Aegis automates creation of a 9,533-trajectory error dataset for multi-agent LLM systems via adaptive fault injection, supporting SFT, RL, and contrastive training that yields models competitive with much larger proprietary ones.
For each agent 's response, critically evaluate its actions against the error definitions
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.RO 1years
2025 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Aegis: Automated Error Generation and Attribution for Multi-Agent Systems
Aegis automates creation of a 9,533-trajectory error dataset for multi-agent LLM systems via adaptive fault injection, supporting SFT, RL, and contrastive training that yields models competitive with much larger proprietary ones.