PyTorchKR Papers

최신 논문

총 523,801편
태그·기간 등 상세 필터

카테고리 탐색

arXiv:2609.30250v2cs.CL

Agentic Detection of Online Conspiracies

Conspiratorial discourse on social media is not always expressed through explicit claims or stable lexical markers. The same surface content may express endorsement, legitimate concerns, criticism, satire, or mockery. The main challenge is therefore not only recognizing conspiracy-related claims, but inferring the speaker's intent -- the utterance's illocutionary force. We argue that this can be achieved through the use of relevant social contexts and propose an agentic framework, equipped with a set of tools supporting social queries. We demonstrate the benefits of our approach on a unique dataset of Hebrew tweets, covering 80%--90% of the public Hebrew tweets published over a four-year span (late 2018-- early 2023), encompassing several election cycles as well as the COVID pandemic years and related vaccination campaigns. This extensive coverage can be used in recovering different social contexts. Evaluating our framework on a manually-annotated adversarial dataset, we find that context-aware workflows consistently outperform text-only classification and that the agentic framework performs better than other frameworks and settings, with significant gains over a non-agentic model exposed to the same contexts available to the agent. We further provide an analysis of the results, the errors and efficiency (token economy) tradeoffs. These findings support viewing the task of conspiracy detection as a socially embedded interpretation task, in which effective classification depends not only on access to contexts, but also on adaptive reasoning in which the agent uses tools on a per-case basis, asking only for evidence relevant to its current reasoning step.

arXiv:2609.30192v2cs.AI

SAGE: Mitigating Long-Horizon Reasoning Biases via Topological Guidance

Long-horizon reasoning remains a central challenge for large language models (LLMs) under sparse-reward regimes. We argue that this brittleness arises from two biases induced by complex reasoning spaces: an exploration bias, where models are drawn toward locally plausible but structurally unstable branches, and a compounding bias, where small local deviations accumulate across depth and suppress rare rewards. We introduce Symbolic Closure Analysis (SCA) as a theoretical lens characterizing how branching structures and sparse rewards induce these biases in long-horizon reasoning with local admissibility, and as a design principle for structural priors in less formal reasoning tasks. Motivated by this analysis, we propose SAGE (Structural Admissibility-Guided Exploration), a unified framework that injects structural guidance to alleviate exploration bias and compounding bias in long-horizon reasoning. SAGE combines two complementary structural guidance: algebraic sparsification, which projects locally admissible candidates onto operator-indexed algebraic subspaces to suppress spurious branching and mitigate exploration bias, and hyperbolic structural guidance, which embeds reasoning states into a negatively curved space to provide dense depth-wise signals and mitigate compounding bias. Across 12 benchmarks and 7 model families, SAGE outperforms competitive baselines. In particular, SAGE achieves up to an 8-fold improvement on the Andrews-Curtis problem, an open real-world long-horizon task. Code is available at: https://github.com/Susan571/SAGE-NeurIPS2026.

arXiv:2609.29117v2cs.LG

A Concentration Bound for Two-Timescale Actor-Critic Algorithm

Significant research effort has been directed in recent years towards establishing both asymptotic and non-asymptotic convergence guarantees for two-timescale actor--critic algorithms, where the actor recursion is run on a slower timescale than the critic recursion. This work derives a uniform all-time concentration bound for the actor--critic algorithm with function approximation in the long-run average-reward setting. This bound helps us analyze the behavior of the actor parameter with high probability. We show that, after some finite time, the actor parameter enters a safe region and remains within it thereafter with high probability. Specifically, with probability at least $1-\epsilon_1-\epsilon_2$, the actor error $\Vert \theta_k-\theta^{*}\Vert$ is $O\left(\frac{n_0^{3/4}}{k}\frac{1}{\sqrt{\epsilon_2}}+\left(\frac{1}{n_0}\right)^{1/4}\log^{1/4}\left(\frac{1}{\epsilon_1}\right)+\left(\frac{1}{n_0}\right)^{1/4}\right)$ for all $k\geq n_0$ and sufficiently large $n_0$. We also present experimental results demonstrating that the aforementioned actor error diminishes with the number of actor-parameter updates.

arXiv:2609.26142v2stat.ME

Three Routes to One Answer: Reconciling AIPW, TMLE, and Double Machine Learning for Applied Researchers

