# paper-forge daily report — 2026-10-16

153 new papers screened; 10 analyzed in depth; 0 reproduction runs.

## Reproductions

- none today (reimplementation is opt-in: `forge daily --reimplement`)

## Top candidates

### [Attention over noisy examples: generalization and memorization in the exponential Hopfield network](https://arxiv.org/abs/2610.22191)
`2610.22191` · score 67.4 · analysis: heuristic · feasibility 7/10

In Hebbian learning from examples a Hopfield network of $N$ neurons is not supplied with the $K$ archetypes it has to retrieve but with $M$ blurred examples of each, of quality $r$, and whether it can infer them depends on the load $α=K/N$ (interference among them) and on the dataset entropy $ρ=(1-r^2)/(Mr^2)$ (how little the examples say about them). We transfer this protocol to the exponential Hopfield model, whose zero-temperature update is a step of softmax attention with the unlabelled examples as keys.

Rubric evidence: + cheap/closed-form ("closed form"); + small/synthetic data ("MNIST"); + concrete method ("closed form"); + formal statement ("theorem"); + code: https://github.com/francescoalemanno/hebb-attention; + keywords: monte carlo

### [Causal Integrated Gradients: Scalable Feature Attribution Along Causal Paths](https://arxiv.org/abs/2610.22126)
`2610.22126` · score 65.8 · analysis: heuristic · feasibility 7/10

Existing feature attribution methods that account for feature dependencies do not scale to high dimensions. We propose causal integrated gradients (CIG) and causal expected gradients (CEG), which extend integrated gradients to incorporate prior causal structure among input features while retaining its computational efficiency and, under conditions on the causal graph, its axiomatic guarantees.

Rubric evidence: + toy/synthetic experiments ("synthetic"); + small/synthetic data ("synthetic"); + public benchmarks ("benchmark"); + formal statement ("theorem"); + claims simplicity ("simple"); + code: https://github.com/sifrimlab/CausalIG; + keywords: ranking

### [Classical methods for modern applications: synthetic tabular data generation using lightweight discrete response continuous latent variable model](https://arxiv.org/abs/2610.21674)
`2610.21674` · score 62.1 · analysis: heuristic · feasibility 8/10

In this paper, we present a lightweight algorithm for modelling tabular data containing both continuous and discrete variables. The method uses empirical marginal distributions and a selected quantile function to transform the data into a continuous latent space, where a distribution of choice, e.g.

Rubric evidence: + toy/synthetic experiments ("synthetic"); + CPU-friendly ("lightweight"); + small/synthetic data ("synthetic"); + formal statement ("Bound"); + claims simplicity ("simple"); no public code found; + keywords: sampling, monte carlo

### [FDE-Gym: The Last Mile of AI Delivery](https://arxiv.org/abs/2610.22350)
`2610.22350` · score 61.1 · analysis: heuristic · feasibility 6/10

Existing agent benchmarks typically evaluate agents on well-specified tasks defined by issue reports, specifications, or task tickets. Real-world customer delivery is different: forward-deployed engineers (FDEs) must clarify ambiguous requirements, work in live systems, and uncover unstated constraints.

Rubric evidence: + toy/synthetic experiments ("simulated"); + small/synthetic data ("simulated"); + public benchmarks ("benchmarks"); + code: https://github.com/deeplethe/fde-gym; + 56 HF upvotes

### [Lean Graph Search for Formal Theorem Proving](https://arxiv.org/abs/2610.22339)
`2610.22339` · score 56.8 · analysis: heuristic · feasibility 6/10

Retrieving relevant lemmas is crucial for formal theorem proving, enabling provers to build machine-checkable proofs from established mathematical results. Recent retrieval systems match library declarations to a target theorem or its intermediate proof steps, but make limited use of multi-hop dependencies among declarations.

Claimed: recall 12.0% (abstract (regex extraction, unverified))
Rubric evidence: + public benchmarks ("benchmarks"); + formal statement ("Theorem"); + claims simplicity ("simple"); + code: https://github.com/PlusLabNLP/LeanGraphSearch; + keywords: retrieval, ranking

### [Robust Covariance Estimation for Entry-wise Contaminated and Partially Observed Data via Expectation Maximization Algorithm](https://arxiv.org/abs/2610.21709)
`2610.21709` · score 56.3 · analysis: heuristic · feasibility 7/10

Robust covariance estimation under entry-wise contamination and missing data is challenging in various signal processing applications. Under the Student-$t$ model, observation-level reweighting induces a uniform attenuation across samples, which can lead to cumulative information loss even when most coordinates remain informative.

Rubric evidence: + cheap/closed-form ("closed form"); + toy/synthetic experiments ("synthetic"); + small/synthetic data ("synthetic"); + concrete method ("closed form"); - system/framework (many moving parts) ("framework"); no public code found; + keywords: sampling

### [RecSoup: Efficient Model Merging Approach for Controllable Multi-Objective Recommendation](https://arxiv.org/abs/2610.21220)
`2610.21220` · score 55.3 · analysis: heuristic · feasibility 6/10

Recommender systems often need to balance multiple conflicting objectives. Controllable Multi-Objective Recommendation (CMOR) enables dynamic adjustment of objective trade-offs at inference time without retraining, adapting to shifting user or system requirements.

Rubric evidence: + CPU-friendly ("lightweight"); + public benchmarks ("benchmarks"); + formal statement ("we prove"); + code: https://github.com/35798642/RECSOUP.git; + keywords: ranking

### [Dimension-Free PAC-Bayes Generalization for Deep Neural Networks via Dirichlet Processes](https://arxiv.org/abs/2610.21340)
`2610.21340` · score 54.7 · analysis: heuristic · feasibility 8/10

Classical PAC-Bayesian generalization theory places prior distributions over Euclidean parameter space $Θ\subseteq \mathbb{R}^D$, causing relative entropy to scale with network dimension $D$. For modern overparameterized neural networks where $D \gg n$, this parameter-counting dependency yields vacuous generalization bounds ($> 1{,}000\%$).

