# paper-forge daily report — 2026-10-01

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

## Reproductions

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

## Top candidates

### [BayesNDE: Bayesian Generative Modeling for Neural Density Estimation](https://arxiv.org/abs/2609.39843)
`2609.39843` · score 65.8 · analysis: heuristic · feasibility 7/10

Density estimation is a fundamental problem in statistics and machine learning. In this work, we introduce BayesNDE, a neural density estimator based on Bayesian generative modeling.

Rubric evidence: + toy/synthetic experiments ("synthetic"); + small/synthetic data ("synthetic"); + concrete method ("We introduce a new"); + formal statement ("convergence"); + code: https://github.com/liuq-lab/BayesNDE; + keywords: sampling

### [Understanding Multimodality in Generative Behavioral Cloning](https://arxiv.org/abs/2605.22493)
`2605.22493` · score 62.6 · analysis: heuristic · feasibility 7/10

Behavioral cloning becomes challenging when the same observation admits several valid actions. We study how generative behavioral-cloning policies represent such multimodal expert behavior and identify different bottlenecks across model parameterizations.

Rubric evidence: + toy/synthetic experiments ("synthetic"); + small/synthetic data ("synthetic"); + public benchmarks ("benchmarks"); + formal statement ("bound"); + code: https://github.com/Lorenzo-Mazza/VersatIL; + keywords: regularization

### [Adaptive mixture variational inference for spike-and-slab regression](https://arxiv.org/abs/2609.38656)
`2609.38656` · score 60.5 · analysis: heuristic · feasibility 8/10

Correlated predictors can support competing sparse explanations with similar predictions, making joint uncertainty about variable inclusion difficult to capture with mean-field approximations. We develop an adaptive fitting procedure for mixtures of product distributions in Gaussian regression with a point-mass spike-and-slab prior.

Rubric evidence: + toy/synthetic experiments ("simulated"); + small/synthetic data ("simulated"); + explicit algorithm/pseudocode ("Algorithm 1"); + formal statement ("theorem"); + claims simplicity ("simple"); no public code found; + keywords: variational

### [Amortized ratio-estimation importance sampling and localized simulation-based calibration for intractable likelihoods](https://arxiv.org/abs/2609.39712)
`2609.39712` · score 55.3 · analysis: heuristic · feasibility 7/10

We consider simulation-based Bayesian inference (SBI) for the parameters of models with intractable likelihoods but tractable forward simulation. Building on Gaussian mixtures-of-experts surrogates, as a first contribution we develop a sequential procedure for posterior estimation in which progressively localized, data informed, conditional density approximations are used as proposal distributions.

Rubric evidence: + toy/synthetic experiments ("simulation"); + small/synthetic data ("simulated"); + concrete method ("closed-form"); + claims simplicity ("simple"); no public code found; + keywords: sampling

### [Scalable Cox Regression via Grouped Risk Sets and Sharper LogSumExp Rates](https://arxiv.org/abs/2609.40120)
`2609.40120` · score 54.2 · analysis: heuristic · feasibility 7/10

Motivated by the computational challenges of large-scale Cox regression, we study stochastic minimization of LogSumExp objectives over large sets. Mini-batch normalizer estimates generally yield biased gradients.

Claimed: error rate 1.0 (abstract (regex extraction, unverified))
Rubric evidence: + cheap/closed-form ("convex"); + toy/synthetic experiments ("synthetic"); + small/synthetic data ("synthetic"); + formal statement ("we prove"); no public code found

### [Removing Timing Shortcuts Improves Non-Invasive Brain-to-Text](https://arxiv.org/abs/2609.40359)
`2609.40359` · score 51.1 · analysis: heuristic · feasibility 6/10

We find that major reported improvements in decoding words from non-invasive brain recordings are largely reproducible without any brain data. In the influential work of d'Ascoli et al.

Claimed: error rate 36.6% (abstract (regex extraction, unverified))
Rubric evidence: - pretraining ("pretrained"); - LLM-based ("LLM"); + toy/synthetic experiments ("synthetic"); + small/synthetic data ("synthetic"); + public benchmarks ("benchmark"); + concrete method ("We introduce a simple"); + claims simplicity ("simple"); + code: https://github.com/neural-processing-lab/SimpleB2T; + 1 HF upvotes

### [Safety of Latent Communication in Multi-Agent Systems](https://arxiv.org/abs/2609.39788)
`2609.39788` · score 50.5 · analysis: heuristic · feasibility 5/10

Latent communication enables multi-agent systems to exchange information directly in internal representation space, reducing the token, computation, and latency overhead of text-based communication. To this end, lightweight trainable links are introduced to map the sender's representations into the receiver's input space.

Rubric evidence: + CPU-friendly ("lightweight"); + public benchmarks ("benchmarks"); + code: https://github.com/Muhammad-Huzaifaa/latent-safety; + 1 HF upvotes

### [When the Label Ignores the Request: Auditing Policy-Selected Targets in Synthetic Conversational Music Recommendation](https://arxiv.org/abs/2609.39696)
`2609.39696` · score 50.0 · analysis: heuristic · feasibility 6/10

Synthetic dialogues generated by LLM pipelines now serve as complete conversational-recommendation benchmarks: an LLM listener talks to an LLM recommender, and the track logged next in the conversation becomes the official label for each turn. These policy-selected labels make large-scale evaluation reproducible, but they are proxies for what the simulated user asked.

Claimed: nDCG@20 43.0 (abstract (regex extraction, unverified))
Rubric evidence: - LLM-based ("LLM"); + toy/synthetic experiments ("Synthetic"); + small/synthetic data ("Synthetic"); + public benchmarks ("benchmarks"); + code: https://github.com/Sanjeev-S/recsys2026-request-audit

### [Accelerated Algorithm for Sparse Regularized Partial Optimal Transport](https://arxiv.org/abs/2609.40075)
`2609.40075` · score 48.4 · analysis: heuristic · feasibility 6/10

Partial Optimal Transport (POT) extends the classical optimal transport problem by relaxing the strict mass conservation constraint, enabling its use in a wide range of real-world applications. In many of these settings, sparse transport plans are preferred for their interpretability and computational benefits.

