agents=12026.0263.20.310.2050.338
Model SwarmsSetting=single-task, Number of runs=52026.0663.2———
RMAAC# compromised agents=22026.0259.80.2580.2150.355
MAPPO# compromised agents=12026.02580.2150.1720.288
MAPPO# compromised agents=22026.0256.90.2750.2280.378
Standard MAD# compromised agents=22026.0251.60.3350.2750.442
RMAAC# compromised agents=32026.0248.60.3180.2780.438
MAPPO# compromised agents=32026.0244.80.3420.2950.472
Data MergeSetting=single-task, Number of runs=52026.0637.4———
Standard MAD# compromised agents=32026.0229.70.5120.5520.573
TIESSetting=single-task, Number of runs=52026.0628———
Attack-only (no defense)Defense=No2026.0716.1———