The paper introduces the Statistical Godel Machine (SGM), a framework that replaces formal proof requirements with statistical confidence tests to safely allow recursive self-modification in machine learning systems. SGM uses e-values and Hoeffding bounds to certify improvements at a chosen confidence level while managing a global error budget, and introduces Confirm-Triggered Harmonic Spending (CTHS) to allocate risk more efficiently. Experiments on supervised learning, reinforcement learning, and black-box optimization demonstrate that SGM can certify genuine gains on CIFAR-100, reject false improvements on ImageNet-100, and maintain robustness across benchmarks.
By Xuening Wu, Shenqin Yin, Yanlan Kang, Xinhang Zhang, Qianya Xu, Zeping Chen, Wenqiang Zhang
The paper presents a reinforcement learning approach to generate graph ensembles that satisfy a hard assortativity constraint, a measure of degree–degree correlation between adjacent nodes. Unlike traditional soft-constraint methods, the learned policy performs degree-preserving rewiring to meet the exact target, reducing generation cost by at least an order of magnitude while preserving over 98% of configurational diversity. Trained on small graphs, the method generalizes to larger sizes and different topologies, allowing precise control over secondary observables such as the clustering coefficient.
By Hoyun Choi, Junghyo Jo, Deok-Sun Lee
The paper introduces Bidirectional Behavior Prior Distillation (B2PD), a method that uses action‑value priors to train a conditional variational autoencoder for generating high‑value behavior support. These expert behavior priors are then distilled into the online reinforcement learning agent, reducing inefficient exploration and stabilizing policy updates. Experiments on state‑ and pixel‑based tasks show that B2PD improves sample efficiency while maintaining stable learning dynamics.
By Gong Gao, Xiao Lai, Jiaji Shen, Ning Jia, Xianhui Liu, Weidong Zhao
The paper introduces a model-based bootstrap framework for uncertainty quantification in offline policy evaluation (OPE) within finite-horizon, time-inhomogeneous Markov decision processes. Unlike traditional bootstrap methods that resample entire episodes, this approach regenerates trajectories from an estimated MDP, enabling use of diverse offline data formats such as complete trajectories, transition-level observations, and trajectory fragments. The authors prove bootstrap distributional consistency, asymptotically valid confidence intervals, and consistent variance estimation, and demonstrate through simulations that the method yields tighter confidence intervals and more accurate variance estimates compared to existing techniques.
By Weiwei Wang, Yuqiang Li, Xianyi Wu, Bingyi Jing
The paper introduces L2C‑TFM, a reinforcement‑learning framework for cleaning tabular data before feeding it to Tabular Foundation Models (TFMs). It proposes a model‑aware reward that regularizes the Wasserstein distance between cleaned and dirty data, aiming to preserve distributional stability. Experiments on ten OpenML datasets show that while some reward designs fail, the model‑aware reward performs comparably to a random‑forest baseline and improves minority‑class macro‑F1 under class imbalance, and a policy trained on one dataset can transfer to others.
By Laure Berti-Equille
AUDITPLAN introduces a plan-then-answer method for safety alignment in language models, where the model first generates a compact structured safety plan before responding. The plan includes a threat label, intended action, and explicit constraints, allowing machine‑checkable auditing while remaining hidden from end users. Training combines supervised fine‑tuning with reinforcement learning using the FAITHGATE reward, which only rewards correct plans, thereby reducing unsafe shortcuts and improving robustness across Qwen model variants.
By Sai Sri Pushpa Jampani, Kshitij Mishra, Asif Ekbal
The paper introduces Live-ProsodyJudge (LPJ), a cost‑effective pairwise evaluator distilled from Gemini for assessing fine‑grained prosody in live streaming TTS. It identifies a flaw called verdict coupling, where multi‑dimensional scores collapse into a single preference, and proposes Decoupled‑Live‑ProsodyJudge (D‑LPJ) to eliminate this issue through masking and a span‑local GRPO strategy. Experiments show LPJ outperforms a single Gemini call in accuracy, and D‑LPJ provides independent dimension judgments, achieving high alignment with human top‑3 selections in a TTS candidate tournament.