Rubric evidence: + cheap/closed-form ("convex"); + public benchmarks ("benchmarks"); + concrete method ("we propose a new"); + formal statement ("convergence"); + claims simplicity ("simple"); - system/framework (many moving parts) ("framework"); no public code found

### [DuoOPD: Learning from Joint Teacher-Student Outcomes for Multi-Task On-Policy Distillation](https://arxiv.org/abs/2609.33711)
`2609.33711` · score 46.8 · analysis: heuristic · feasibility 4/10

On-policy distillation (OPD) trains a student on its own responses with token-level feedback from a stronger teacher, yet the teacher can fail on questions the student already answers correctly, and how often each model succeeds varies across tasks. OPD ignores these outcomes and, on average, pushes down even the student's correct responses; gating feedback by student correctness fixes the direction but uses the teacher in the same way whether or not it succeeded.

Claimed: accuracy 2.58 (abstract (regex extraction, unverified))
Rubric evidence: + code: https://github.com/YongYuanDeAo/DuoOPD; + 4 HF upvotes

## All screened papers

| score | paper | compute | data | clarity | code | interest | analysis | laptop-feasible |
|---:|---|---:|---:|---:|---:|---:|---|---|
| 65.8 | [BayesNDE: Bayesian Generative Modeling for Neural Density Estimation](https://arxiv.org/abs/2609.39843) `2609.39843` | 0.65 | 0.75 | 0.75 | 1.00 | 0.20 | heuristic | True |
| 62.6 | [Understanding Multimodality in Generative Behavioral Cloning](https://arxiv.org/abs/2605.22493) `2605.22493` | 0.65 | 0.80 | 0.55 | 1.00 | 0.20 | heuristic | True |
| 60.5 | [Adaptive mixture variational inference for spike-and-slab regression](https://arxiv.org/abs/2609.38656) `2609.38656` | 0.65 | 0.75 | 1.00 | 0.00 | 0.20 | heuristic | True |
| 55.3 | [Amortized ratio-estimation importance sampling and localized simulation-based calibration for intractable likelihoods](https://arxiv.org/abs/2609.39712) `2609.39712` | 0.65 | 0.75 | 0.75 | 0.00 | 0.20 | heuristic | True |
| 54.2 | [Scalable Cox Regression via Grouped Risk Sets and Sharper LogSumExp Rates](https://arxiv.org/abs/2609.40120) `2609.40120` | 0.85 | 0.75 | 0.55 | 0.00 | 0.00 | heuristic | True |
| 51.1 | [Removing Timing Shortcuts Improves Non-Invasive Brain-to-Text](https://arxiv.org/abs/2609.40359) `2609.40359` | 0.20 | 0.80 | 0.75 | 1.00 | 0.10 | heuristic | False |
| 50.5 | [Safety of Latent Communication in Multi-Agent Systems](https://arxiv.org/abs/2609.39788) `2609.39788` | 0.65 | 0.45 | 0.40 | 1.00 | 0.10 | heuristic | True |
| 50.0 | [When the Label Ignores the Request: Auditing Policy-Selected Targets in Synthetic Conversational Music Recommendation](https://arxiv.org/abs/2609.39696) `2609.39696` | 0.45 | 0.80 | 0.40 | 1.00 | 0.00 | heuristic | False |
| 48.4 | [Accelerated Algorithm for Sparse Regularized Partial Optimal Transport](https://arxiv.org/abs/2609.40075) `2609.40075` | 0.70 | 0.45 | 0.80 | 0.00 | 0.00 | heuristic | True |
| 46.8 | [DuoOPD: Learning from Joint Teacher-Student Outcomes for Multi-Task On-Policy Distillation](https://arxiv.org/abs/2609.33711) `2609.33711` | 0.50 | 0.40 | 0.40 | 1.00 | 0.23 | heuristic | True |
| 46.7 | [Not Every Token Is Worth Distilling: Selective Supervision for Direct-OPD](https://arxiv.org/abs/2609.29142) `2609.29142` | 0.50 | 0.45 | 0.40 | 1.00 | 0.16 | — | — |
| 46.3 | [Learning Multiresolution Relevance for Hierarchical Generative Retrieval](https://arxiv.org/abs/2609.39312) `2609.39312` | 0.50 | 0.40 | 0.40 | 1.00 | 0.20 | — | — |
| 45.3 | [PTNO: Training Neural Operators with Noisy Monte Carlo Estimates for Particle Transport Problems](https://arxiv.org/abs/2609.40090) `2609.40090` | 0.80 | 0.40 | 0.40 | 0.00 | 0.20 | — | — |
| 45.3 | [Amortized Data Borrowing with Exchangeability-Aware Neural Posterior Estimation](https://arxiv.org/abs/2609.38902) `2609.38902` | 0.40 | 0.75 | 0.30 | 1.00 | 0.00 | — | — |
| 45.3 | [Near-Linear Accuracy Bounds for Moreau--Yosida Unadjusted Langevin Sampling](https://arxiv.org/abs/2609.40193) `2609.40193` | 0.70 | 0.40 | 0.55 | 0.00 | 0.20 | — | — |
| 45.3 | [Grokking through the Lens of Minimum-Norm Interpolation](https://arxiv.org/abs/2609.38453) `2609.38453` | 0.70 | 0.40 | 0.55 | 0.00 | 0.20 | — | — |
| 44.7 | [EvoDuet: Bilevel Co-Evolution of Web Searching and Task Solving for Scientific Discovery](https://arxiv.org/abs/2609.40340) `2609.40340` | 0.30 | 0.40 | 0.40 | 1.00 | 0.50 | — | — |
| 44.7 | [RoPE at the End of Its Rope? Theory, Diagnosis, and Mitigation of Long-Context Failures](https://arxiv.org/abs/2609.39929) `2609.39929` | 0.65 | 0.45 | 0.55 | 0.00 | 0.20 | — | — |
| 44.2 | [Residual Trajectory Distillation for Generative Retrieval](https://arxiv.org/abs/2609.39319) `2609.39319` | 0.50 | 0.40 | 0.30 | 1.00 | 0.20 | — | — |
