ASRN: a learned-hash-table copy layer for LLMs with linear memory in sequence length
Mean-Disaster8380 · reddit · 2026-10-05
A brief [N] method announcement: ASRN (Adaptive Sparse Recurrence Network) is a copy layer for language models that uses learned hash tables to find earlier occurrences of the current context and copies what came next, with memory growing linearly in sequence length. The post gives only the one-line core mechanism, without paper details or results.
More from Research
- Stanford kundajelab releases MotifCompendium, GPU-accelerated motif analysis toolkit — anshulkundaje · 2026-10-05
- ibUMAP paper: coherent field evaluation speeds up UMAP up to 5.79x with better stability — leland_mcinnes · 2026-10-05
- Distilling big-RL models into small ones breeds overconfident agents without calibration RL — willcb · 2026-10-05
- DeepMind paper shows a single model can rank without an index, ending the results page — dejanseo · 2026-10-05
- New Paper IGP-Bench Teaches AI Agents When to Stop Chasing a Wrong Idea — _sathvikr · 2026-10-05
- Kissing number lower bound in 21 dimensions pushed to 30,779, possibly by AI agents — felpix_ · 2026-10-05