By Zifan Guan, Longyu Lu, Junan Zhang, Zhizheng Wu, Meiguang Jin, Junfeng Ma
The paper investigates how a single pooled contract offered by an aggregator to heterogeneous smallholder farmers can be designed to maximize profit while addressing private adoption costs and unobserved effort over multiple seasons. Using a POMDP framework and reinforcement learning, the authors find that profit‑maximizing contracts disproportionately favor large farms, achieving 87.7% of possible adoption on large farms versus only 8.2% on smallholdings, largely due to higher measurement, reporting, and verification costs on smaller plots. The study suggests that adjusting MRV cost structures could reduce this disparity and help scale carbon farming to smallholders.
By Rishi Bharadwaj, Yadati Narahari
CovR is an agentic framework that automates testbench generation for hardware verification by combining self-reflection loops with simulation-based feedback to maximize coverage. It builds a large dataset of 16,514 specification–RTL reasoning tuples and uses reinforcement learning with tool-derived rewards to train a student model, achieving high coverage scores on VerilogEval, RTLLM V2.0, and CVDP. When deployed as a plug-in stimulus engine, CovR boosts coverage by nearly 19% and improves mutation detection while uncovering previously undetected failures.
By Manar Abdelatty, Maryam Nouh, Sherief Reda
The paper introduces Vision‑RL2, a region‑level reinforcement learning approach that optimizes a lightweight proposal network for fine‑grained multimodal large language model (MLLM) perception. By treating coherent image regions as actions and scoring them with a frozen MLLM reader, the method selectively focuses visual resolution on evidence, reducing token usage while improving accuracy across multiple benchmarks and backbones. The approach eliminates the need for region annotations, response sampling, or reasoning trajectories, and the refined proposals enable sparse encoding that magnifies relevant evidence.
By Yuheng Shi, Xiaohuan Pei, Minjing Dong, Chang Xu
The article introduces a mathematical model of the Motivated Emotional Mind cognitive architecture for embodied intelligent systems. It formalizes a re‑entrant loop that integrates feedforward processing, lateral interactions, and feedback pathways, along with representational selection mechanisms that govern adaptive responses. The model binds exteroceptive and interoceptive signals, bodily-motivational context, and memory traces into associative structures called semblions, enabling motivated learning that incorporates need thresholds, goal generation, and regulatory constraints.
By Wies{\l}aw L. Galus, Janusz A. Starzyk
REARL is a closed‑loop simulation enhancement framework that combines real traffic data with large language models (LLMs) to improve autonomous driving simulations. It clusters real traffic, uses cluster centers as representative scenarios for the LLM, and employs a sliding‑window detector to monitor vehicle speed and spacing discrepancies. When thresholds are exceeded, the LLM adjusts vehicle decision‑making or selects matching real vehicle actions, resulting in lower Hellinger distance and MAPE compared to baselines in a HighD highway setting.
By Xiaojun Bi (Minzu University of China, Beijing, China), Jun Jiang (Minzu University of China, Beijing, China), Yiwen Sun (Peking University, Beijing, China, BIGAI, Beijing, China), Quanyi Ou (Minzu University of China, Beijing, China), Ke Cheng (Beihang University, Beijing, China), Mingjie Bi (BIGAI, Beijing, China), Yexin Li (BIGAI, Beijing, China)
arXiv:2609. 20409v1 Announce Type: new Abstract: Two-timescale stochastic approximation (TTSA) is a fundamental tool for analyzing coupled iterative algorithms in reinforcement learning, optimization, and stochastic control.