Rubric evidence: + cheap/closed-form ("closed form"); + small/synthetic data ("MNIST"); + public benchmarks ("benchmark"); + concrete method ("closed form"); + formal statement ("Theorem"); no public code found

### [Sign Betting for Misspecified Regression: Variance Adaptivity and Optimal Approximation](https://arxiv.org/abs/2610.20903)
`2610.20903` · score 54.2 · analysis: heuristic · feasibility 7/10

We study bounded regression under model misspecification, evaluating a function $f$ by its $L_1(P_X)$ distance to the conditional mean $η(x)=\mathbb{E}[Y\mid X=x]$, rather than the standard $L_2(P_X)$. This $L_1$ metric is natural for downstream decision problems such as bandits and reinforcement learning.

Rubric evidence: + toy/synthetic experiments ("synthetic"); + small/synthetic data ("synthetic"); + formal statement ("bound"); + claims simplicity ("simple"); no public code found; + keywords: bandit

### [OmniVerifier-C1: Decomposed Graded Verifier for Rule-Governed Generative World Modeling](https://arxiv.org/abs/2610.22355)
`2610.22355` · score 53.7 · analysis: heuristic · feasibility 5/10

Generative visual verifiers play a crucial role in advancing multimodal intelligence. To enhance the robustness of visual verifiers in complex multimodal scenarios involving compositional requirements, we move beyond conventional binary verifiers that produce only True/False discrete judgments and introduce a verifier that automatically decomposes complex prompts or environmental rules into fine-grained components and assigns pointwise scores to provide a continuous and more informative critique signal.

Rubric evidence: + toy/synthetic experiments ("simulation"); + claims simplicity ("simple"); - system/framework (many moving parts) ("framework"); + code: https://github.com/Cominclip/OmniVerifier; + 25 HF upvotes