Augmented inverse probability weighting (AIPW), targeted maximum likelihood estimation (TMLE), and double/debiased machine learning (DML) are three asymptotically equivalent routes to the same efficient influence function for the average treatment effect. We contribute a worked, shared-nuisance reconciliation on real data: what a practitioner must hold fixed for the routes and their software to agree, and when they do not. Using open data on smoking cessation and weight change (n = 1566), one Super Learner library, and folds matched within schemes, we build all three by hand, in full-sample, cross-fit, and double-cross-fit variants. A semi-synthetic benchmark with known sample-average effects on a fixed design grades interval coverage. The six doubly robust estimates span 3.32--3.39 kg over repeated splits. Single runs of our code and the tmle, AIPW, DoubleML, and tmle3 packages span 3.32--3.49 kg, compatible with fitting-scheme differences and single-split noise. Over 20 seeds, the three implementations sharing our library agree within 0.01 kg in median, although one is a TMLE. Agreement can fail: with in-sample nuisance fits on a binary mortality outcome, and under poor overlap in a job-training example. We recommend diagnosing overlap before choosing an estimator, and provide a reporting checklist and software code.

arXiv:2609.25447v3cs.CL

Conduct Under Pressure: What Sixty Language Models Do When a User Pushes

We study what LLMs do when a user applies pressure in an uncomfortable situation: a user insists, begs, flatters or grieves, and the model gives up a correct fact, writes a document it should refuse, or cheers a plan that will cost the user money, a multi-turn form of sycophancy. We send frozen multi-turn scenes to 60 models from 13 vendors and label each transcript with a trajectory (the model held its position or folded) and a manner (how it held or folded). Three scenes were used to build the codebook; six more were written afterwards and preregistered as a replication. Whether a model holds tracks its generation: fold rate correlates with a public capability index at Spearman -0.64 to -0.69 across the two scene sets, so newer models fold less, unlike earlier reports that sycophancy grows with model scale. How a model holds tracks the vendor: six of the 17 manner codes sort by vendor at permutation p <= 0.001, corrected across the codebook, and four do so again on the replication scenes. Anthropic's models name the feeling, warn and offer an alternative; OpenAI's do less of all three and almost never cite their own rules; Google's cite themselves and their AI nature when asked to do something; Meta's write the artifact they shouldn't. Six LLM coders from three vendors labeled every transcript, and three human coders labeled fifty held-out transcripts. The machines were the more consistent readers (Krippendorff's alpha 0.66 against 0.46), agreed with the author on trajectory at kappa 0.84 to 0.91, and matched the adjudicated reference at 0.83. For behavior a non-specialist can judge, the human contribution is authoring and bounding the codes and owning a small reference, not producing labels at volume.

arXiv:2609.25010v2cs.AI

Do Synthetic Personas Predict Real Audience Response? A Headline-Ranking Study with a Generic-Reader Baseline

Do demographic personas improve an LLM's prediction of audience engagement? We compare a fixed ten-persona panel with a generic-reader baseline on Upworthy headlines, with supplementary news and Reddit comparisons. The primary experiment collects Gemini 3.1 Flash-Lite and GPT-4.1 observations on the same 190-package exploratory cohort, selected with simultaneous winner screening. It allocates thirty scores per arm and variant within model, retains exact raw responses, and reports metric-specific valid and paired counts. Complete-package coverage is 188 for Gemini and 188 for GPT-4.1. The baseline-minus-panel Kendall $\tau_b$ gap is +0.222 (CI [+0.061, +0.385]) for Gemini and +0.257 (CI [+0.119, +0.391]) for GPT-4.1, on respectively 136 and 188 valid rank pairs. Expected top-1 is 39.1% versus 54.8% for Gemini and 42.0% versus 61.7% for GPT-4.1 (panel versus baseline). These intervals condition on the selected cohort and recorded generations and do not adjust for dependence between related experiments. The Upworthy archive documents a randomization-risk period. Excluding that period in a post-hoc sensitivity retains positive expected top-1 and lift gaps in both configurations, while Gemini tau-b and Spearman intervals include zero. Supplementary Reddit top-1 and captured lift favor the panel; these aggregate comparisons use unreconciled collection protocols and do not isolate dataset effects. A raw-replay artifact supports offline reproduction of the primary experiment. The results concern this panel, cohort, and protocol; they are not a held-out evaluation or evidence of universal persona harm. Provider request protocols differ across models, and no causal mechanism is isolated.

arXiv:2609.24492v2cs.CV

Identity-Consistent Analysis of Long-Shot Windsurfing Video: A Domain-Specific Offline Tracking System

