🧪 Test?View on arXiv
MARCH: Scaling Recurrent Memory with Content-Routed State Anchors
Not provided in the abstract
memory-augmentedrecurrent networkslong-context retrievalattention mechanisms
2608.12435
Builder Relevance
6h ago80%
Abstract
MARCH introduces a network architecture that scales state-space models beyond fixed-size dimensions while maintaining computational efficiency over long sequences.
Reality Card
Core Claim
MARCH outperforms multiple linear attention variants in commonsense reasoning, LongBench, and in-context retrieval tasks after standard pretraining.
Method / Result
MARCH maintains a memory bank that grows with context length, providing a controllable trade-off between historical resolution and memory cost.
Limitations
The paper does not specify limitations or reproducibility concerns.