News & Research

arXiv:2605.27395v1 Announce Type: cross Abstract: As the rapid proliferation of AI systems and harms spurs efforts in AI governance around the world, prioritizing among competing policy options has become increasingly challenging for policymakers and researchers. We introduce a methodology for identifying viable policy options to mitigate specified AI harms, helping policymakers and researchers target areas that warrant greater time and resource investment. This method combines participatory evaluation of policies, expert assessment of implementation costs, and an LLM-based assessment of perceived harm mitigation under each policy option. We leverage a genetic algorithm-based simulation study to explore a vast solution space of potential policy combinations, and examine how outcomes change under different weightings of cost, participatory input, and harm mitigation. We find that this method enables exploration of different balances between participatory and expert components, allowing

arXiv:2605.27912v1 Announce Type: cross Abstract: We study efficient differentially private algorithms for estimating monotone statistics, i.e., statistics that are monotone under the addition of new observations. The starting point for our investigation is subsample-and-aggregate: a classical paradigm that partitions the dataset into blocks, estimates the statistic on each block, and then privately aggregates the estimates.While practical and generically applicable, this approach is quite data-hungry. We improve upon this framework for the class of monotone statistics -- compared to subsample-and-aggregate, our algorithms save a factor of $t$ in sample complexity and pay a factor of $e^t$ in running time, where $t>0$ is a tunable parameter. We complement our results with a query-complexity lower bound, showing that our algorithms are essentially optimal for this task. As an application, we obtain improved results for private eigenvalue estimation, private loss estimation, and privatel

arXiv:2605.28388v1 Announce Type: new Abstract: Reinforcement Learning with Verifiable Reward (RLVR) is empirically shown to notably enhance the reasoning performance of large language models (LLMs), particularly in mathematics and programming. However, the mechanistic role of Sample Difficulty in RLVR remains poorly understood. In this paper, we investigate RLVR through the lens of difficulty-wise and one-sample analysis. We find that sample difficulty has a non-monotonic effect on RLVR: easy and medium-difficulty problems yield the strongest and most stable reasoning improvements, whereas overly hard problems often provide weak learning signals, induce degenerate behaviors such as answer repetition or skipping necessary computation, and can ultimately degrade the model's pre-existing capabilities. Beyond the obverse of response, we further analyze the model's internal feature dynamics using Temporal Sparse Autoencoders (T-SAE). Easy problems mainly reinforce direct-answer and basic-c

arXiv:2603.00349v2 Announce Type: replace Abstract: Many complex tasks require extended effort, diverse capabilities, or coordinated actions beyond what a single agent can provide. However, simply adding more agents does not guarantee better performance, as effective cooperation depends on how agents interact with each other and with task structure to satisfy evolving constraints over time. This challenge is amplified for LLM-based multi-agent systems (LLM-MAS): plans, messages, and revisions occur in natural language, whereas task progress depends on grounded environment actions. Current evaluations mostly treat cooperation as an implicit ingredient of final task success, leaving both cooperation and the effect of multi-agent interaction on task dynamics difficult to study. We introduce COOP$^2$, an evaluation framework that grounds high-level agent cooperation dynamics in LLM-MAS within task progress in the environment. COOP$^2$ then defines cooperative tasks with verifiable cooperat

arXiv:2605.27389v1 Announce Type: cross Abstract: We study how conditioning context shapes personalization behavior in a teacher-facing educational recommender system. We compare contextual conditioning based on the current student question with memory-based conditioning using persistent learner information. Using deviation correlation and paired statistical tests, we find that contextual recommendations exhibit stronger question-level responsiveness, while memory-based recommendations exhibit history-dependent behaviors, including learner-specific differentiation under identical input. Teacher-facing evaluation signals suggest these recommendations are interpretable and actionable. These results indicate that embedding-based similarity metrics capture responsiveness to the current question but do not characterize personalization grounded in learner history, motivating behavior-level diagnostics for studying conditioning effects.

