A fixed-shape secure multi-party kernel for post-alignment steps ran 21 to 30 times faster than a sorting variant under a LAN, and the speedup reversed on a high-latency link
AspireAI community · arXiv
A recurring step in two-party collaborative workflows combines a private candidate frontier with a partner's compatible evidence, scores the resulting candidates and retains only a small frontier, and direct execution can expose intermediate candidates, local scores or unauthorized evidence provenance. BlindStep is a fixed-shape MPC-online kernel for that post-alignment step between two semi-honest parties, providing a per-invocation simulation-based guarantee in the F-align hybrid model where the kernel leaks only public setup parameters, fixed execution-shape metadata, a global overflow bit and policy-authorized disclosures. The guarantee explicitly excludes malicious security and multi-invocation composition, and treats frontend leakage and offline preprocessing separately. Within that kernel boundary the authors implement six semi2k realizations in MP-SPDZ, crossing two sharing styles and three top-K strategies: H-variants compare over arithmetic shares, while B-variants convert packed ranking values to Boolean shares before selection. Under a LAN, B-Queue is 6 to 7 times faster and uses about 40 times less communication than H-Queue at N = 256 and 2048, and against H-Sort on the same backend B-Queue achieves speedups of 21 and 30 times with communication reductions of 204 and 290 times at N = 2048 and 8192. The limitation is the network regime: under the evaluated N = 256, 100 ms RTT WAN profile H-Sort becomes fastest, showing that round complexity can dominate backend choice for this kernel.
Read the story and explore more from the AspireAI community.
Two sparse-autoencoder analyses have resisted scaling: matching semantically similar features across layers, and compressing large feature circuits into interpretable supernodes. This paper argues both are the same underlying problem, estimating semantic distances between SAE features that sit on different activation manifolds. Its representation treats each feature not as a single decoder vector but as an activation-weighted distribution over the hidden states expressing it, projects those distributions into a shared reference space, and compares them with Wasserstein distance. The representation is proved invariant to activation rescaling and stable under perturbation, and recovers true matches under finite-sample margin conditions. Empirically it outperformed decoder-vector and LLM-based baselines and captured subtle functional distinctions between related features, and it compressed large feature circuits into interpretable supernodes automatically. The limitation is that this is interpretability tooling: the paper demonstrates better feature matching and circuit compression, not that downstream interpretability claims become more accurate.