Researchlong term memorylong contextassociative memorygoogle research
Google Research Debuts Titans And MIRAS Memory Framework
10.0
Google Research introduces two papers, Titans and MIRAS, proposing memory-driven sequence models to handle extremely long context. Titans uses a surprise metric, momentum, and adaptive forgetting to build a long-term memory module, while MIRAS offers a framework of four design choices for associative memory; evaluations show Titans scales beyond two million tokens and outperforms larger baselines, including GPT-4, on BABILong.
Key Points
- 1Introduce Titans architecture with surprise metric, momentum, and adaptive forgetting for sustained long-term memory
- 2Demonstrate Titans scales past 2 million tokens and outperforms larger baselines on BABILong
- 3Enable practitioners to design memory-driven sequence models via MIRAS’ four choices and test-time learning
Scoring Rationale
Strong novelty and broad scope with official Google Research validation and clear experimental evidence across 2M-token contexts.
Sources
Public references used for this report.
Practice interview problems based on real data
1,625 SQL & Python problems across 15 industry datasets — the exact type of data you work with.
Try 250 free problems