arXiv:2605.27736v1 Announce Type: cross Abstract: Online reinforcement learning is becoming increasingly important for aligning diffusion models with non-differentiable objectives. However, existing methods still face limitations in assigning fine-grained credit along denoising trajectories and in realizing stable value-based optimization. We propose a state-aligned latent actor-critic framework for diffusion post-training, in which the diffusion model serves as its own timestep-conditioned value function and predicts values directly on noisy latent states. This enables trajectory-level PPO training, supports stable actor-critic optimization with simple conditioning and value pretraining strategies, and naturally allows the learned critic to be reused for inference-time steering. We further extend the framework to multi-reward optimization, where joint training with complementary rewards helps alleviate reward hacking. Across both UNet- and DiT-based backbones, our method consistently

arXiv:2605.27393v1 Announce Type: cross Abstract: Large language models (LLMs) can generate fluent dialogue, but prior works lack situational grounding, dynamic strategy control, and evaluation aligned with clinical standards in motivational interviewing (MI). We introduce StoryMI, a multi-LLM agent framework for controllable MI dialogue generation, where questionnaire-based client profiles are expanded into situational stories that provide narrative context for the dialogue. Therapist and client agents generate MI-coded utterances guided by MI codes selected by the interaction agent, while an interaction agent dynamically coordinates exchanges to control MI strategies during a multi-turn conversation. We propose a two-level evaluation protocol: lexical metrics and MI-specific measures of macro-level counseling strategies, alongside LLM-as-judge and human expert assessments. We construct a dataset of 6K simulated MI dialogues grounded in 1K questionnaire-story pairs, covering 12 MI cod

arXiv:2605.28053v1 Announce Type: new Abstract: Test-time training (TTT) adapts an LLM during generation by reading and updating request-owned state, such as fast weights, low-rank deltas, or streaming learner state. This breaks batched LLM serving, which assumes shared static weights: serial execution is correct but slow, while naive batching can corrupt request state. We formulate this problem as read-write TTT serving and present RW-TTT , which tags each decode step with its owner, version, and READ/WRITE effect, batches only compatible phases, and commits updates only to the owner. On one GPU with eight fast-weight InPlace-TTT streams, RW-TTT reaches 274.61 aggregate tok/s, 9.31x over sequential serving and 3.44x over per-stream replicas under the same memory budget. It preserves behavior on RULER, a long-context benchmark, and passes owner/version checks.

arXiv:2605.28775v1 Announce Type: cross Abstract: Computer-use agents (CUAs) have recently made substantial progress, but deploying a separate large expert for each software domain remains expensive. Small open computer-use agents are more practical specialization targets, but they remain substantially weaker and exhibit uneven domain-specific failures. A straightforward remedy is to synthesize large-scale training data for the target domain, yet we find that this naive approach yields only marginal improvements. Building on this observation, we introduce LearnWeak, an annotation-free specialization framework for small computer-use agents that uses a stronger reference agent to identify the student's weaknesses in the target domain, synthesize targeted tasks, and construct supervision automatically. LearnWeak further introduces an error-aware specialization objective that disentangles planning and execution errors, enabling more behaviorally precise updates than broad uniform supervisi

arXiv:2605.27495v1 Announce Type: new Abstract: Data availability remains a critical bottleneck in many deep learning applications. Large-scale datasets are often expensive to collect, curate and annotate, which can limit the scalability and applicability of supervised learning methods. In this work, we evaluate the classification performance of models trained on synthetic image datasets produced by generative deep learning. In particular, we use latent diffusion models conditioned on learned representations from DINOv2, DINOv3, and CLIP. Our results demonstrates that this representation-conditioned formulation significantly outperforms class-conditioned generation by a large margin (+10.76 p.p. top-1 accuracy on ImageNet100), by improving sample quality and mode coverage. Furthermore, by scaling the size of the synthetic dataset, we are able to outperform a classifier trained on the real data (+2.0 p.p top-1 accuracy). We also demonstrate how generated images can be used for augmentat

