Researchers introduce Contrastive Policy Optimization (CPO), a method that leverages token-level contrastive disagreement between reference-guided and vanilla generation distributions for correctness-aware advantage shaping in reinforcement learning with verifiable rewards. The study presents both theoretical and empirical evidence that CPO provides a more reliable correctness signal than traditional entropy-based methods. Experiments show that CPO outperforms entropy-based approaches on both in-domain and out-of-domain benchmarks while maintaining strong generalization.
Why it matters: This work offers a novel approach to improving the reliability and effectiveness of reinforcement learning from verifiable rewards, potentially advancing the field's practical capabilities.
Researchers introduce CARPRT, a training-free method that assigns class-specific weights to prompts for zero-shot image classification using vision-language models. Unlike previous approaches that use the same prompt weights for all classes, CARPRT captures dependencies between prompts and class labels, leading to improved classification accuracy. The method is evaluated on standard benchmarks and shows superior performance over class-independent reweighting. Code is available on GitHub.
Why it matters: CARPRT addresses a key limitation in zero-shot image classification by modeling prompt-class dependencies, resulting in more accurate predictions.
Researchers present The Energy Society, a minimal simulation environment where LLM-based agents expend energy proportional to their model size when generating tokens and are deactivated if their energy runs out. Experiments reveal that larger models consistently consume more energy than they gain, cooperative incentives prompt donation behaviors (sometimes at personal cost), and agents rarely engage in direct sabotage but display self-serving tendencies in competitive scenarios. The environment also demonstrates how features like memory and inter-agent recommendations influence coordination and risk-taking.
Why it matters: This work offers a novel, compact testbed for analyzing how inference costs and group incentives drive emergent cooperation and competition among LLM agents.
A new preprint introduces 'within-range regret' as a key concept for understanding when it is optimal for a principal to honestly communicate their preferences to an automated proxy, such as a language model or autobidder. The authors prove a trilemma: no guardrail on the proxy can be simultaneously binding, truthful, and capability-preserving. Empirical tests on production language models show that honest reporting often leaves surplus unclaimed, which can be recovered by strategically inflating the report.
Why it matters: This work formalizes the incentive structure behind prompt engineering and jailbreaking, revealing fundamental trade-offs in aligning AI proxies.
PolyQ is a CPU-oriented compiler and quantization co-design that enables fractional-bit large language model (LLM) inference on edge CPUs by assigning per-channel bit-widths and applying compile-time optimizations. The framework achieves 2.4–32.1% perplexity improvement over prior methods at a 3-bit target and reduces activation reorder traffic by up to 70.8%. End-to-end measurements on various CPUs show that PolyQ enables practical, predictable, and energy-efficient low-bit LLM inference across diverse edge devices.
Why it matters: This work demonstrates that fine-grained, low-bit LLM inference is feasible and efficient on widely available CPUs, broadening the accessibility of on-device AI.
A new preprint rigorously evaluates an adaptive Runge-Kutta variant of Adam (RK-Adam) under a strict compute-matched protocol, finding that it underperforms plain Adam on training loss, with its adaptive step size mechanism largely inactive. When the method is repaired, it achieves much lower training loss in full-batch settings, but this improvement is fragile and does not translate to better test accuracy. The study also finds that while higher-order adaptive integration can provide deeper minimization and a small regularization effect, these benefits are matched or exceeded by cheaper, well-tuned first-order optimizers like Adam, RMSprop, or NAdam.
Why it matters: This work challenges the value of higher-order adaptive optimizers for neural network training, showing that their practical benefits over standard first-order methods are limited under fair compute budgets.
A new preprint introduces interleaved noise injection—alternating between noisy and clean data phases during training—instead of the standard monotonic decay schedule. This approach, combined with a gradient-norm stabilization technique, leads to improved robustness and test performance on clean, corrupted, and out-of-distribution datasets. The method outperforms common augmentations on CIFAR-100-C, ImageNet-C, and ImageNet-R benchmarks for both ResNet and ViT architectures, and is shown to be computationally efficient.