## All screened papers

| score | paper | compute | data | clarity | code | interest | analysis | laptop-feasible |
|---:|---|---:|---:|---:|---:|---:|---|---|
| 67.4 | [Attention over noisy examples: generalization and memorization in the exponential Hopfield network](https://arxiv.org/abs/2610.22191) `2610.22191` | 0.70 | 0.75 | 0.75 | 1.00 | 0.20 | heuristic | True |
| 65.8 | [Causal Integrated Gradients: Scalable Feature Attribution Along Causal Paths](https://arxiv.org/abs/2610.22126) `2610.22126` | 0.65 | 0.80 | 0.70 | 1.00 | 0.20 | heuristic | True |
| 62.1 | [Classical methods for modern applications: synthetic tabular data generation using lightweight discrete response continuous latent variable model](https://arxiv.org/abs/2610.21674) `2610.21674` | 0.80 | 0.75 | 0.70 | 0.00 | 0.40 | heuristic | True |
| 61.1 | [FDE-Gym: The Last Mile of AI Delivery](https://arxiv.org/abs/2610.22350) `2610.22350` | 0.65 | 0.80 | 0.40 | 1.00 | 0.30 | heuristic | True |
| 56.8 | [Lean Graph Search for Formal Theorem Proving](https://arxiv.org/abs/2610.22339) `2610.22339` | 0.50 | 0.45 | 0.70 | 1.00 | 0.40 | heuristic | True |
| 56.3 | [Robust Covariance Estimation for Entry-wise Contaminated and Partially Observed Data via Expectation Maximization Algorithm](https://arxiv.org/abs/2610.21709) `2610.21709` | 0.85 | 0.75 | 0.50 | 0.00 | 0.20 | heuristic | True |
| 55.3 | [RecSoup: Efficient Model Merging Approach for Controllable Multi-Objective Recommendation](https://arxiv.org/abs/2610.21220) `2610.21220` | 0.65 | 0.45 | 0.55 | 1.00 | 0.20 | heuristic | True |
| 54.7 | [Dimension-Free PAC-Bayes Generalization for Deep Neural Networks via Dirichlet Processes](https://arxiv.org/abs/2610.21340) `2610.21340` | 0.70 | 0.80 | 0.75 | 0.00 | 0.00 | heuristic | True |
| 54.2 | [Sign Betting for Misspecified Regression: Variance Adaptivity and Optimal Approximation](https://arxiv.org/abs/2610.20903) `2610.20903` | 0.65 | 0.75 | 0.70 | 0.00 | 0.20 | heuristic | True |
| 53.7 | [OmniVerifier-C1: Decomposed Graded Verifier for Rule-Governed Generative World Modeling](https://arxiv.org/abs/2610.22355) `2610.22355` | 0.65 | 0.40 | 0.45 | 1.00 | 0.30 | heuristic | True |
| 50.2 | [Memory as Weights: Internalizing Long-Term History for Streaming Videos](https://arxiv.org/abs/2610.22070) `2610.22070` | 0.65 | 0.40 | 0.30 | 1.00 | 0.28 | — | — |
| 50.0 | [Rethinking Successive Rejects Algorithms](https://arxiv.org/abs/2610.22051) `2610.22051` | 0.65 | 0.80 | 0.45 | 0.00 | 0.20 | — | — |
| 49.5 | [Idea2Prior: Auditable Paper Retrieval and Ranking from Research Ideas](https://arxiv.org/abs/2610.20857) `2610.20857` | 0.50 | 0.40 | 0.40 | 1.00 | 0.40 | — | — |
| 47.9 | [Intern-Decision: Efficient Multimodal Decision Making with Autoregressive Masked Language Modeling](https://arxiv.org/abs/2610.22131) `2610.22131` | 0.50 | 0.40 | 0.40 | 1.00 | 0.30 | — | — |
| 47.9 | [Each Reward to Its Relevant Tokens: Reward-Specific Token-Level Credit Assignment for Multi-Reward RL](https://arxiv.org/abs/2610.21439) `2610.21439` | 0.50 | 0.40 | 0.40 | 1.00 | 0.30 | — | — |
| 47.4 | [Patch where the text is new: a compression view of where byte-level models should spend compute](https://arxiv.org/abs/2610.22100) `2610.22100` | 0.40 | 0.75 | 0.40 | 1.00 | 0.00 | — | — |
| 46.8 | [Halve-Then-Greedy: Adaptive Sampling for Posterior Success with Near-Optimal Instance Dependence](https://arxiv.org/abs/2610.20997) `2610.20997` | 0.65 | 0.40 | 0.55 | 0.00 | 0.40 | — | — |
| 46.8 | [Flexible Doubly Robust Proximal Causal Inference for Censored Outcomes](https://arxiv.org/abs/2610.20893) `2610.20893` | 0.65 | 0.80 | 0.30 | 0.00 | 0.20 | — | — |
| 46.3 | [Effectiveness of Hidden-State Probes for Localizing Reasoning Failures](https://arxiv.org/abs/2610.22046) `2610.22046` | 0.50 | 0.40 | 0.40 | 1.00 | 0.20 | — | — |
| 45.8 | [Efficient Agnostic Learning of Univariate Densities Under Local Differential Privacy](https://arxiv.org/abs/2610.21033) `2610.21033` | 0.65 | 0.75 | 0.45 | 0.00 | 0.00 | — | — |
| 45.8 | [Mosaic: Multi-Session Agentic Reinforcement Learning](https://arxiv.org/abs/2610.22174) `2610.22174` | 0.50 | 0.40 | 0.30 | 1.00 | 0.30 | — | — |
| 45.8 | [AutoSFT: An Agentic System for Automated Data Research in Supervised Fine-Tuning](https://arxiv.org/abs/2610.22194) `2610.22194` | 0.50 | 0.40 | 0.30 | 1.00 | 0.30 | — | — |
| 45.5 | [Learning Self-Orchestration in Decentralized Agent Swarms](https://arxiv.org/abs/2610.22322) `2610.22322` | 0.50 | 0.40 | 0.30 | 1.00 | 0.28 | — | — |
| 45.3 | [Explore Once, Evaluate Many: Multiple Policy Evaluation for Stochastic Linear Bandits](https://arxiv.org/abs/2610.22252) `2610.22252` | 0.50 | 0.40 | 0.55 | 0.00 | 0.60 | — | — |