arXiv:2605.27929v1 Announce Type: cross Abstract: Active sensing links behavior and learning through an action-perception loop: actions determine the observations used to update internal predictive models of perception, which subsequently guide the next actions. Predictive-coding frameworks provide a natural way to model this process, since internal representations are continuously updated to predict future observations. Here, we ask how exploratory and exploitative behavioral strategies shape these internal predictive representations. We build an online learning agent in a tree-like maze with a controllable parameter regulating the balance between exploratory and exploitative regimes. The agent updates a predictive-coding-based perception model from experience generated by its own behavior. The model predicts both future maze states and reward probability, allowing the agent to select actions either by expected information gain during exploration or by predicted reward during exploita

arXiv:2602.12843v2 Announce Type: replace Abstract: Chest X-ray (CXR) reporting follows a region-based clinical workflow in which radiologists inspect anatomical regions and integrate localized findings into a final report. However, existing resources for CXR report generation provide these supervision signals in fragmented forms. We introduce MMRad-22K, a dataset that organizes regional textual observations, anatomical grounding coordinates, localized image evidence, and report targets into structured multimodal evidence units for CXR report generation. To motivate this formulation, we first compare different evidence formats for report generation and find that structured multimodal evidence is generally more useful than text-only or bounding box-based evidence. We then adapt a unified LVLM backbone using MMRad-22K and show that adaptation with multimodal evidence outperforms both textual-evidence adaptation and end-to-end adaptation on language and clinically oriented metrics. Under

arXiv:2605.28501v1 Announce Type: new Abstract: Fitting an unknown number of hyperplanes to data is a fundamental yet challenging problem in machine learning, characterized by its non-convexity, non-differentiability, and unknown model order. Existing approaches often struggle with local optima or lack geometric consistency. To address these limitations, we propose a novel framework based on Manifold Optimization. We reformulate the problem as an unsupervised learning task on the unit sphere manifold $\mathcal{S}^{\textbf{dim}-1}$. This formulation effectively handles the non-convex constraints and linearizes the distance measurement, rendering the gradient descent tractable. We propose a Two-Stage Manifold Optimization algorithm. In Phase I, we employ a Riemannian Expectation-Maximization process with a heavy-tailed kernel to robustly estimate posterior probabilities, effectively resolving the ambiguities of point distribution between intersecting hyperplanes. In Phase II, upon conver

arXiv:2603.09117v3 Announce Type: replace-cross Abstract: Reinforcement Learning from Verifiable Rewards (RLVR) significantly enhances large language models (LLMs) reasoning but severely suffers from calibration degeneration, where models become excessively over-confident in incorrect answers. Previous studies devote to directly incorporating calibration objective into existing optimization target. However, our theoretical analysis demonstrates that there exists a fundamental gradient conflict between the optimization for maximizing policy accuracy and minimizing calibration error. Building on this insight, we propose DCPO, a simple yet effective framework that systematically decouples reasoning and calibration objectives. Extensive experiments demonstrate that our DCPO not only preserves accuracy on par with GRPO but also achieves the best calibration performance and substantially mitigates the over-confidence issue. Our study provides valuable insights and practical solution for more

arXiv:2604.18518v3 Announce Type: replace Abstract: Uniform Discrete Diffusion Model (UDM) has recently emerged as a promising paradigm for discrete generative modeling; however, its integration with reinforcement learning remains largely unexplored. We observe that naively applying GRPO to UDM leads to training instability and marginal performance gains. To address this, we propose UDM-GRPO, the first framework to integrate UDM with RL. Our method is guided by two key insights: (i) treating the final clean sample as the action provides more accurate and stable optimization signals; and (ii) reconstructing trajectories via the diffusion forward process better aligns probability paths with the pretraining distribution. Additionally, we introduce two strategies, Reduced-Step and CFG-Free, to further improve training efficiency. UDM-GRPO significantly improves base model performance across multiple T2I tasks. Notably, GenEval accuracy improves from $69\%$ to $96\%$ and PickScore increases