MIRA maintains a persistent research record and uses an outer-loop meta-reasoner to curate the relevant evidence.
Separating research choices from execution improves long-horizon agent performance
MIRA learns at investigation boundaries, directing future work while leaving lengthy execution traces outside the policy optimization loop.
Big Tech
Ankur Samanta · Yonathan Efroni · Paul Sajda · Kaveh Hassani · Anirudh Goyal
Meta AI · Columbia University · Tel Aviv University
Research Digest··2 min read
Samanta et al.
Why this paper
From Meta AI and 2 others
In one line
Explicitly separating research allocation from execution lets agents learn which investigation to pursue next and improves long-horizon outcomes from delayed and proxy feedback.
What we could check
- ·No code link found
- ·No weights link found
- ·No dataset link found
- ·No compute details found
- ·No stated limitations found
- ·No benchmark numbers found
Observed from the paper text and links we have. Absence here means we did not find it, not that it does not exist.
§