| 45.3 | [Beyond Downstream QA: Rethinking How We Evaluate Agent Memory](https://arxiv.org/abs/2610.22209) `2610.22209` | 0.50 | 0.45 | 0.30 | 1.00 | 0.20 | — | — |
| 45.3 | [On the Last-Iterate Convergence of Clipped Gradient Methods](https://arxiv.org/abs/2610.21026) `2610.21026` | 0.70 | 0.40 | 0.70 | 0.00 | 0.00 | — | — |
| 45.3 | [Acceleration and its Limits in Gaussian Variational Inference](https://arxiv.org/abs/2610.21020) `2610.21020` | 0.70 | 0.40 | 0.55 | 0.00 | 0.20 | — | — |
| 44.7 | [Spectral Indistinguishability and Learning-Curve Identifiability in Quadratic Networks](https://arxiv.org/abs/2610.20678) `2610.20678` | 0.65 | 0.75 | 0.40 | 0.00 | 0.00 | — | — |
| 43.7 | [EVA-Cap: Optimizing Audiovisual Video Captioning via Event-Centric Alignment](https://arxiv.org/abs/2610.22186) `2610.22186` | 0.50 | 0.45 | 0.30 | 1.00 | 0.10 | — | — |
| 43.7 | [Conditional Mean Flows: Few-Step Densities for Simulation-Based Inference](https://arxiv.org/abs/2610.21657) `2610.21657` | 0.65 | 0.40 | 0.40 | 0.00 | 0.40 | — | — |
| 43.2 | [REKI: Registering Knowledge in VLAs](https://arxiv.org/abs/2610.22178) `2610.22178` | 0.40 | 0.40 | 0.55 | 1.00 | 0.00 | — | — |
| 43.2 | [When Is Muon Optimal? Radial Fragility and Minimax Geometry](https://arxiv.org/abs/2610.21549) `2610.21549` | 0.70 | 0.40 | 0.45 | 0.00 | 0.20 | — | — |
| 42.6 | [SSR: Structural Signal Retriever for Scalable Agent Discovery in LLM-Orchestrated Multi-Agent Systems](https://arxiv.org/abs/2610.21113) `2610.21113` | 0.45 | 0.80 | 0.40 | 0.00 | 0.20 | — | — |
| 42.1 | [Two-Stage Score Estimation for Group-Specific Diffusion Models](https://arxiv.org/abs/2610.20979) `2610.20979` | 0.50 | 0.80 | 0.45 | 0.00 | 0.00 | — | — |
| 40.5 | [Complete or Sparse: A Tale of Two Identifiabilities](https://arxiv.org/abs/2610.20851) `2610.20851` | 0.45 | 0.80 | 0.45 | 0.00 | 0.00 | — | — |
| 39.5 | [Learning When to Think: Utility-Guided Adaptive Reasoning for Efficient Text Retrieval](https://arxiv.org/abs/2610.21030) `2610.21030` | 0.65 | 0.45 | 0.30 | 0.00 | 0.20 | — | — |
| 39.5 | [Librarian Agents: Training LLMs to Curate Long-Context Workspaces that Anticipate Future Queries](https://arxiv.org/abs/2610.22273) `2610.22273` | 0.45 | 0.75 | 0.30 | 0.00 | 0.20 | — | — |
| 39.5 | [MindEval2: Clinically Grounded Patients and Rubrics for Evaluating Language Models in Mental Health Support](https://arxiv.org/abs/2610.22314) `2610.22314` | 0.45 | 0.40 | 0.30 | 1.00 | 0.00 | — | — |
| 39.5 | [StochLOB: Heavy-Tailed Score Matching for Robust Limit Order Book Trend Prediction](https://arxiv.org/abs/2610.21081) `2610.21081` | 0.35 | 0.40 | 0.30 | 1.00 | 0.20 | — | — |
| 39.5 | [Trajectory-Level Speculation with Semantic Verification for Tool-Using Agents](https://arxiv.org/abs/2610.21463) `2610.21463` | 0.30 | 0.40 | 0.30 | 1.00 | 0.30 | — | — |
| 38.9 | [Diminishing Returns of Data in Sparse Autoencoders: Functional Behavior does not Guarantee Feature Reproducibility](https://arxiv.org/abs/2610.20680) `2610.20680` | 0.50 | 0.40 | 0.55 | 0.00 | 0.20 | — | — |
| 38.4 | [LIBERO-PeRM: Benchmarking Personalized Robotic Manipulation](https://arxiv.org/abs/2610.22372) `2610.22372` | 0.65 | 0.45 | 0.40 | 0.00 | 0.00 | — | — |
| 38.4 | [Time-partitioning Gaussian Process Mixture Model](https://arxiv.org/abs/2610.20860) `2610.20860` | 0.65 | 0.40 | 0.30 | 0.00 | 0.20 | — | — |
| 37.9 | [Training Reasoning Models to Generate Diverse Scientific Ideas](https://arxiv.org/abs/2610.22324) `2610.22324` | 0.30 | 0.45 | 0.40 | 1.00 | 0.00 | — | — |
| 37.9 | [Grid-Bench: Can LLMs Predict How Discrete Worlds Change?](https://arxiv.org/abs/2610.22312) `2610.22312` | 0.30 | 0.45 | 0.40 | 1.00 | 0.00 | — | — |
| 37.9 | [GACC-Bench: Benchmarking Goal-Adaptive Context Compression Ability across Large Language Models](https://arxiv.org/abs/2610.21893) `2610.21893` | 0.30 | 0.45 | 0.40 | 1.00 | 0.00 | — | — |
| 37.4 | [Asking Smarter Questions: Targeted Active Preference Learning for Underspecified Rewards](https://arxiv.org/abs/2610.22285) `2610.22285` | 0.65 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 37.4 | [Unfragmenting Nawatl NMT: Single-Variant Specialization for Masewalkopa](https://arxiv.org/abs/2610.22110) `2610.22110` | 0.65 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 37.4 | [Controlling the False Discovery Rate via Model-Agnostic Importance Measures](https://arxiv.org/abs/2610.21360) `2610.21360` | 0.65 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 36.8 | [Whole-Body Dexterity](https://arxiv.org/abs/2610.22184) `2610.22184` | 0.50 | 0.40 | 0.45 | 0.00 | 0.20 | — | — |
| 36.8 | [LCT: Losslessly Compressed Tensors with GPU friendly execution for Training](https://arxiv.org/abs/2610.22179) `2610.22179` | 0.50 | 0.40 | 0.30 | 0.00 | 0.40 | — | — |
