Daily report — 2026-09-30

Modern learning systems often acquire supervision at multiple resolutions, trading annotation cost against information content. We study cost-aware two-resolution learning, where expensive fine labels reveal a vector response and cheaper coarse labels reveal a scalar aggregate formed with unknown weights, while the target remains the full response.

Claimed: none extracted

Evidence: + cheap/closed-form ("closed-form"); + toy/synthetic experiments ("Synthetic"); + small/synthetic data ("Synthetic"); + public benchmarks ("benchmark"); + concrete method ("closed-form"); + formal statement ("bound"); + claims simplicity ("simple"); no public code found

heuristic

An agent that uses tools typically responds to what the user explicitly asks, yet completing the task may require information the user never requested. Work on proactive agents mainly studies whether and when an agent should act on its own, not what information it should pursue.

Claimed: none extracted

Evidence: + toy/synthetic experiments ("simulated"); + small/synthetic data ("simulated"); + public benchmarks ("benchmark"); - system/framework (many moving parts) ("agent that uses tools"); + code: https://github.com/dolev31/ProactiveInquirer; + keywords: retrieval; + 13 HF upvotes

heuristic

In Bayesian inference problems with non-Gaussian observation noise, the posterior is only as accurate as the noise density, and gradient-based samplers need that density and its gradient evaluable pointwise, whether from an explicit expression or from code, and without an inner solve. We propose Copula Active Subspaces (CAS) to represent this noise density.

Claimed: none extracted

Evidence: + cheap/closed-form ("closed form"); + concrete method ("closed form"); + formal statement ("bound"); + code: https://github.com/joshuawchen/copula_active_subspaces; + keywords: regularization

heuristic

Dense retrievers decide which documents reach users and the language models that use them, yet they are typically evaluated with neutral queries. We ask whether the identity signals that real users express in their queries---political ideology and dialect---bias what a retriever returns.

Claimed: none extracted

Evidence: + toy/synthetic experiments ("synthetic"); + small/synthetic data ("synthetic"); + code: https://github.com/Andrewtcr/bias-ret; + keywords: retrieval

heuristic

Mixture-of-Experts (MoE) layers are central to frontier language models, and recent architectures push toward more and smaller experts. In this regime, the standard linear router becomes a bottleneck: with $M$ experts and hidden dimension $h$, its per-token cost $Θ(Mh)$ dominates the MoE layer once $M$ is large.

Claimed: none extracted

Evidence: - pretraining ("pretraining"); + toy/synthetic experiments ("synthetic"); + small/synthetic data ("synthetic"); + public benchmarks ("benchmarks"); + formal statement ("We prove"); + claims simplicity ("simple"); + code: https://github.com/Matheart/MoRE_code; + keywords: kernel

heuristic

Visual-token compression is effective for improving the efficiency of vision-language models, but under extreme compression budgets, token pruning can break visual grounding while learned resamplers increase parameter count, attention cost, and training complexity. We revisit compression through a token parameterization lens, separating (i) basis transformation and structured truncation (retained

Claimed: accuracy 23.0; accuracy 95.2%; accuracy 84.2%; accuracy 36.0%

Evidence: + CPU-friendly ("lightweight"); + formal statement ("bound"); + claims simplicity ("straightforward"); + code: https://github.com/zrrraa/Braco; + 3 HF upvotes

heuristic

Private Evolution (PE) is a differentially private algorithm for synthetic data generation. While it can be viewed as a Wasserstein learning algorithm, it performs much better in practice than worst-case Wasserstein analyses would predict.

Claimed: none extracted

Evidence: + toy/synthetic experiments ("synthetic"); + small/synthetic data ("synthetic"); + explicit algorithm/pseudocode ("Algorithm 2"); + formal statement ("convergence"); + claims simplicity ("simple"); no public code found

heuristic

Reliable evaluation of large language models (LLMs) is essential for their development and deployment, yet is often costly, risky, and difficult to perform safely online. We study off-policy evaluation for LLMs, where limited human-labeled data from a behavior model are used to evaluate a newer target LLM.

Claimed: none extracted

Evidence: - LLM-based ("Large Language Model"); + toy/synthetic experiments ("synthetic"); + small/synthetic data ("synthetic"); + concrete method ("we propose a new"); + formal statement ("convergence"); + code: https://github.com/LinerXiang/OTROPE

heuristic

Graph-based retrieval-augmented generation supports multi-hop retrieval by organizing corpus information into graphs. However, existing relation-free graph retrieval methods rely primarily on query-sentence similarity to search for evidence.

Claimed: none extracted

Evidence: + public benchmarks ("benchmarks"); + concrete method ("we propose a simple"); + claims simplicity ("simple"); + code: https://github.com/Jacob-biu/NexusRAG; + keywords: retrieval

heuristic

Closed-form recourse moves a rejected user along the unit gradient $\hat g$ of the classifier score $f$ by the promised distance $d_p=|f(x)|/\|\nabla f(x)\|$, at which the linearized score reaches zero. We ask when this one-shot step succeeds and what additional model queries change.

Claimed: none extracted

Evidence: + cheap/closed-form ("Closed-form"); + small/synthetic data ("MNIST"); + concrete method ("Closed-form"); + formal statement ("bound"); no public code found

heuristic