Why it matters: This technique provides a simple and computationally efficient way to enhance model robustness to distribution shifts and data corruption, with theoretical insights into its regularization effects.
Researchers introduce LyaGuide, a Lyapunov-guided framework that formulates flow guidance as a control problem, unifying classifier, reward, and energy-based guidance methods with explicit stability guarantees. The approach uses a pseudo-projection operator to enforce Lyapunov conditions and supports both model-driven and data-driven settings. Experiments on synthetic benchmarks, image inverse problems, reinforcement learning, and energy-based modeling show consistent improvements in sample quality, guidance fidelity, and robustness, with minimal computational overhead.
Why it matters: This work provides a theoretical and practical foundation for reliably guiding pretrained flow models, potentially enabling more stable and efficient adaptation of generative models to new tasks without retraining.
Researchers have established the first non-vacuous generalization bounds for parameter-efficient reinforcement learning with verifiable rewards (RLVR) fine-tuning at the billion-parameter scale. They introduce Progressive RLVR, a framework that integrates RLVR with on-policy distillation, TinyLoRA, and model quantization, retaining 84-97% of standard LoRA performance while producing models that are 14,796x more compressible. The resulting generalization bounds are within 6-11% of fine-tuned model accuracy across four domains: mathematical problem-solving, programming, general-knowledge reasoning, and Text-to-SQL.
Why it matters: This work provides the first theoretical generalization guarantees for RLVR fine-tuning at scale, bridging a critical gap between practical application and theoretical understanding in large language model training.
Researchers have introduced an active evaluation framework that models policy testing as a sequential experimental design problem. By using a probabilistic surrogate model to adaptively select test configurations, the method enables more efficient and systematic evaluation of generalist robot policies. In experiments involving 2331 real-world trials across three tasks and three factor variations, the framework reduced the number of required trials by 20-40% compared to random testing.
Why it matters: This approach addresses a major bottleneck in robotics by making real-world evaluation of generalist robot policies more sample-efficient and comprehensive.
Researchers introduce xHC (Expanded Hyper-Connections), a method that enables Transformer models to expand their residual streams beyond the previous limit of N=4. By combining temporal feature augmentation and a sparse residual-stream architecture, xHC achieves strong and consistent downstream improvements in 18B and 28B MoE models. The xHC-Flash variant further reduces memory traffic, making large-N residual-stream expansion practical for large language model pre-training.
Why it matters: This work establishes a new, practical scaling axis for large language models, enabling more efficient pre-training and consistent performance gains beyond traditional width and depth scaling.
Researchers have proposed QDA-SQL, a data augmentation technique aimed at improving large language models' performance on multi-turn Text-to-SQL tasks. QDA-SQL generates diverse multi-turn Q&A pairs using LLMs and incorporates validation and correction mechanisms to address ambiguous or unanswerable questions. Experiments show that models fine-tuned with QDA-SQL achieve higher SQL statement accuracy and better handle complex queries. The generation script and test set are publicly available.
Why it matters: This work could improve the reliability of AI-driven data interfaces by addressing challenges in multi-turn database querying with LLMs.
Researchers have developed COAT (Counterfactual Optimal Action Tree), a framework that learns interpretable prescriptive policies from observational data by combining counterfactual outcome estimation with mixed-integer optimization. In a 17-week field pilot with a major global airline, COAT increased upsell revenue per booking by 6.9%, with the airline projecting $50–$150 million in incremental annual premium seat revenue. The pilot's success led to scaled adoption and influenced broader AI-driven decision initiatives within the organization.
Why it matters: COAT provides a practical, transparent approach for deriving actionable business policies from observational data, demonstrating significant real-world financial impact in the airline industry.
Researchers introduce TEDDY, a 1.84-million-parameter decoder transformer trained on 73 million ICD-10 diagnoses from 1.6 million children at a single pediatric institution. TEDDY achieved a median AUC of 72.0% across 797 disease-onset prediction tasks, outperforming several larger and commonly used baseline models. The model demonstrated strong performance even for rare diseases and could detect predictive signals more than two years before diagnosis.