| 36.8 | [ReviewPilot: A Multi-Agent System for Screening and Data Extraction in Systematic Reviews](https://arxiv.org/abs/2610.21042) `2610.21042` | 0.30 | 0.40 | 0.40 | 1.00 | 0.00 | — | — |
| 36.8 | [TransX Technical Report: Scaling Sequence Transduction for Recommendation](https://arxiv.org/abs/2610.20913) `2610.20913` | 0.50 | 0.40 | 0.30 | 0.00 | 0.40 | — | — |
| 36.8 | [The Tilted Sampling Problem: Optimism, Posterior Sampling, and Logarithmic Regret](https://arxiv.org/abs/2610.21908) `2610.21908` | 0.50 | 0.40 | 0.30 | 0.00 | 0.40 | — | — |
| 35.8 | [Learning Rational Strategies for Temporal Objectives in Multi-Agent Systems](https://arxiv.org/abs/2610.22287) `2610.22287` | 0.50 | 0.40 | 0.55 | 0.00 | 0.00 | — | — |
| 35.8 | [Sharp and Stable Label-Free Calibration Certificates from Logical Coherence](https://arxiv.org/abs/2610.21567) `2610.21567` | 0.50 | 0.40 | 0.55 | 0.00 | 0.00 | — | — |
| 35.8 | [Regret Lower Bound for Online Learning of Quantum States with Logarithmic Loss, with Applications to Quantum State Tomography and Confidence Sequences for Matrix Means](https://arxiv.org/abs/2610.21273) `2610.21273` | 0.50 | 0.40 | 0.55 | 0.00 | 0.00 | — | — |
| 35.8 | [Ranked Right, Sized Wrong: Ensemble Spread Fails Silently at Streamflow Extremes](https://arxiv.org/abs/2610.20947) `2610.20947` | 0.50 | 0.40 | 0.40 | 0.00 | 0.20 | — | — |
| 35.8 | [Selection-Safe Conformal Early Exit: Distribution-Free Coverage under Adaptive Depth](https://arxiv.org/abs/2610.20773) `2610.20773` | 0.50 | 0.40 | 0.55 | 0.00 | 0.00 | — | — |
| 35.8 | [PremiseSQL: Evaluating Premise Critique and Abstention in Text-to-SQL Systems](https://arxiv.org/abs/2610.22158) `2610.22158` | 0.30 | 0.45 | 0.30 | 1.00 | 0.00 | — | — |
| 35.8 | [Stochastic Gradient Descent: A Nonlinear Time Series Perspective](https://arxiv.org/abs/2610.20894) `2610.20894` | 0.70 | 0.40 | 0.25 | 0.00 | 0.00 | — | — |
| 35.8 | [InfoVis: Unified Slide Image Generation and Editing](https://arxiv.org/abs/2610.16784) `2610.16784` | 0.25 | 0.05 | 0.40 | 1.00 | 0.50 | — | — |
| 35.3 | [A Mixture-of-Experts Framework for Multi-Stream Quickest Change Detection](https://arxiv.org/abs/2610.20876) `2610.20876` | 0.65 | 0.40 | 0.30 | 0.00 | 0.00 | — | — |
| 35.3 | [Knowledge That Is Linear Before Fine-Tuning Generalizes After It](https://arxiv.org/abs/2610.22300) `2610.22300` | 0.25 | 0.40 | 0.40 | 1.00 | 0.00 | — | — |
| 35.3 | [Athanor: An Agent-Native System for Autonomous Post-Training Research](https://arxiv.org/abs/2610.21055) `2610.21055` | 0.50 | 0.40 | 0.30 | 0.00 | 0.30 | — | — |
| 35.3 | [One Bias Score, Two Factors: How Mitigations Affect Option Bias in Multiple-Choice Evaluation](https://arxiv.org/abs/2610.20993) `2610.20993` | 0.50 | 0.45 | 0.40 | 0.00 | 0.10 | — | — |
| 34.7 | [Lexical Reference Frames: Interpreting LM Residual Streams in Native Semantics](https://arxiv.org/abs/2610.22169) `2610.22169` | 0.50 | 0.40 | 0.00 | 1.00 | 0.00 | — | — |
| 34.7 | [EMO: A Closed-Loop Test-Time Computation Framework via Alternating Context Optimization and Logits Steering](https://arxiv.org/abs/2610.22160) `2610.22160` | 0.50 | 0.45 | 0.45 | 0.00 | 0.00 | — | — |
| 33.7 | [Backward Compatibility in Zero-Bit Neural Audio Watermarking](https://arxiv.org/abs/2610.22296) `2610.22296` | 0.50 | 0.40 | 0.30 | 0.00 | 0.20 | — | — |
| 33.7 | [Generalization Guarantees of Randomized Block Coordinate Descent for Nonconvex Learning](https://arxiv.org/abs/2610.22213) `2610.22213` | 0.70 | 0.40 | 0.15 | 0.00 | 0.00 | — | — |
| 33.7 | [Learning Speech Attribute Representations via Multi-Teacher Distillation](https://arxiv.org/abs/2610.21941) `2610.21941` | 0.50 | 0.40 | 0.30 | 0.00 | 0.20 | — | — |
| 33.7 | [CARA: Causally Guided Retrieval Agent for Question Answering](https://arxiv.org/abs/2610.22017) `2610.22017` | 0.50 | 0.40 | 0.30 | 0.00 | 0.20 | — | — |
| 33.7 | [SelectPFN: Amortized Bayesian Variable Selection for Sparse Nonlinear Regression with Prior-Data Fitted Networks](https://arxiv.org/abs/2610.21950) `2610.21950` | 0.20 | 0.75 | 0.55 | 0.00 | 0.00 | — | — |
| 33.7 | [Failure Is the Mother of Success: Recovering Supervision with Grounding Hindsight](https://arxiv.org/abs/2610.21018) `2610.21018` | 0.50 | 0.45 | 0.40 | 0.00 | 0.00 | — | — |
| 33.7 | [RoLaQ: Joint Rotation and Range Calibration for 2-Bit KV Caches in Multi-Head Latent Attention](https://arxiv.org/abs/2610.20686) `2610.20686` | 0.50 | 0.45 | 0.40 | 0.00 | 0.00 | — | — |
| 33.7 | [FlashMask-3: Efficient and Expressive Mask-Aware Distributed Attention](https://arxiv.org/abs/2610.17901) `2610.17901` | 0.00 | 0.40 | 0.40 | 1.00 | 0.40 | — | — |