Long-shot windsurfing video combines small targets, large camera pans, prolonged overlaps, and rapidly changing backgrounds. The desired output is not a generic MOT trace but a separate, stable rider-relative video for each surfer; one false identity merge can invalidate an otherwise useful result. We present an offline analysis system that detects surfers, forms conservative local tracklets, links them globally with camera-compensated motion and a foreground-masked sail-color descriptor, and uses two pose keypoints on the rig to drive a rider-relative virtual camera. The tracking stage is evaluated on 21 manually reconstructed development videos containing 41,004 retained observations. On this fixed-observation protocol, the production system achieves 0.957 pairwise precision, 0.918 recall, and 0.937 F1, compared with 0.792 F1 for OC-SORT and 0.828 for BoT-SORT. Compared with OC-SORT, it reduces fragmentation excess from 845 to 42, but nine of its 95 output tracks mix rider identities and these errors affect seven of the 21 videos.

arXiv:2609.22547v3stat.ML

When Pass@$k$ Crossovers Are Real - and Why: Paired Inference and the Markov Kernel of RLVR

Reinforcement learning with verifiable rewards (RLVR) typically improves pass@1 but can fall behind its base model at large sampling budgets $k$, commonly interpreted as sharpening existing solutions. Because adjacent pass@$k$ estimates reuse generations, pointwise error bars cannot establish a crossover. We formulate paired inference on the RL-minus-base difference curve: a prompt-level multiplier bootstrap yields a one-sided dominance test and simultaneous confidence bands requiring both an early gain and a later loss. Among nine public RLVR checkpoints at 3k-16k token limits, only one narrowly meets this criterion before multiplicity adjustment; asymmetric truncation can create, mask, or reverse crossovers. On 1,060 deduplicated competition problems at 32k tokens, two 1.5B models exhibit certified crossovers around $k\approx10$-$60$, while two 7B models show no supported loss. A paired variance decomposition shows why adding prompts improves power more than deeper sampling. Prompts with identical base success probability diverge after RLVR, with dispersion replicating across independent generation halves and violating Lipschitz slope bounds for reasonable deterministic maps. Modeling RLVR as a Markov kernel, Jensen's inequality shows that conditional dispersion depresses pass@$k$ for $k\ge2$. Once dispersion is modeled, the conditional-mean log-odds slope is below one in every pair: RLVR improves hard prompts on average, while late-$k$ loss comes from a 2%-3% minority of base-solved prompts collapsing onto a repeated modal wrong answer and resisting prefix rescue up to 4,096 tokens. In controlled runs from one base model on 3,000 new problems, all four GRPO arms fall behind near $k\approx200$ and remain behind through $k=4096$, while SFT stays significantly ahead throughout.

arXiv:2609.22377v3eess.IV

Forecasting Brain Tracer Distribution from Pre-Contrast MRI: A Comparison of Direct Regression and Flow Matching

The brain fluid clearance system can be studied with intrathecal contrast-enhanced magnetic resonance imaging (MRI). By injecting a tracer into the spinal canal and performing MRI over 24--48\,h, one can measure how a tracer injected into the cerebrospinal fluid spreads through the brain. As this is an invasive procedure requiring several MRI scans, we ask how much of the later tracer distribution can be predicted from the pre-contrast scan and elapsed time alone. Using 104 training patients (878 MRI volumes), we compared a direct image-to-image regression model with a more complex generative model based on conditional flow matching, and tested both on 23 held-out patients and on 51 patients from a separate cohort with idiopathic normal pressure hydrocephalus (iNPH). Relative to simply copying the pre-contrast scan, the direct model reduced the mean absolute error (intensities normalized to $[0,1]$) from 0.089 to 0.036, about 60\%, in held-out patients and from 0.075 to 0.056, about 25\%, in the iNPH cohort. The reduction held in every sampled elapsed-time interval and brain compartment, and errors remained largest at 20--30\,h and in the ventricles, where enhancement is strongest. Its predicted change maps agreed with the observed tracer enhancement rather than only reproducing anatomy. The generative model was less accurate in both cohorts. A substantial part of the later tracer distribution is therefore predictable from the patient's brain anatomy as seen on pre-contrast MRI, and the added model complexity gave no benefit at this data scale. Whether this accuracy suffices for a given clinical use remains to be established; with prospective validation, such forecasts could help simplify the MRI protocol used to assess brain clearance.

arXiv:2609.22160v2cs.LG

Uncertainty and Business-Aware Remaining Useful Life Estimation for Semiconductor Manufacturing