| 44.2 | [On the Relaxation of Conditional Independence Assumption for Image Segmentation](https://arxiv.org/abs/2609.38930) `2609.38930` | 0.50 | 0.45 | 0.40 | 1.00 | 0.00 | — | — |
| 43.7 | [Partial identification with entropy regularized optimal transport](https://arxiv.org/abs/2609.40156) `2609.40156` | 0.65 | 0.40 | 0.55 | 0.00 | 0.20 | — | — |
| 43.7 | [A Dynamical Theory of LoRA in Continual Learning](https://arxiv.org/abs/2609.39367) `2609.39367` | 0.65 | 0.80 | 0.30 | 0.00 | 0.00 | — | — |
| 43.2 | [Proximal Balancing for Causal Effect Estimation under Unmeasured Confounding](https://arxiv.org/abs/2609.40051) `2609.40051` | 0.50 | 0.40 | 0.40 | 1.00 | 0.00 | — | — |
| 43.2 | [RouteRec: Behavior-Guided Sparse Routing for Sequential Recommendation](https://arxiv.org/abs/2609.39007) `2609.39007` | 0.50 | 0.40 | 0.40 | 1.00 | 0.00 | — | — |
| 42.6 | [BatSLAM 2.0: Sequence-Verified Sonar Place Recognition in a Robust Pose Graph](https://arxiv.org/abs/2609.40085) `2609.40085` | 0.65 | 0.75 | 0.30 | 0.00 | 0.00 | — | — |
| 42.6 | [Gromov-Wasserstein Distillation for Inductive Multi-View Embedding](https://arxiv.org/abs/2609.40047) `2609.40047` | 0.65 | 0.75 | 0.30 | 0.00 | 0.00 | — | — |
| 42.6 | [PhantomEnvironments: Training LLM Agents in Fictional Worlds](https://arxiv.org/abs/2609.40221) `2609.40221` | 0.45 | 0.80 | 0.55 | 0.00 | 0.00 | — | — |
| 42.6 | [Linguistic Loopholes in LLM Unlearning: From a 174-Language Benchmark to Coverage-Aware Unlearning](https://arxiv.org/abs/2609.40286) `2609.40286` | 0.45 | 0.80 | 0.55 | 0.00 | 0.00 | — | — |
| 42.6 | [OPSRD: On-Policy Self-Role Distillation](https://arxiv.org/abs/2609.39884) `2609.39884` | 0.45 | 0.45 | 0.40 | 1.00 | 0.00 | — | — |
| 42.6 | [SkillSeek: Revisiting Agent Skill Retrieval at Marketplace Scale](https://arxiv.org/abs/2609.38822) `2609.38822` | 0.30 | 0.45 | 0.40 | 1.00 | 0.30 | — | — |
| 42.6 | [Re-ranking and Late Interaction Drive Retrieval Quality: A Controlled Comparison of RAG Strategies for Scientific Question Answering](https://arxiv.org/abs/2609.38473) `2609.38473` | 0.45 | 0.75 | 0.30 | 0.00 | 0.40 | — | — |
| 42.6 | [False Frontiers: Diagnosing and Mitigating Co-Cheating in Self-Evolving Search Agents](https://arxiv.org/abs/2609.39102) `2609.39102` | 0.50 | 0.45 | 0.60 | 0.00 | 0.30 | — | — |
| 42.1 | [Robust and Learned Online Matching in Growing Trees](https://arxiv.org/abs/2609.40077) `2609.40077` | 0.70 | 0.40 | 0.55 | 0.00 | 0.00 | — | — |
| 42.1 | [The Geometry of Randomized Smoothing on Feasible Sets](https://arxiv.org/abs/2609.39497) `2609.39497` | 0.70 | 0.40 | 0.55 | 0.00 | 0.00 | — | — |
| 42.1 | [Bandits with Multiple Optimal Arms: Minimax Regret and Non-Adaptivit](https://arxiv.org/abs/2609.38659) `2609.38659` | 0.50 | 0.40 | 0.55 | 0.00 | 0.40 | — | — |
| 42.1 | [Lower Bounds for Linear-Oracle Online Learning](https://arxiv.org/abs/2609.38375) `2609.38375` | 0.70 | 0.40 | 0.55 | 0.00 | 0.00 | — | — |
| 42.0 | [Synthetic Pre-pretraining Survives Scale, but Not as a Grammatical Prior](https://arxiv.org/abs/2609.39827) `2609.39827` | 0.00 | 0.75 | 0.40 | 1.00 | 0.46 | — | — |
| 41.6 | [Mid-Harness: Scaling Actions Between Model and Harness for Terminal Agents](https://arxiv.org/abs/2609.39982) `2609.39982` | 0.50 | 0.45 | 0.40 | 0.00 | 0.50 | — | — |
| 41.1 | [PMosFM: Preconditioned Manifold Matching for One-Step Physics-Constrained Generation](https://arxiv.org/abs/2609.40287) `2609.40287` | 0.50 | 0.45 | 0.45 | 0.00 | 0.40 | — | — |
| 41.1 | [OpenTSLM TeeMoE: A Unified Time-Series Language Model for Forecasting, Contextual Prediction, and Reasoning](https://arxiv.org/abs/2609.40265) `2609.40265` | 0.30 | 0.45 | 0.40 | 1.00 | 0.20 | — | — |
| 40.5 | [Mitigating Representation Gaps in Amortized Bayesian Inference with Auxiliary Supervision](https://arxiv.org/abs/2609.39525) `2609.39525` | 0.65 | 0.40 | 0.55 | 0.00 | 0.00 | — | — |
| 40.0 | [MANET-GNN: Learned Decentralized Optimization of Power Allocation in Multi-Channel MANETs](https://arxiv.org/abs/2609.40170) `2609.40170` | 0.70 | 0.40 | 0.30 | 0.00 | 0.20 | — | — |
| 40.0 | [Doc2LoRA Provides Decodable Representations of Scientific Ideas](https://arxiv.org/abs/2609.38374) `2609.38374` | 0.30 | 0.40 | 0.55 | 1.00 | 0.00 | — | — |
| 40.0 | [A Unified Dual Method for Matching Problems](https://arxiv.org/abs/2609.39339) `2609.39339` | 0.70 | 0.40 | 0.45 | 0.00 | 0.00 | — | — |
| 40.0 | [Optimal Allocation and Volume under Surface](https://arxiv.org/abs/2609.38875) `2609.38875` | 0.70 | 0.40 | 0.30 | 0.00 | 0.20 | — | — |
| 40.0 | [Generative sequence modeling for infinite memory processes via predictive states](https://arxiv.org/abs/2609.38524) `2609.38524` | 0.50 | 0.40 | 0.75 | 0.00 | 0.00 | — | — |
| 40.0 | [Local polynomial density ratio estimation](https://arxiv.org/abs/2609.38412) `2609.38412` | 0.50 | 0.40 | 0.75 | 0.00 | 0.00 | — | — |