| 32.6 | [Rethinking Routing Exploration in Modern Mixture-of-Experts](https://arxiv.org/abs/2610.22353) `2610.22353` | 0.50 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 32.6 | [Twice Upon a Time: Complementary Representations for Time-Series Reasoning](https://arxiv.org/abs/2610.22327) `2610.22327` | 0.50 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 32.6 | [How Much Physics Can Vision Learn from Audio? Probing and Distilling Damping Structure](https://arxiv.org/abs/2610.22221) `2610.22221` | 0.50 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 32.6 | [FutureLift: Lifting Trajectory Proposals into Bidirectional Future Graphs for Motion Forecasting](https://arxiv.org/abs/2610.22215) `2610.22215` | 0.50 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 32.6 | [Specification Gaming and a Constrained Reinforcement Learning Mitigation in a Cleaning Gridworld](https://arxiv.org/abs/2610.22152) `2610.22152` | 0.50 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 32.6 | [Beyond Fixed Acquisition Functions: Temporal Difference Active Learning for Perturbation Discovery](https://arxiv.org/abs/2610.22122) `2610.22122` | 0.50 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 32.6 | [Learning Changes the World and Then What Was Learned has Changed: Training and Deploying Neural Networks in the Control Loop](https://arxiv.org/abs/2610.22084) `2610.22084` | 0.50 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 32.6 | [The Right Byte First: How UTF-16 Helps Multilingual Boundary Prediction](https://arxiv.org/abs/2610.22071) `2610.22071` | 0.50 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 32.6 | [CLUE: Uncovering Social Context for Pragmatic Understanding](https://arxiv.org/abs/2610.22167) `2610.22167` | 0.50 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 32.6 | [Dynamic ordering in sequence generation in Transformers: experiments on low-resource Tagging scenarios](https://arxiv.org/abs/2610.21862) `2610.21862` | 0.50 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 32.6 | [Content associations with answer selection and engagement in supply chain community question answering](https://arxiv.org/abs/2610.21807) `2610.21807` | 0.50 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 32.6 | [Making every stage of medical device document assessment with large language models deterministic](https://arxiv.org/abs/2610.21300) `2610.21300` | 0.30 | 0.40 | 0.40 | 0.00 | 0.40 | — | — |
| 32.6 | [Provably Efficient Risk-Sensitive Offline Reinforcement Learning via Optimized Certainty Equivalents](https://arxiv.org/abs/2610.21870) `2610.21870` | 0.50 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 32.6 | [A One-Sided McDiarmid-Type Inequality for Heavy Tails via Reverse KL Tensorization](https://arxiv.org/abs/2610.21103) `2610.21103` | 0.50 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 32.1 | [Score Mismatch and Set-Constrained Sampling in Zero-Shot Conditional Discrete Diffusion](https://arxiv.org/abs/2610.21061) `2610.21061` | 0.35 | 0.40 | 0.45 | 0.00 | 0.20 | — | — |
| 32.1 | [Waypoint-OPD: Transferring RL-Induced Policy Shifts via Gated On-Policy Distillation](https://arxiv.org/abs/2610.21601) `2610.21601` | 0.00 | 0.40 | 0.40 | 1.00 | 0.30 | — | — |
| 32.1 | [Beyond Policy Rollouts: Embedding-Space Directional Exploration with Anchored Noise for LLM RLVR](https://arxiv.org/abs/2610.16105) `2610.16105` | 0.30 | 0.45 | 0.40 | 0.00 | 0.30 | — | — |
| 31.6 | [Shortcut Sensitivity in Agent-Safety Evaluations](https://arxiv.org/abs/2610.22243) `2610.22243` | 0.50 | 0.45 | 0.30 | 0.00 | 0.00 | — | — |
| 31.6 | [SurvICL: a Foundation Model for in-context Survival Analysis](https://arxiv.org/abs/2610.22242) `2610.22242` | 0.20 | 0.80 | 0.40 | 0.00 | 0.00 | — | — |
| 31.6 | [HTN-Orchestrate: A Hierarchical Task Network Framework for Multi-Agent Orchestration](https://arxiv.org/abs/2610.22195) `2610.22195` | 0.50 | 0.45 | 0.30 | 0.00 | 0.00 | — | — |
| 31.6 | [WEAVE: Weaving Real-World Code Evolution into Dependency-Structured Tasks for Benchmarking and Training Code Agents](https://arxiv.org/abs/2610.22039) `2610.22039` | 0.50 | 0.45 | 0.30 | 0.00 | 0.00 | — | — |
| 31.6 | [P2TPO: Structure-Aligned Page-to-Token Credit Assignment for Full-Page Document Parsing](https://arxiv.org/abs/2610.21979) `2610.21979` | 0.50 | 0.45 | 0.30 | 0.00 | 0.00 | — | — |
| 31.6 | [Retrieved but Not Read: Understanding Why Search-Augmented LLMs Fail to Use Available Evidence](https://arxiv.org/abs/2610.21731) `2610.21731` | 0.30 | 0.00 | 0.40 | 1.00 | 0.20 | — | — |
| 31.6 | [Virtual Head Attention](https://arxiv.org/abs/2610.17959) `2610.17959` | 0.00 | 0.45 | 0.40 | 1.00 | 0.20 | — | — |
| 31.1 | [Query-Efficient Rejection Sampling for Parallel Masked Diffusion](https://arxiv.org/abs/2610.22238) `2610.22238` | 0.35 | 0.40 | 0.40 | 0.00 | 0.20 | — | — |