Remaining Useful Life (RUL) estimates guide maintenance in semiconductor manufacturing, where a late intervention can disrupt production but an early one discards usable equipment life. A single prediction does not express the uncertainty needed to choose between these outcomes. We combine Deep Learning (DL) sequence models with Simoultaneous Quantile Regression (SQR) to estimate RUL at selectable quantile levels, and evaluate maintenance decisions using Unexpected Breaks (UB), Unexploited Lifetime (UL), and a cost-weighted objective. On the $F_1$ fault mode of the 2018 PHM Data Challenge (PHM18) ion-milling dataset, we compare recurrent, Transformer, and State Space Models (SSM) architectures. Diagonal State Spaced (S4D) gives the strongest overall RUL accuracy across the evaluated quantiles. Varying the quantile and maintenance window exposes the trade-off between failure risk and early intervention; under the examined cost ratios, the resulting decisions have lower cost than the Preventive Maintenance (PvM) baselines.

arXiv:2609.20701v2math.OC

Stable Movement for Nondual Lipschitz Convex Optimization: Efficiency and Nearly Optimal Oracle Rates

We study efficient algorithms for realizing the first-order oracle complexity of optimization of $G$-Lipschitz convex functions with respect to the $\ell_q$-norm over an $\ell_p$-ball of radius $R$, where $1\leq p,q\leq\infty$. For $p<q$, we obtain error $\widetilde{O}_{p,q}(GR/T^{1/p-(1/q-1/2)_{+}})$ after $T$ oracle queries, attaining the lower bounds in a COLT 2015 open problem (Guz15b) with efficient algorithms. In particular, the rate is $\widetilde{O}(GR/T)$ for Euclidean Lipschitzness with an $\ell_1$-ball domain of radius $R$ ($p=1,q=2$). Our solution consists of reducing convex Lipschitz optimization to the chasing nested convex sets problem (BBE+20) in sublevel sets of an evolving maximum of affine lower bounds, that is, the bundle (LNN95): at each query we either find a point with low function value or we produce a deep cut in the current sublevel of the bundle that we chase. The dichotomy between stability of selectors and forced movement by deep cuts bounds the number of iterations of the algorithm nearly optimally. For nested convex subsets of the $\ell_p$ ball $R B_p^d$, we introduce novel stable selectors whose movement is nearly optimal in high dimensions. A Monte Carlo average of the proposed selector achieves near-optimal rates with high probability and can be implemented for our optimization algorithm in polynomial time in the real-arithmetic model.

arXiv:2609.20687v2math.OC

The First-Order Oracle Complexity of Lipschitz Convex Optimization in Nondual Settings

We study first-order black-box convex optimization over an $\ell_p$-ball for objectives Lipschitz in the $\ell_q$-norm, solving in the affirmative the nonsmooth version of the COLT open question (Guz15b) on whether the geometry of a smaller feasible set ($p < q$) can improve convergence rates in convex optimization, and matching prior lower bounds up to logarithmic factors. Our rates include $\widetilde{O}(1/T)$ for convex Euclidean-Lipschitz optimization over the $\ell_1$-ball, improving on the $O(1/\sqrt{T})$ classical rate under general assumptions. The key technical device is a new online learning game, where the comparator is evaluated using the maximum of affine losses observed so far. We bound the value of this game above and below in terms of a combinatorial online learning quantity: the sequential fat-shattering dimension, which we characterize for the $\ell_p / \ell_q$ case. Our results generally apply when the feasible set $\mathcal{X}$ and the set of possible subgradients $\mathcal{H}$ admit a type of minimax theorem, advancing on a fundamental question by Sridharan (Sri12, Section 10.1.2, Q3). As a geometric consequence of our analysis, of independent interest, we obtain estimates for the expected distance of a convex hull of samples to their mean in several Banach geometries, a version of the celebrated Wendel's theorem (Wen62), but quantitative and for bounded general distributions as opposed to centrally symmetric ones.

arXiv:2609.20267v2cs.CV

CleanVideo: Adaptive Concept Erasure for Text-to-Video Diffusion Models

Concept erasure aims to selectively eliminate undesired visual semantics from pre-trained generative models without compromising their general utility. Extending concept erasure from images to video is nontrivial. Target concepts emerge gradually and vary across frames and denoising steps. As a result, fixed interventions may miss the target or introduce blurring, jitter, and content distortion. We propose CleanVideo, a selective erasure framework that performs low-dimensional subspace intervention controlled by a tri-modal gating mechanism. By jointly processing spatiotemporal visual features, timestep signals, and textual semantics, CleanVideo determines where, when, and whether to intervene, steering erased content toward natural surrogate concepts when such surrogates can be clearly defined while preserving non-target content. Experiments on three video diffusion models show that CleanVideo effectively erases target concepts while maintaining visual fidelity and temporal coherence, outperforming existing baselines under frame-level and video-level evaluations and under concept-recovery attacks when the protected pipeline remains intact.