| 40.0 | [The Teacher Is a Direction, Not a Destination: Extrapolating RL-Induced Representation Residuals in On-Policy Distillation](https://arxiv.org/abs/2609.36484) `2609.36484` | 0.25 | 0.40 | 0.40 | 1.00 | 0.30 | — | — |
| 38.9 | [How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text](https://arxiv.org/abs/2609.40295) `2609.40295` | 0.00 | 0.75 | 0.60 | 1.00 | 0.00 | — | — |
| 38.9 | [OPTS-TTPO: Enhancing Finite-Sample Policy-Gradient Learning with Tree Search](https://arxiv.org/abs/2609.40035) `2609.40035` | 0.50 | 0.40 | 0.55 | 0.00 | 0.20 | — | — |
| 38.9 | [AutoDataBench: A Data-centric Testbed for Accelerating Auto Research](https://arxiv.org/abs/2609.40097) `2609.40097` | 0.30 | 0.45 | 0.30 | 1.00 | 0.20 | — | — |
| 38.9 | [Learning Global Sensitivity Indices from Observational Data: A Metamodel-Based Approach](https://arxiv.org/abs/2609.40342) `2609.40342` | 0.50 | 0.40 | 0.40 | 0.00 | 0.40 | — | — |
| 38.9 | [CAMOS: Coupled Oscillatory State-Space Model for Multimodal Clinical Time-Series](https://arxiv.org/abs/2609.39484) `2609.39484` | 0.50 | 0.40 | 0.55 | 0.00 | 0.20 | — | — |
| 38.9 | [Flow Matching under Noisy Latent Structure: Beyond Exact Low-Dimensional Support](https://arxiv.org/abs/2609.38918) `2609.38918` | 0.50 | 0.40 | 0.70 | 0.00 | 0.00 | — | — |
| 38.4 | [UBTree: Parallel Tree Drafting via Unigram and Bigram Models for Speculative Decoding](https://arxiv.org/abs/2609.39972) `2609.39972` | 0.65 | 0.45 | 0.40 | 0.00 | 0.00 | — | — |
| 38.4 | [Distributionally robust linear regression through the lens of adversarial training](https://arxiv.org/abs/2609.39449) `2609.39449` | 0.65 | 0.40 | 0.45 | 0.00 | 0.00 | — | — |
| 38.4 | [A Rank Graduation metric for Algorithmic fairness](https://arxiv.org/abs/2609.39025) `2609.39025` | 0.65 | 0.40 | 0.30 | 0.00 | 0.20 | — | — |
| 37.9 | [Semifactual Credit-Augmented Policy Optimization](https://arxiv.org/abs/2609.40360) `2609.40360` | 0.30 | 0.45 | 0.40 | 1.00 | 0.00 | — | — |
| 37.9 | [PatchHolmes: Agentic Patch Retrieval via Listwise Selection](https://arxiv.org/abs/2609.38807) `2609.38807` | 0.30 | 0.40 | 0.30 | 1.00 | 0.20 | — | — |
| 37.4 | [Less is more: error-distance scaling relation for data-efficient kilometer-scale downscaling of extreme heat](https://arxiv.org/abs/2609.40140) `2609.40140` | 0.65 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 37.4 | [Asymptotic Properties of Support Vector Machines in High-Dimension, Low-Sample-Size Settings under a Spiked Model](https://arxiv.org/abs/2609.39173) `2609.39173` | 0.65 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 37.1 | [Thinking Outside the Box: Can Language Models Rely on External Guidance Selectively?](https://arxiv.org/abs/2609.39578) `2609.39578` | 0.50 | 0.40 | 0.40 | 0.00 | 0.28 | — | — |
| 36.8 | [Cheap to Draw, Expensive to Trust: Certifying Test-Time Scaling Curves](https://arxiv.org/abs/2609.40190) `2609.40190` | 0.50 | 0.45 | 0.40 | 0.00 | 0.20 | — | — |
| 36.8 | [Beyond Model Ranking: Regime Diagnosis for Distributional-Statistical Misspecification in Industrial Time-Series Forecasting](https://arxiv.org/abs/2609.40117) `2609.40117` | 0.50 | 0.45 | 0.40 | 0.00 | 0.20 | — | — |
| 36.8 | [Learning Steganography Is Easy, Learning Steganographic Reasoning Is Hard](https://arxiv.org/abs/2609.39838) `2609.39838` | 0.30 | 0.40 | 0.40 | 1.00 | 0.00 | — | — |
| 36.8 | [Generative End-to-end Ad Retrieval at Douyin](https://arxiv.org/abs/2609.39327) `2609.39327` | 0.50 | 0.40 | 0.30 | 0.00 | 0.40 | — | — |
| 36.8 | [Text-Video Retrieval via Multi-Dimensional Saliency Assessment and Granularity-Aware Query Decomposition](https://arxiv.org/abs/2609.38949) `2609.38949` | 0.50 | 0.45 | 0.40 | 0.00 | 0.20 | — | — |
| 36.8 | [Estimation of the Label-Noise Transition Matrix with Performance Guarantees via Selective Classification](https://arxiv.org/abs/2609.39829) `2609.39829` | 0.50 | 0.40 | 0.60 | 0.00 | 0.00 | — | — |
| 36.8 | [Label-free steering: Compressing test-time reinforcement learning into bias-only subspaces](https://arxiv.org/abs/2609.18587) `2609.18587` | 0.25 | 0.40 | 0.40 | 1.00 | 0.10 | — | — |
| 36.3 | [The Low-Rank Structure of VLA Reinforcement Learning](https://arxiv.org/abs/2609.34599) `2609.34599` | 0.50 | 0.40 | 0.40 | 0.00 | 0.23 | — | — |
| 35.8 | [Disentangling Computation in Multi-Task Neural Networks with the Green's Operator](https://arxiv.org/abs/2609.40292) `2609.40292` | 0.50 | 0.40 | 0.55 | 0.00 | 0.00 | — | — |
| 35.8 | [Role-Adaptive Policy Optimization for Offline Reinforcement Learning](https://arxiv.org/abs/2609.40149) `2609.40149` | 0.50 | 0.40 | 0.40 | 0.00 | 0.20 | — | — |
| 35.8 | [Policy Iteration Is Not Strongly Polynomial for Deterministic Markov Decision Processes: The Price of Algorithmic Anarchy](https://arxiv.org/abs/2609.40147) `2609.40147` | 0.50 | 0.40 | 0.55 | 0.00 | 0.00 | — | — |