| 30.5 | [Robust Reinforcement Learning for Dynamic Treatment Regimes under Model and Measurement Uncertainty](https://arxiv.org/abs/2610.22336) `2610.22336` | 0.50 | 0.40 | 0.30 | 0.00 | 0.00 | — | — |
| 30.5 | [Beyond the Dominant Label: Recovering Coexisting Lesion Evidence with Multiple Instance Learning](https://arxiv.org/abs/2610.22091) `2610.22091` | 0.50 | 0.40 | 0.30 | 0.00 | 0.00 | — | — |
| 30.5 | [ABNAT: Attention-Based Noise-Aware Training for Robust Transformers on Analog Compute-in-Memory Systems](https://arxiv.org/abs/2610.22065) `2610.22065` | 0.50 | 0.40 | 0.30 | 0.00 | 0.00 | — | — |
| 30.5 | [Selection Is Not Learning: Choosing Model Futures](https://arxiv.org/abs/2610.22029) `2610.22029` | 0.50 | 0.40 | 0.30 | 0.00 | 0.00 | — | — |
| 30.5 | [Aligning LLMs with Sparse Retrieval via Adaptive Sparse Supervision](https://arxiv.org/abs/2610.21877) `2610.21877` | 0.30 | 0.40 | 0.30 | 0.00 | 0.40 | — | — |
| 30.5 | [Designing Experiments for External Decision Makers](https://arxiv.org/abs/2610.22032) `2610.22032` | 0.50 | 0.40 | 0.30 | 0.00 | 0.00 | — | — |
| 30.5 | [Why Shallow Neural Networks Learn Sinusoidal Components First: An Orthogonal Polynomial Dynamics Framework](https://arxiv.org/abs/2610.21964) `2610.21964` | 0.50 | 0.40 | 0.30 | 0.00 | 0.00 | — | — |
| 30.5 | [Learning Topology-Aware Representations for Mechanics-Relevant TPMS Metamaterials](https://arxiv.org/abs/2610.20684) `2610.20684` | 0.50 | 0.40 | 0.30 | 0.00 | 0.00 | — | — |
| 30.0 | [MLExpert: Towards a Generalist Agentic Model for Autonomous Machine Learning Engineering](https://arxiv.org/abs/2610.22344) `2610.22344` | 0.00 | 0.40 | 0.30 | 1.00 | 0.30 | — | — |
| 29.5 | [BitSparse: Compact Activation Storage for ReLU Feed-Forward Networks](https://arxiv.org/abs/2610.22256) `2610.22256` | 0.30 | 0.40 | 0.40 | 0.00 | 0.20 | — | — |
| 29.5 | [Two Sample Sizes in Test-Time LLM Ensembling: Buffered ERM and Minimax Theory](https://arxiv.org/abs/2610.21363) `2610.21363` | 0.30 | 0.40 | 0.55 | 0.00 | 0.00 | — | — |
| 29.5 | [CleanRM: Reward Modeling from Noisy Feedback with a Few Clean Feedback for RLHF](https://arxiv.org/abs/2610.21138) `2610.21138` | 0.30 | 0.40 | 0.55 | 0.00 | 0.00 | — | — |
| 29.5 | [When the KV Cache Is Slower Than the GPU: Scaling Remote KV Tiers for Context-Parallel Prefill](https://arxiv.org/abs/2610.21500) `2610.21500` | 0.30 | 0.40 | 0.40 | 0.00 | 0.20 | — | — |
| 29.3 | [H-DDCM: Hadamard Codebooks for Fast Generative Image and Video Compression](https://arxiv.org/abs/2610.18125) `2610.18125` | 0.35 | 0.45 | 0.30 | 0.00 | 0.16 | — | — |
| 28.9 | [Multi-Target Few-Shot Prompting with Diffusion Language Models](https://arxiv.org/abs/2610.22294) `2610.22294` | 0.35 | 0.45 | 0.40 | 0.00 | 0.00 | — | — |
| 27.9 | [TIPS: A Tiny Inference-time Policy for Scaling in Diffusion Language Models](https://arxiv.org/abs/2610.22099) `2610.22099` | 0.35 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 27.4 | [PathFM: Fast Generative Forecasting of Multivariate Trajectories](https://arxiv.org/abs/2610.22358) `2610.22358` | 0.30 | 0.45 | 0.40 | 0.00 | 0.00 | — | — |
| 27.4 | [Learning to Steer LLM Agents and Multi-Agent Systems](https://arxiv.org/abs/2610.22316) `2610.22316` | 0.30 | 0.40 | 0.30 | 0.00 | 0.20 | — | — |
| 27.4 | [AI Consilium: a Benchmark of LLM Clinical Multidisciplinary Second Opinions](https://arxiv.org/abs/2610.22298) `2610.22298` | 0.30 | 0.45 | 0.40 | 0.00 | 0.00 | — | — |
| 27.4 | [TEM: Temporal State Tracking Memory for Large Language Model Agents](https://arxiv.org/abs/2610.21900) `2610.21900` | 0.30 | 0.40 | 0.30 | 0.00 | 0.20 | — | — |
| 27.4 | [VoteGEO-Bench: Measuring the Impact of Generative Engine Optimization on Electoral Visibility](https://arxiv.org/abs/2610.21330) `2610.21330` | 0.30 | 0.45 | 0.40 | 0.00 | 0.00 | — | — |
| 27.4 | [TRACER: Trajectory-Aware Tool Retrieval for LLM Agents](https://arxiv.org/abs/2610.21267) `2610.21267` | 0.30 | 0.40 | 0.30 | 0.00 | 0.20 | — | — |
| 27.4 | [Market Incentives for AI Safety Investment](https://arxiv.org/abs/2610.21607) `2610.21607` | 0.30 | 0.45 | 0.40 | 0.00 | 0.00 | — | — |
| 26.3 | [GeoState: Learning View-Decoupled Geometry States for 3D World Modeling](https://arxiv.org/abs/2610.22343) `2610.22343` | 0.30 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 26.3 | [Not X, but When: Diagnosing a Mandarin LLM Writing Tic through Contextual Appropriateness](https://arxiv.org/abs/2610.22249) `2610.22249` | 0.30 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 26.3 | [Change the Number, Change the Answer: Assessing Rating Scale Sensitivity in LLM-Based Text Scoring](https://arxiv.org/abs/2610.22187) `2610.22187` | 0.30 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 26.3 | [Believed but Not Said: Measuring Hidden Belief Collapse in Multi-Agent LLM Discussions](https://arxiv.org/abs/2610.22138) `2610.22138` | 