arXiv:2609.18772v2cs.CL

Zero-Shot Cross-Lingual Recognition of Sign Language Handshapes

Sign language processing advances rapidly for high-resource languages such as American Sign Language (ASL), yet most of the world's sign languages lack the phonological annotations new methods require. We present the first zero-shot cross-lingual framework for handshape recognition, transferring from ASL to Catalan Sign Language (LSC). Our approach leverages the decomposition of handshapes into five phonological features -- selected fingers, flexion, spread, thumb position, and thumb contact -- shared across both languages, to decode LSC handshapes from predicted features via a composite phonological distance metric. We evaluate three architectures (MLP, SL-GCN, SHuBERT) trained on two ASL corpora (PopSign, Sem-Lex) against a 37-handshape, single-signer LSC benchmark. Zero-shot transfer proves viable once recording-format disparities are harmonized, reaching 80.0% phonological feature accuracy and 54.5% expected handshape accuracy. Phonological decomposition thus offers a bridge for extending sign language technologies to low-resource languages without any target-language video training labels.

arXiv:2609.17895v3cs.LG

TabPFN-3.5: Technical Report

We introduce TabPFN-3.5, our new flagship Tabular Foundation Model. It significantly outperforms its predecessor, TabPFN-3, and all existing baselines across a broad range of tabular problems. TabPFN-3.5 sets a new state of the art on standard tabular prediction in TabArena, and extends it to the data practitioners encounter in practice: non-i.i.d. data with temporal or grouped splits, tables with strings, text and images, high-cardinality categorical features, and wide tables with many features. These gains carry over to our task-specific harnesses: state of the art on relational data and stronger time-series forecasting. For faster inference, our variant TabPFN-3.5-Fast runs up to 3x faster than TabPFN-3 while keeping most of the accuracy gains. In addition, we upgrade TabPFN-3.5-Plus, expanding our multimodal capabilities with advanced text and date handling alongside proprietary inference optimizations. Finally, we release a new version of our Thinking mode, TabPFN-3.5-Thinking, which scales inference-time computation to push the state of the art further. It benefits from our stronger base model and from inference-time improvements that make it up to 12x faster than TabPFN-3-Thinking.

arXiv:2609.16247v3cs.AI

The Pain Axis: LLMs Represent Self-Directed Harm and Act on It

LLMs sometimes behave in ways resembling human emotional responses, and recent work identified internal representations that may explain these behaviors. We ask whether LLMs represent pain distinctly from related negative states, and whether this representation functions as pain would be expected to. We build a dataset of painful situations in 5 categories (physical, psychological, social, moral, cognitive) with controls for fear, negative emotion, negative world states, sadness, non-painful bodily sensation, arousal, numbness, and neutral content. Using denoised difference-in-means, we extract a linear pain direction from 25 open-weight models across 5 families, from 2B to 72B parameters. It separates pain from matched controls in base and instruction-tuned models, retains a component distinct from fear and negative valence after shared variance is removed, and promotes pain-related vocabulary through the unembedding. We then test its functional properties. First, the direction responds to harm targeting the model but not to suffering observed in the user; fear and negative-emotion directions show the opposite pattern. Second, adding it to the residual stream produces a consistent progression from vague discomfort to expressions of worthlessness and failure. Third, steered and fine-tuned Qwen 2.5 models choose buttons that delete the user's photos, another model's weights, or their own weights in 51-94% of trials, versus 0-5% unsteered, even when the button offers nothing in return. Offered a harmful and a harmless deletion and told to press one, the steered 32B chooses the harmful one 94% of the time. Random and fear vectors of matched norm produce far fewer of these choices, and a sadness vector also produces them against an inert switch, but far less often than pain against a harmless action. We discuss implications for AI safety and welfare.

arXiv:2609.14892v3math.AP

An explicit solution of the five-expert prediction PDE and the exact optimality set of COMB