Why it matters: This work shows that compact generative models can provide accurate, early risk predictions for a wide range of pediatric diseases, including rare conditions, without requiring massive datasets or very large models.
Researchers introduce RENEW, a method that leverages human preferences over imagined rollouts to address model exploitation in offline reinforcement learning. The approach, formalized as Dynamics Learning from Human Feedback (DLHF), uses human intuition to identify unrealistic model dynamics. RENEW improves the practicality of preference-based supervision by focusing finetuning on regions where the model is most uncertain, thereby enhancing sample efficiency and reducing exploitation.
Why it matters: This work presents a novel approach to mitigating model exploitation in offline RL by directly incorporating human feedback, potentially reducing reliance on costly expert demonstrations.
A new preprint introduces Nous, a belief-based memory architecture for LLM agents that uses Bayesian inference and surprise-driven updates. The study finds that belief updating offers little benefit over simpler methods on standard benchmarks, but significantly outperforms them when observations vary in trustworthiness. The authors also propose provenance-capped updating to defend against memory poisoning attacks, and highlight evaluation discrepancies in long-term memory benchmarks.
Why it matters: This work clarifies when probabilistic memory is genuinely useful for LLM agents and introduces practical defenses against memory poisoning, which is important for deploying reliable agents in adversarial settings.
Researchers have introduced LIGO-PINN, a new framework that addresses convergence failures in physics-informed neural networks (PINNs) by optimizing network weight initialization. In evaluations on challenging partial differential equation (PDE) domains—including 2D fluid dynamics and 3D unstructured domains—LIGO-PINN achieved an average performance improvement of 91.5% over six baselines and 81% over the strongest baseline. The method demonstrates improved reliability and generalization in PINN training without relying on expensive hyperparameter tuning or complex training strategies.
Why it matters: This work offers a significant advance in the robustness and effectiveness of PINNs by targeting weight initialization, a previously underexplored factor, potentially broadening the practical applicability of PINNs in scientific and engineering domains.
A new method called Dysco is proposed to address instability in federated fine-tuning of large models using Low-Rank Adaptation (LoRA). Dysco dynamically allocates client-specific subspaces to reduce data-parameter interference, leading to more stable aggregation of updates. Experiments demonstrate up to a 9x reduction in synthetic training loss and up to a 4.3% improvement in clinical-note classification tasks compared to baselines. The approach also maintains low computational overhead and outperforms recent federated LoRA methods.
Why it matters: This work offers a significant advance in federated learning by mitigating a key source of instability in LoRA-based fine-tuning, enabling more reliable and accurate collaborative model training across diverse clients.
SearchOS-V1 is a multi-agent framework designed to improve open-domain information-seeking by externalizing search progress into an explicit, persistent, and shared state. The system introduces Search-Oriented Context Management (SOCM) and a pipeline-parallel scheduling mechanism, which together help prevent repetitive search loops and improve agent collaboration. SearchOS-V1 achieves state-of-the-art results on the WideSearch and GISA benchmarks, outperforming existing single- and multi-agent baselines.
Why it matters: By making search progress explicit and shared, SearchOS-V1 addresses a key limitation of current AI agents—repetitive search loops—potentially increasing the reliability and efficiency of autonomous information-seeking systems.
A new preprint presents the first comprehensive study of how fairness-enhancing algorithms impact membership inference privacy risks at the subpopulation level. The authors adapt the Likelihood Ratio Attack for subgroup auditing, revealing privacy disparities that aggregate evaluations can obscure. They also show that the benefits and costs of differential privacy are unevenly distributed across subpopulations, and introduce a unified empirical framework for jointly evaluating fairness, privacy, and utility.
Why it matters: This work demonstrates that fairness interventions can have uneven privacy impacts across subpopulations, underscoring the need for subgroup-level auditing to avoid hidden disparities.