Action Judge
The judging module in the AEWM framework that classifies each action of an agent into one of three categories: critical progress, effective exploration, or invalid repetition. It is used to decide whether a step should be retained or rewritten.
Timeline (last 90 days)
AEWM's Action Judge component classifies each step into three categories: critical progress (Critical), effective exploration (Exploratory), and ineffective repetition (Noisy).
On the Action Judge benchmark, AEWM achieves a macro-F1 of 70.5%, 10.6 percentage points higher than the strongest competing method.
All Facts (2)
AEWM's Action Judge component classifies each step into three categories: critical progress (Critical), effective exploration (Exploratory), and ineffective repetition (Noisy).
50%UnverifiedOn the Action Judge benchmark, AEWM achieves a macro-F1 of 70.5%, 10.6 percentage points higher than the strongest competing method.
50%