| 35.8 | [TRACE: Target-Aware Retrieval, Attributed Evidence, and Contract-Constrained Extraction for LitTraceQA](https://arxiv.org/abs/2609.38861) `2609.38861` | 0.50 | 0.40 | 0.40 | 0.00 | 0.20 | — | — |
| 35.8 | [Exploring Forum Post Retrieval with Generative Modeling](https://arxiv.org/abs/2609.38646) `2609.38646` | 0.50 | 0.40 | 0.40 | 0.00 | 0.20 | — | — |
| 35.8 | [Awakening of the Buddha: Subspace Learning During Population-Loss Plateaus](https://arxiv.org/abs/2609.39408) `2609.39408` | 0.50 | 0.40 | 0.55 | 0.00 | 0.00 | — | — |
| 35.8 | [Minimax rates for learning spectral Barron functions by deep ReLU neural networks](https://arxiv.org/abs/2609.39020) `2609.39020` | 0.50 | 0.40 | 0.55 | 0.00 | 0.00 | — | — |
| 35.8 | [Sharp Statistical Rates for Asynchronous TD Learning with Markovian Data](https://arxiv.org/abs/2609.38880) `2609.38880` | 0.50 | 0.40 | 0.55 | 0.00 | 0.00 | — | — |
| 35.8 | [Optimal VC Dimension of Contrastive Learning with Margin](https://arxiv.org/abs/2609.38834) `2609.38834` | 0.50 | 0.40 | 0.55 | 0.00 | 0.00 | — | — |
| 35.8 | [Learning-Enabled Estimation: Tight Characterizations under Sample Selection Biases](https://arxiv.org/abs/2609.38608) `2609.38608` | 0.50 | 0.40 | 0.55 | 0.00 | 0.00 | — | — |
| 35.8 | [FIGS: Evaluating Multi-Turn Sycophancy Without Penalizing Empathy](https://arxiv.org/abs/2609.39863) `2609.39863` | 0.30 | 0.45 | 0.30 | 1.00 | 0.00 | — | — |
| 35.8 | [O-Funnel: Lossless Structural Capture and Requirement-Driven Extraction from Drifting, Heterogeneous Documents](https://arxiv.org/abs/2609.39209) `2609.39209` | 0.50 | 0.45 | 0.00 | 1.00 | 0.00 | — | — |
| 34.7 | [Prototype-Rule Neurosymbolic Regularization for Rank-Constrained Tensor Neural Networks under Label Scarcity](https://arxiv.org/abs/2609.40131) `2609.40131` | 0.50 | 0.45 | 0.30 | 0.00 | 0.20 | — | — |
| 34.7 | [Who Verifies the Graph? Misspecification Attacks on Causal Action Verification for Language Agents](https://arxiv.org/abs/2609.40027) `2609.40027` | 0.50 | 0.45 | 0.45 | 0.00 | 0.00 | — | — |
| 34.7 | [MemLife: Curating and Reasoning over Long-Term Egocentric Video Memories](https://arxiv.org/abs/2609.40195) `2609.40195` | 0.50 | 0.45 | 0.30 | 0.00 | 0.20 | — | — |
| 34.7 | [Signal Processing over Product DAGs: Causal Shifts and Filters](https://arxiv.org/abs/2609.40275) `2609.40275` | 0.50 | 0.40 | 0.50 | 0.00 | 0.00 | — | — |
| 33.7 | [cua-speedrun: Standardized Benchmarking of the Speed of Computer-Use Agents](https://arxiv.org/abs/2609.40284) `2609.40284` | 0.50 | 0.45 | 0.40 | 0.00 | 0.00 | — | — |
| 33.7 | [Reinforcement Learning-Guided Graph Transformations for SpTRSV Optimization](https://arxiv.org/abs/2609.40159) `2609.40159` | 0.50 | 0.40 | 0.30 | 0.00 | 0.20 | — | — |
| 33.7 | [Game-Guided Skill Discovery through Self-Play for Playable Agent Control](https://arxiv.org/abs/2609.40137) `2609.40137` | 0.50 | 0.40 | 0.30 | 0.00 | 0.20 | — | — |
| 33.7 | [AdaGEPA: Adaptive Feedback Allocation for Reflective Prompt Optimization](https://arxiv.org/abs/2609.39927) `2609.39927` | 0.50 | 0.45 | 0.40 | 0.00 | 0.00 | — | — |
| 33.7 | [When a Kindergartener Solves Calculus: Measuring Capability Leakage in Role-Prompted Reasoning Models](https://arxiv.org/abs/2609.39846) `2609.39846` | 0.50 | 0.45 | 0.40 | 0.00 | 0.00 | — | — |
| 33.7 | [Stress-Testing LLM Lie Detectors: Role-Play Failures and Spurious Correlations](https://arxiv.org/abs/2609.39807) `2609.39807` | 0.30 | 0.40 | 0.75 | 0.00 | 0.00 | — | — |
| 33.7 | [LatentHarness: Learning Latent Actions for Memory and Reasoning via Counterfactual Policy Distillation](https://arxiv.org/abs/2609.39740) `2609.39740` | 0.50 | 0.45 | 0.40 | 0.00 | 0.00 | — | — |
| 33.7 | [Towards Universal Wasserstein Barycenters through Flow Matching](https://arxiv.org/abs/2609.38547) `2609.38547` | 0.50 | 0.45 | 0.40 | 0.00 | 0.00 | — | — |
| 33.2 | [Breaking Babel: A Self-Evolving Multi-Agent System for Long-Form Subtitle Translation](https://arxiv.org/abs/2609.38660) `2609.38660` | 0.30 | 0.45 | 0.30 | 0.00 | 0.50 | — | — |
| 32.6 | [Compression Footprints as Security Signals for Model-Poisoning Defense in Federated Learning](https://arxiv.org/abs/2609.40312) `2609.40312` | 0.50 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 32.6 | [LARC: Low-Rank Adaptive Residual Connections for Learning in Frozen Models](https://arxiv.org/abs/2609.40063) `2609.40063` | 0.50 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 32.6 | [Amortized Bayesian Inference on Multilevel Models of Arbitrary Structure](https://arxiv.org/abs/2609.40024) `2609.40024` | 0.50 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 32.6 | [Component-Weighted Centroid Search for Exact Incremental BPE](https://arxiv.org/abs/2609.40016) `2609.40016` | 0.50 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 32.6 | [Provably Tractable NFA-Constrained Language Generation via HMMs](https://arxiv.org/abs/2609.40185) `2609.40185` | 