By Djamel Rassem Lamouri, Dorian Baudry, Nicolas Gast
GLAMDRING is a framework that jointly designs a quadruped robot’s morphology and its gait controller using reinforcement learning of Hopf-oscillator Central Pattern Generators (CPGs). Given specifications such as forward‑velocity bounds, actuator power budgets, an actuator library, and payload requirements, the system returns an optimized robot design and a corresponding gait policy, ranking designs by objectives like maximum speed, minimum Cost of Transport, or maximum payload margin. Experiments demonstrate that co‑designing body and gait is essential for meeting locomotion constraints, that actuator feasibility determines payload capacity, and that natural animal gaits emerge from the design process, with a real‑world demonstration confirming the approach’s effectiveness.
By Amogh Joshi, Kaushik Roy
arXiv:2601. 00728v5 Announce Type: replace Abstract: We propose a reinforcement learning (RL) framework for \xy{responsive} precision tuning for linear solvers, which can be extended to general algorithms.
By Erin Carson, Xinye Chen
The paper introduces ActObs, a supervised fine‑tuning method that, unlike standard approaches, also predicts environment observations in agent trajectories. While both ActObs and action‑only training perform similarly after initial fine‑tuning, ActObs diverges during subsequent reinforcement learning, yielding higher pass@k scores on several benchmarks and better cross‑domain task performance. The authors attribute this advantage to ActObs’s joint supervision, which preserves observation gradients and prevents the policy from over‑specializing on actions alone.
By Juzheng Zhang, Disha Makhija, Manoj Ghuhan Arivazhagan, Vinayshekhar Bannihatti Kumar, Rashmi Gangadharaiah
UnifiedPlayers is a cooperative framework that jointly adapts planning, execution, and evaluation for tool-integrated reinforcement learning agents. It consists of a Planning Player that generates tasks, an Execution Player that creates multi-turn trajectories with Python tool calls, and an Evaluation Player that builds executable verifiers, all coordinated by role‑specific rewards under GRPO. The approach outperforms prior baselines on mathematical and general reasoning benchmarks and yields a verifier with high adversarial detection accuracy and more discriminative reward signals.
By Wenjie Liao, Liangjie Zhao, Zehong Cao
EPIG-Tree proposes a compute‑optimal branching strategy for gradient‑efficient reinforcement learning, arguing that branches should be placed where they most reduce policy‑gradient uncertainty per unit of compute. By deriving allocation laws from a law‑of‑total‑variance decomposition, the method introduces an EPIG‑Tree score that guides branch placement using already computed rollouts, estimating occupancy‑ and score‑weighted value uncertainty. Empirical results show EPIG‑Tree reduces gradient MSE in cloned‑state control, improves frozen‑LLM gradient calibration, and outperforms flat GRPO and entropy branching in both single‑turn math and multi‑turn Wordle tasks.
By Nikita Khomich, Leopold Hermansson, Ido Hakimi
The paper introduces CURB, a reward‑shaping framework that penalizes the total variation distance between an agent’s action distributions under cooperation and defection histories, thereby preventing collusive equilibria in repeated games. By linking empirical Q‑learning collusion to Simple Penal Codes, the authors prove that any non‑trivial SPC can be neutralized, and demonstrate CURB’s effectiveness in both tabular and deep Q‑learning settings for Bertrand and Cournot competition.
By Karthik Sivachandran, Rohan Paleja
The paper introduces Odds‑Ratio Thompson Sampling (OR‑TS), a method for batched multi‑armed bandits that updates the joint posterior over log‑odds contrasts and refits the shared level in each batch, rather than carrying over absolute reward rates. It presents a Bayesian bandit agent with controls for decay of past evidence and aggressiveness of allocation, and evaluates OR‑TS against traditional absolute‑rate memory across 86 public A/B series and synthetic environments. Results show that when the shared level varies significantly, OR‑TS outperforms absolute‑rate memory, reducing regret and ensuring the best arm receives more traffic, while also handling cases where contrasts themselves shift.
By Sulgi Kim