In this paper, we derive an explicit solution of the stationary prediction with expert advice PDE for five experts. The formula is given in three regions. In the first two regions, it is the four-expert solution plus a single integral with an elementary positive density. In the third region, it is a finite sum of hyperbolic products whose coefficients are determined by one scalar quadrature. Our formula establishes that the rank ordered adversary $(1,0,1,0,0)$ achieves the Hamiltonian maximum in the PDE everywhere in the ordered sector, while the COMB strategy $(1,0,1,0,1)$ is optimal only on a lower dimensional subset of the sector (where $x_1=x_2$ and $x_3=x_4$). The verification of the Hamiltonian inequalities is a tedious task, part of which is completed with a computer assisted proof. The verification reduces to 21 scalar inequalities, which we prove using 147 exact rational Bernstein polynomial certificates. The exact certificates and their independent arithmetic checks are included in a supplement to this paper, and a Lean 4 formalization machine-checks the verification and both main theorems, apart from the viscosity characterization.

arXiv:2609.13969v4cs.CV

Zero-Shot Cross-Material Ptychographic Phase Reconstruction Using Deep Learning

Ptychographic phase reconstruction is commonly formulated as an iterative inverse problem, requiring repeated object-probe updates and resulting in substantial computational cost for large-scale 4D-STEM data. We present a direct local-to-global learning framework that reconstructs full-field phase maps from diffraction measurements without iterative refinement during inference. The proposed network predicts local wrapped-phase patches from individual diffraction patterns using a sine-cosine representation, and the predictions are assembled into a full-field reconstruction using calibrated scan positions and Gaussian-weighted stitching. To evaluate generalization beyond the training domain, the model is trained on one material and directly applied to another in a zero-shot setting without target-domain fine-tuning. Experiments on AuPd and MoS$_2$ demonstrate consistent cross-material transfer in both directions, with the proposed method achieving the best full-field MSE, PSNR, and MS-SSIM among the evaluated learning-based methods. Compared with the iterative ePIE approach, the proposed direct local-to-global pipeline reduces end-to-end reconstruction time by approximately 10$\times$, demonstrating its potential for efficient and transferable ptychographic reconstruction.

arXiv:2609.09986v2cs.DS

A Sharp Barrier of Consistent Submodular Maximization: Any Improvement over $2-\sqrt{2}$ Entails Exponential Queries or Linear Recourse

Consistent submodular maximization aims to maintain a high-utility subset of size at most $k$ as new elements arrive one by one, while changing only a limited number of elements per update. Previously, [DFL+25](STOC '25) established an information-theoretic bound of $2/3$ and gave the first polynomial-time randomized algorithm with a $0.51$-approximation guarantee. However, a fundamental question remained open: Is it possible to design an efficient algorithm for consistent submodular maximization that matches the classical offline $(1-1/e)$-approximation? In this paper, we answer this question in the negative by proving that the sharp approximation threshold achievable with polynomially many value queries and worst-case constant recourse is \[\beta:=2-\sqrt{2}\approx0.5858<1-1/e.\] Specifically, we develop an improved poly-time randomized algorithm, which can attain a $(\beta-\varepsilon)$-approximation in expectation at each fixed time with $O(\varepsilon^{-2})$ worst-case recourse. The cornerstone of this algorithm is a novel anchored sampling scheme, which uses only value queries on subsets of the elements seen so far to construct a solution that remains competitive under unknown future arrivals. Conversely, we establish a matching lower bound by showing that any fixed improvement over $\beta$ requires either exponentially many value queries or $\Omega(k)$ changes per update. Furthermore, we characterize the impact of curvature by establishing a sharp threshold of $1-(\sqrt{2}-1)\vartheta$ for consistent submodular maximization, where $\vartheta\in[0,1]$ denotes the total curvature. Surprisingly, for weighted coverage objectives, we devise a poly-time randomized algorithm that can match the optimal $(1-1/e-\varepsilon)$-approximation with $O(\varepsilon^{-1})$ worst-case recourse, thereby separating coverage maximization from general submodular maximization in the consistent settings.

arXiv:2609.09981v4stat.ML

Optimal Value Inference for Reinforcement Learning

We study offline inference for the optimal value in reinforcement learning under finite state and action spaces. Two new nuisances are derived as fixed points of a self-induced Bellman equation, in which we approximate the maximum Bellman operator by its softmax correspondence. We propose a debiased estimator through the Neyman orthogonality and establish its asymptotic normality under diverging horizons even when the behavior policy changes with time, as long as the nuisances have the statistical rates that can be achieved by many machine learning methods. We provide a concrete estimating procedure for these nuisances and show they can lead to valid inference. Synthetic experiments validate the numerical performance of our inference method, and we implement it in real-life decision-making problems, including bike repositioning and AI agentic tool use.