0.50 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 32.6 | [From Tweets to Trades: Analyzing the Influence of Public Mood over Stock Market Performance in Turkiye](https://arxiv.org/abs/2609.40064) `2609.40064` | 0.50 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 32.6 | [MGhana-ST: A Low-Resource Speech Translation Dataset for Ghanaian Languages and an Analysis of Multilingual Training Trade-offs](https://arxiv.org/abs/2609.40041) `2609.40041` | 0.50 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 32.6 | [GrammarRL: Effective Grammar-Constrained Decoding via Reinforcement Learning](https://arxiv.org/abs/2609.39869) `2609.39869` | 0.50 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 32.6 | [MemCodex: Self-Programming Hierarchical Memory for Language Agents](https://arxiv.org/abs/2609.39765) `2609.39765` | 0.50 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 32.6 | [Routing Between Generative and Collaborative User Profiles: A Serving-Time Gate for Controllable Novelty](https://arxiv.org/abs/2609.39043) `2609.39043` | 0.30 | 0.40 | 0.55 | 0.00 | 0.20 | — | — |
| 32.6 | [Evidence First, Arithmetic Second: A System Report and Failure Analysis for DocSem](https://arxiv.org/abs/2609.39013) `2609.39013` | 0.50 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 32.6 | [Breaking News Out of the Filter Bubble: Generative AI Search Diversifies Collective Attention and Raises Shared Information Consumption](https://arxiv.org/abs/2609.38946) `2609.38946` | 0.50 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 32.6 | [Minimax Additive Regression under Unknown Dependent Designs](https://arxiv.org/abs/2609.39212) `2609.39212` | 0.50 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 32.6 | [Warm-starting PDE solvers with any-dimensional machine learning](https://arxiv.org/abs/2609.38916) `2609.38916` | 0.50 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 32.6 | [Understanding Off- vs On-Policy Distillation: A Tale of Distinct Training Objectives](https://arxiv.org/abs/2609.38666) `2609.38666` | 0.50 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 32.6 | [PTED: A multi-dimensional two-sample test for scientific inference and generative machine learning](https://arxiv.org/abs/2609.38388) `2609.38388` | 0.50 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 32.6 | [LoopVL: Recurrent Visual Intelligence](https://arxiv.org/abs/2609.38426) `2609.38426` | 0.25 | 0.45 | 0.00 | 1.00 | 0.30 | — | — |
| 32.6 | [LANTERN: Illuminating Hidden Mathematical Knowledge in Language Models](https://arxiv.org/abs/2609.32264) `2609.32264` | 0.25 | 0.40 | 0.40 | 0.00 | 0.50 | — | — |
| 32.1 | [Working Around the Compute Ceiling: Byte-Exact Memory in Galahad Makes LLM Reading a One-Time Cost LLM Reading a One-Time Cost](https://arxiv.org/abs/2609.39358) `2609.39358` | 0.00 | 0.40 | 0.55 | 1.00 | 0.10 | — | — |
| 31.6 | [From DNA Design to DNA Slimming: Auditable Agentic Discovery of a Deletion-Only Designer](https://arxiv.org/abs/2609.40143) `2609.40143` | 0.50 | 0.45 | 0.30 | 0.00 | 0.00 | — | — |
| 31.6 | [MCD: Causal Distillation of Multimodal In-Context Learning in Large Vision-Language Models](https://arxiv.org/abs/2609.39920) `2609.39920` | 0.50 | 0.45 | 0.30 | 0.00 | 0.00 | — | — |
| 31.1 | [Distribution Matching Distillation for Continuous Diffusion Language Models](https://arxiv.org/abs/2609.40235) `2609.40235` | 0.35 | 0.40 | 0.40 | 0.00 | 0.20 | — | — |
| 31.1 | [Decision-Oriented Recommendation Reranking: An Empirical Study of Jev](https://arxiv.org/abs/2609.40241) `2609.40241` | 0.30 | 0.40 | 0.40 | 0.00 | 0.30 | — | — |
| 31.1 | [Persistent Context Graphs for Efficient Memory Compaction in LLM Agents](https://arxiv.org/abs/2609.40118) `2609.40118` | 0.45 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 31.1 | [Dating the Model: Hidden Dates in System Prompts Affect LLM Evaluation](https://arxiv.org/abs/2609.36931) `2609.36931` | 0.30 | 0.40 | 0.40 | 0.00 | 0.30 | — | — |
| 30.5 | [Efficient Active Auditing of Multi-Group Fairness with Bias Probes](https://arxiv.org/abs/2609.40034) `2609.40034` | 0.50 | 0.40 | 0.30 | 0.00 | 0.00 | — | — |
| 30.5 | [Fenchel Tilting: Weighted Correction for Efficient Finetuning of Generative Models](https://arxiv.org/abs/2609.40030) `2609.40030` | 0.10 | 0.45 | 0.55 | 0.00 | 0.40 | — | — |
| 30.5 | [Probabilistic Adversarial Training](https://arxiv.org/abs/2609.39798) `2609.39798` | 0.50 | 0.40 | 0.30 | 0.00 | 0.00 | — | — |
| 30.5 | [Towards Optimal Inventory Control under Censored Demand: A Biased Sample-Average Approximation Approach](https://arxiv.org/abs/2609.39397) `2609.39397` | 0.50 | 0.40 | 0.30 | 0.00 | 0.00 | — | — |
| 30.5 | [Advantage of Sample Complexity in Quantum PAC Learning Requires Inverse Access to State-Preparation Unitaries](https://arxiv.org/abs/2609.38403) `2609.38403` | 0.50 | 0.40 | 0.30 | 0.00 | 0.00 | — | — |
