MIRAS
Summary
MIRAS tracks the associative-memory framework for designing sequence models through attentional bias, retention, and online memory learning choices.
Role In The Wiki
Use this page as the object card for the model or method. The source page carries the evidence details, limitations, and agenda mapping.
Relation To Foundation TSFM Agenda
Use the source-level agenda mapping in miras-2025 rather than duplicating verdict rows here. At the entity level, this page should stay as the object card; source pages carry slot-level evidence, limitations, and missing pieces.
Evidence
Overlap Notes
MIRAS is the comparison frame for memory mechanisms rather than one narrow model family. It connects Titans, ATLAS, retention gates, attentional bias, and online memory learning. It should also be cross-read with ARMT when the question is associative key/value memory, and with Mamba when the question is whether state should be an explicit memory object or an implicit recurrent hidden state.