Reproduce “Top-$k$ Retrieval Leaves Memory Value Undetermined” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Arman Behnam
The authors introduce LEDGER, which tests whether retrieval logs distinguish memories’ Shapley values before estimating them, and varies retrieval choices to reduce nesting. Under the set-function model, fixed-size retrieval leaves each value undetermined. On HotpotQA, LEDGER determines 0.797 of memories versus 0.001 for deterministic top-k, with 2.6-point lower accuracy; real-data estimates assume d=1 and are not causal without exogenous logs.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.