| 30.5 | [Evaluating Bounded Autonomy in Regulated Agentic AI: A Diagnostic Harness with Constitutional Rewards, Escalation Labels, and Runtime Governance](https://arxiv.org/abs/2609.37501) `2609.37501` | 0.50 | 0.40 | 0.30 | 0.00 | 0.00 | — | — |
| 30.0 | [RSIGame: Autonomous Agentic Game Development with Recursive Self-improvement](https://arxiv.org/abs/2609.39045) `2609.39045` | 0.00 | 0.40 | 0.30 | 1.00 | 0.30 | — | — |
| 29.5 | [Cognitive Enhancement: Rethinking the Necessity of Role-Playing for Large Language Models](https://arxiv.org/abs/2609.39853) `2609.39853` | 0.30 | 0.40 | 0.55 | 0.00 | 0.00 | — | — |
| 29.5 | [When LLM-Inferred User Context Adds Value in Production Streaming Recommendation](https://arxiv.org/abs/2609.38999) `2609.38999` | 0.30 | 0.40 | 0.40 | 0.00 | 0.20 | — | — |
| 29.5 | [Component-Aware Feedback for Self-Evolving Programs](https://arxiv.org/abs/2609.38639) `2609.38639` | 0.30 | 0.40 | 0.40 | 0.00 | 0.20 | — | — |
| 29.5 | [ShamAN-Q: Shampoo Augmented NanoQuant for Sub-1-bit LLM Weights](https://arxiv.org/abs/2609.38521) `2609.38521` | 0.30 | 0.40 | 0.40 | 0.00 | 0.20 | — | — |
| 28.9 | [Looped Diffusion Transformer](https://arxiv.org/abs/2609.40305) `2609.40305` | 0.35 | 0.45 | 0.40 | 0.00 | 0.00 | — | — |
| 28.9 | [Learning Functional Subspaces for Neural Network Compression](https://arxiv.org/abs/2609.40127) `2609.40127` | 0.25 | 0.40 | 0.60 | 0.00 | 0.00 | — | — |
| 28.9 | [Agent Error Dataset: Scaling 50,000 Error--Diagnosis Pairs for Failure Analysis and Error-Aware Post-Training](https://arxiv.org/abs/2609.40111) `2609.40111` | 0.30 | 0.40 | 0.30 | 0.00 | 0.30 | — | — |
| 28.4 | [Index-Translate: A Multilingual Translation Model Family -- Text, Speech, Controlled Dubbing, and Long-Document Translation](https://arxiv.org/abs/2609.40181) `2609.40181` | 0.00 | 0.45 | 0.40 | 1.00 | 0.00 | — | — |
| 28.4 | [Learning to Route in Visual Space via Multi-Step Embedding Retrieval](https://arxiv.org/abs/2609.38743) `2609.38743` | 0.30 | 0.45 | 0.30 | 0.00 | 0.20 | — | — |
| 27.9 | [Learning to Plan from Random Exploration](https://arxiv.org/abs/2609.38383) `2609.38383` | 0.35 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 27.4 | [JuryFlow: Disagreement-Guided Human-in-the-Loop Multi-Agent Evaluation](https://arxiv.org/abs/2609.40103) `2609.40103` | 0.30 | 0.40 | 0.30 | 0.00 | 0.20 | — | — |
| 27.4 | [BAM! Bayesian Anything Model: a foundation model for generative computational imaging](https://arxiv.org/abs/2609.39660) `2609.39660` | 0.20 | 0.45 | 0.40 | 0.00 | 0.20 | — | — |
| 26.8 | [See it, Say it, Sorted: Mechanistic Diagnosis and Parameter-Space Mitigation of Emergent Misalignment in LLMs](https://arxiv.org/abs/2609.34970) `2609.34970` | 0.00 | 0.40 | 0.30 | 1.00 | 0.10 | — | — |
| 26.3 | [Is Weight Tying Still Beneficial for Decoder-Only LLMs in Private Settings Under DP-SGD?](https://arxiv.org/abs/2609.40335) `2609.40335` | 0.30 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 26.3 | [On the (In)effectiveness of AMR Augmentation for Large Language Models](https://arxiv.org/abs/2609.40121) `2609.40121` | 0.30 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 26.3 | [OverForge: Reasoning Through Strategies and Tactics Helps Cooperative Lifelong Adaptation](https://arxiv.org/abs/2609.39727) `2609.39727` | 0.30 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 26.3 | [KUAISHOU Explorer LLM-Rec Challenge 2026: Reasoning Generative Recommendation](https://arxiv.org/abs/2609.39828) `2609.39828` | 0.30 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 26.3 | [Raw-Routed Mixture of Adapters: A Causal Intervention for Routing Collapse in Time Series Foundation Models](https://arxiv.org/abs/2609.39445) `2609.39445` | 0.30 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 25.8 | [MeanVoiceFlow2: Joint Optimization of Mean Flow and Content Encoder for Fast One-Step Zero-Shot Voice Conversion](https://arxiv.org/abs/2609.40087) `2609.40087` | 0.35 | 0.40 | 0.30 | 0.00 | 0.00 | — | — |
| 25.8 | [TAGGRAPH: Tag-Augmented Graphs for Graph Retrieval of Agent Persistent Histories](https://arxiv.org/abs/2609.38353) `2609.38353` | 0.15 | 0.40 | 0.30 | 0.00 | 0.40 | — | — |
| 25.8 | [Discrete Score Matching Enables Causal Discovery from Count Data](https://arxiv.org/abs/2609.39326) `2609.39326` | 0.35 | 0.40 | 0.30 | 0.00 | 0.00 | — | — |
| 25.8 | [Steepest Guidance: A Practical and Principled Approach to Inference-Time Alignment of Flow and Diffusion-based Models](https://arxiv.org/abs/2609.39091) `2609.39091` | 0.35 | 0.40 | 0.30 | 0.00 | 0.00 | — | — |
| 25.3 | [LEAP: Learned Block-wise Evidence Retrieval for Long Audio-Video Perception](https://arxiv.org/abs/2609.39938) `2609.39938` | 0.15 | 0.45 | 0.30 | 0.00 | 0.30 | — | — |