0.30 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 26.3 | [Explaining the Roles of Experts in Mixture-of-Experts LLMs for Personalization](https://arxiv.org/abs/2610.22127) `2610.22127` | 0.30 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 26.3 | [Documenting Trofinetide Treatment Phenotypes in Rett Syndrome with a Dual-LLM Consensus Pipeline](https://arxiv.org/abs/2610.22062) `2610.22062` | 0.30 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 26.3 | [Name-Bias in Large Language Models](https://arxiv.org/abs/2610.22053) `2610.22053` | 0.30 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 26.3 | [N-Gram Memories at Two Bits: Quantizing Lookup Tables in Large Language Models](https://arxiv.org/abs/2610.21810) `2610.21810` | 0.30 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 25.8 | [Topology-Inspired Fine-Tuning for Wind-Field Super-Resolution](https://arxiv.org/abs/2610.22262) `2610.22262` | 0.25 | 0.45 | 0.40 | 0.00 | 0.00 | — | — |
| 25.8 | [Can AI Forecast the Extremes? Evaluating Probabilistic Weather Models on Tropical Cyclones](https://arxiv.org/abs/2610.22077) `2610.22077` | 0.25 | 0.45 | 0.40 | 0.00 | 0.00 | — | — |
| 25.3 | [Architecting for Hardware Diversity: Meta's End-to-End Co-Design for High-Performance AI Training on AMD GPUs](https://arxiv.org/abs/2610.22230) `2610.22230` | 0.50 | 0.00 | 0.30 | 0.00 | 0.20 | — | — |
| 25.3 | [Momentum, Commentary Tone, and Earnings Surprise in LLM Financial Forecasts](https://arxiv.org/abs/2610.22043) `2610.22043` | 0.00 | 0.80 | 0.40 | 0.00 | 0.00 | — | — |
| 25.3 | [Beyond Context Length: Characterising the Difficulty of Long-Context Reasoning](https://arxiv.org/abs/2610.22067) `2610.22067` | 0.50 | 0.05 | 0.40 | 0.00 | 0.00 | — | — |
| 24.7 | [Towards Continual Learning: Mitigating Forgetting With Subspace Tuning](https://arxiv.org/abs/2610.22293) `2610.22293` | 0.05 | 0.40 | 0.55 | 0.00 | 0.20 | — | — |
| 24.7 | [In-Context Learning Before Specialisation: Theory of Learning in a Solvable Topic Model](https://arxiv.org/abs/2610.21901) `2610.21901` | 0.25 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 23.2 | [Priming Base Models for Agentic Post-Training: An Open Midtraining Recipe](https://arxiv.org/abs/2610.22019) `2610.22019` | 0.00 | 0.80 | 0.30 | 0.00 | 0.00 | — | — |
| 22.1 | [AI-Driven, Timing-Aware Power Optimization](https://arxiv.org/abs/2610.18890) `2610.18890` | 0.50 | 0.00 | 0.30 | 0.00 | 0.00 | — | — |
| 21.1 | [Baselines and Score Centering in Policy Gradient Estimation](https://arxiv.org/abs/2610.22263) `2610.22263` | 0.30 | 0.40 | 0.00 | 0.00 | 0.20 | — | — |
| 17.9 | [CAT-alyzing Honesty: Continuous Adversarial Training for Transferable LLM Honesty](https://arxiv.org/abs/2610.22292) `2610.22292` | 0.00 | 0.45 | 0.40 | 0.00 | 0.00 | — | — |
| 17.9 | [Training-Free On-Policy Distillation](https://arxiv.org/abs/2610.22236) `2610.22236` | 0.00 | 0.45 | 0.40 | 0.00 | 0.00 | — | — |
| 16.8 | [The importance of tensor handling and implementation of physics-informed DeepONets](https://arxiv.org/abs/2610.22157) `2610.22157` | 0.00 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 16.8 | [Evaluation Awareness Beyond English: Do LLMs Act Differently When Tested in Indic Languages?](https://arxiv.org/abs/2610.22333) `2610.22333` | 0.00 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 15.8 | [Language Is a Security Surface for Agent Guards](https://arxiv.org/abs/2610.22244) `2610.22244` | 0.00 | 0.45 | 0.30 | 0.00 | 0.00 | — | — |
| 15.8 | [Hy-Memory-30B-A3B: Attributed Reinforcement Learning for Long-Term Memory Writing](https://arxiv.org/abs/2610.22196) `2610.22196` | 0.00 | 0.45 | 0.30 | 0.00 | 0.00 | — | — |
| 14.7 | [OS-Helix: Unifying GUI Interaction and Code Execution in Native Computer-Using Agents](https://arxiv.org/abs/2610.22234) `2610.22234` | 0.00 | 0.40 | 0.30 | 0.00 | 0.00 | — | — |
| 14.7 | [Scene Reconstruction Using Ubiquitous Wireless Signals via Diffusion Posterior Inference Over physics-Grounded Representations](https://arxiv.org/abs/2610.22120) `2610.22120` | 0.00 | 0.40 | 0.30 | 0.00 | 0.00 | — | — |
| 14.7 | [RoleQ: Role-Aligned Scales for Converting Language Models to Ternary](https://arxiv.org/abs/2610.21808) `2610.21808` | 0.00 | 0.40 | 0.30 | 0.00 | 0.00 | — | — |
| 14.7 | [ANTE: Anticipate, Then Execute -- Learning Internal World Models for Coding Agents from Verified Beliefs](https://arxiv.org/abs/2610.15368) `2610.15368` | 0.00 | 0.40 | 0.30 | 0.00 | 0.00 | — | — |
| 9.5 | [Multi-SFT: Permissively-Licensed Multilingual Instruction-Following](https://arxiv.org/abs/2610.22086) `2610.22086` | 0.00 | 0.05 | 0.40 | 0.00 | 0.00 | — | — |

Scores: weighted mean of rubric criteria (0–1 each) × 100; see README for the rubric.