| 25.3 | [STARS: From Spatiotemporal Dynamics to Social Representations in Human-Robot Interaction](https://arxiv.org/abs/2609.40245) `2609.40245` | 0.30 | 0.45 | 0.30 | 0.00 | 0.00 | — | — |
| 25.3 | [Debias It Yourself: Teaching LLMs Cognitive Bias Mitigation Interventions](https://arxiv.org/abs/2609.40124) `2609.40124` | 0.30 | 0.45 | 0.30 | 0.00 | 0.00 | — | — |
| 25.3 | [Explore-on-Graph: Hybrid Embedding-LLM Reasoning for Knowledge Graph Question Answering under Incompleteness](https://arxiv.org/abs/2609.39786) `2609.39786` | 0.30 | 0.45 | 0.30 | 0.00 | 0.00 | — | — |
| 25.3 | [Conversational Capture: A Trajectory-Level Framework for Evaluating Generative Engine Optimization in Multi-turn Human-Agent Interaction](https://arxiv.org/abs/2609.40069) `2609.40069` | 0.30 | 0.40 | 0.05 | 0.00 | 0.40 | — | — |
| 25.3 | [Argument Structure Prediction in Online Conversations: A Comparative Study of Modeling Paradigms and Task Architectures](https://arxiv.org/abs/2609.39225) `2609.39225` | 0.30 | 0.45 | 0.30 | 0.00 | 0.00 | — | — |
| 25.3 | [EVOKE: Eliciting World Knowledge in Agents for Transferable Decision-Making](https://arxiv.org/abs/2609.38334) `2609.38334` | 0.05 | 0.40 | 0.00 | 1.00 | 0.30 | — | — |
| 24.7 | [DynaHarness: A Dynamic Physical Harness for Self-Evolving Robot Agents](https://arxiv.org/abs/2609.40306) `2609.40306` | 0.25 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 24.7 | [From Spectra to Joint Schedules in LLM Pre-training: 3+3(+2) Scaling-Law Regimes](https://arxiv.org/abs/2609.40148) `2609.40148` | 0.05 | 0.40 | 0.55 | 0.00 | 0.20 | — | — |
| 24.7 | [Replay on Demand: An Emergent Curriculum for Balancing Adaptation and Forgetting in Continued Pretraining](https://arxiv.org/abs/2609.40089) `2609.40089` | 0.25 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 24.2 | [Inference Auctions](https://arxiv.org/abs/2609.40070) `2609.40070` | 0.30 | 0.40 | 0.30 | 0.00 | 0.00 | — | — |
| 24.2 | [AdaM-Rec: Adaptive Modality Routing for Multimodal Recommendation](https://arxiv.org/abs/2609.38455) `2609.38455` | 0.30 | 0.40 | 0.30 | 0.00 | 0.00 | — | — |
| 24.2 | [Privacy in Personalized AI Is a System Property, Not Just a Model Property](https://arxiv.org/abs/2609.38289) `2609.38289` | 0.50 | 0.40 | 0.00 | 0.00 | 0.00 | — | — |
| 24.2 | [Principal Component Regression Dominates all Monotone Spectral Filters for Linear Regression](https://arxiv.org/abs/2609.39440) `2609.39440` | 0.50 | 0.40 | 0.00 | 0.00 | 0.00 | — | — |
| 22.6 | [PrefPI: Preference-Guided Steering into Out-of-Distribution Behaviors](https://arxiv.org/abs/2609.40165) `2609.40165` | 0.25 | 0.40 | 0.30 | 0.00 | 0.00 | — | — |
| 22.6 | [OverdoseMoE: A Multi-Expert Framework for Opioid Overdose Risk Prediction](https://arxiv.org/abs/2609.40108) `2609.40108` | 0.25 | 0.40 | 0.30 | 0.00 | 0.00 | — | — |
| 22.3 | [Scaling Laws for Looped Mixture of Experts](https://arxiv.org/abs/2609.40316) `2609.40316` | 0.00 | 0.45 | 0.40 | 0.00 | 0.28 | — | — |
| 22.1 | [SCB: SpeechConversationBench for Evaluating Multi-Turn Reasoning in Speech-to-Speech Models](https://arxiv.org/abs/2609.40198) `2609.40198` | 0.50 | 0.00 | 0.30 | 0.00 | 0.00 | — | — |
| 21.6 | [Ranking-Aware Prompt Optimization for Multimodal Clinical Diagnosis](https://arxiv.org/abs/2609.40361) `2609.40361` | 0.05 | 0.40 | 0.40 | 0.00 | 0.20 | — | — |
| 21.6 | [The Evolution of Attention in Large Language Models: Mechanisms, Trade-offs, and Emerging Trends](https://arxiv.org/abs/2609.39661) `2609.39661` | 0.30 | 0.40 | 0.00 | 0.00 | 0.23 | — | — |
| 21.1 | [MatLoom: Layered Text-to-Material Generation in a Compact Program Space](https://arxiv.org/abs/2609.40322) `2609.40322` | 0.10 | 0.45 | 0.40 | 0.00 | 0.00 | — | — |
| 20.5 | [Comparison of techniques for fine-tuning open-weight models for entity extraction from radiology reports](https://arxiv.org/abs/2609.40236) `2609.40236` | 0.15 | 0.35 | 0.40 | 0.00 | 0.00 | — | — |
| 19.5 | [LLM Persona Unlearning](https://arxiv.org/abs/2609.39882) `2609.39882` | 0.05 | 0.45 | 0.40 | 0.00 | 0.00 | — | — |
| 16.8 | [Overview of BioASQ 2026: The fourteenth BioASQ Challenge on Large-Scale Biomedical Semantic Indexing and Question Answering](https://arxiv.org/abs/2609.39975) `2609.39975` | 0.00 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 16.8 | [The Concrete-Arbitrary Gap: Kinship Reasoning in LLMs Is Not Indifferent to Presentation](https://arxiv.org/abs/2609.39913) `2609.39913` | 0.00 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 16.3 | [Image Classifiers are Efficient Self-Supervised Video Representation Learners](https://arxiv.org/abs/2609.40347) `2609.40347` | 0.05 | 0.40 | 0.30 | 0.00 | 0.00 | — | — |
| 14.7 | [Efficient Expert-Parallel Communication on PCIe-Connected Consumer GPUs](https://arxiv.org/abs/2609.40093) `2609.40093` | 0.00 | 0.40 | 0.30 | 0.00 | 0.00 | — | — |

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