This paper studies reinforcement learning from human feedback (RLHF) for aligning large language models with human preferences. While RLHF has demonstrated promising results, many algorithms are highly sensitive to misspecifications in the underlying preference model (e.g., the Bradley-Terry model), the reference policy, or the reward function, resulting in undesirable fine-tuning. To address model misspecification, we propose a doubly robust preference optimization algorithm that remains consistent when either the preference model or the reference policy is correctly specified (without requiring both). Our proposal demonstrates superior and more robust performance than state-of-the-art algorithms, both in theory and in practice. The code is available at https://github.com/DRPO4LLM/DRPO4LLM

We investigate the finite-time convergence properties of Temporal Difference (TD) learning with linear function approximation, a cornerstone algorithm in reinforcement learning. While prior work has established convergence guarantees, these results typically rely on the assumption that each iterate is projected onto a bounded set or that the learning rate is set according to the unknown strong convexity constant -- conditions that are both artificial and do not match the current practice. In this paper, we challenge the necessity of such assumptions and present a refined analysis of TD learning. We show that the simple projection-free variant converges with a rate of $\tilde{\mathcal{O}}(\frac{||θ^*||^2_2}{\sqrt{T}})$, even in the presence of Markovian noise. Our analysis reveals a novel self-bounding property of the TD updates and exploits it to guarantee bounded iterates.

Estimating epistemic uncertainty in value functions is a crucial challenge for many aspects of reinforcement learning (RL), including efficient exploration, safe decision-making, and offline RL. While deep ensembles provide a robust method for quantifying value uncertainty, they come with significant computational overhead. Single-model methods, while computationally favorable, often rely on heuristics and typically require additional propagation mechanisms for myopic uncertainty estimates. In this work we introduce universal value-function uncertainties (UVU), which, similar in spirit to random network distillation (RND), quantify uncertainty as squared prediction errors between an online learner and a fixed, randomly initialized target network. Unlike RND, UVU errors reflect policy-conditional value uncertainty, incorporating the future uncertainties any given policy may encounter. This is due to the training procedure employed in UVU: the online network is trained using temporal difference learning with a synthetic reward derived from the fixed, randomly initialized target network. We provide an extensive theoretical analysis of our approach using neural tangent kernel (NTK) theory and show that in the limit of infinite network width, UVU errors are exactly equivalent to the variance of an ensemble of independent universal value functions. Empirically, we show that UVU achieves equal performance to large ensembles on challenging multi-task offline RL settings, while offering simplicity and substantial computational savings.

In recent years, deep reinforcement learning (DRL) algorithms have gained traction in home energy management systems. However, their adoption by energy management companies remains limited due to the black-box nature of DRL, which fails to provide transparent decision-making feedback. To address this, explainable reinforcement learning (XRL) techniques have emerged, aiming to make DRL decisions more transparent. Among these, soft differential decision tree (DDT) distillation provides a promising approach due to the clear decision rules they are based on, which can be efficiently computed. However, achieving high performance often requires deep, and completely full, trees, which reduces interpretability. To overcome this, we propose a novel asymmetric soft DDT construction method. Unlike traditional soft DDTs, our approach adaptively constructs trees by expanding nodes only when necessary. This improves the efficient use of decision nodes, which require a predetermined depth to construct full symmetric trees, enhancing both interpretability and performance. We demonstrate the potential of asymmetric DDTs to provide transparent, efficient, and high-performing decision-making in home energy management systems.

--Advanced wireless networks must support highly dynamic and heterogeneous service demands. Open Radio Access Network (O-RAN) architecture enables this flexibility by adopting modular, disaggregated components, such as the RAN Intelligent Controller (RIC), Centralized Unit (CU), and Distributed Unit (DU), that can support intelligent control via machine learning (ML). While deep reinforcement learning (DRL) is a powerful tool for managing dynamic resource allocation and slicing, it often struggles to process raw, unstructured input like RF features, QoS metrics, and traffic trends. These limitations hinder policy generalization and decision efficiency in partially observable and evolving environments. T o address this, we propose ORAN-GUIDE, a dual-LLM framework that enhances multi-agent RL (MARL) with task-relevant, semantically enriched state representations. The architecture employs a domain-specific language model, ORANSight, pretrained on O-RAN control and configuration data, to generate structured, context-aware prompts. These prompts are fused with learnable tokens and passed to a frozen GPT -based encoder that outputs high-level semantic representations for DRL agents. This design adopts a retrieval-augmented generation (RAG) style pipeline tailored for technical decision-making in wireless systems. Experimental results show that ORAN-GUIDE improves sample efficiency, policy convergence, and performance generalization over standard MARL and single-LLM baselines. ETWORK slicing is a critical technology for enabling flexible and intelligent radio access networks (RAN) in next-generation wireless networks, as it allows efficient support for highly dynamic and heterogeneous service demands such as enhanced Mobile Broadband (eMBB), Ultra-Reliable Low-Latency Communications (URLLC), and massive Machine-Type Communications (mMTC) [1].

Tables are a fundamental structure for organizing and analyzing data, making effective table understanding a critical capability for intelligent systems. While large language models (LMs) demonstrate strong general reasoning abilities, they continue to struggle with accurate numerical or symbolic reasoning over tabular data, especially in complex scenarios. Spreadsheet formulas provide a powerful and expressive medium for representing executable symbolic operations, encoding rich reasoning patterns that remain largely underutilized. In this paper, we propose Formula Tuning (Fortune), a reinforcement learning (RL) framework that trains LMs to generate executable spreadsheet formulas for question answering over general tabular data. Formula Tuning reduces the reliance on supervised formula annotations by using binary answer correctness as a reward signal, guiding the model to learn formula derivation through reasoning. We provide a theoretical analysis of its advantages and demonstrate its effectiveness through extensive experiments on seven table reasoning benchmarks. Formula Tuning substantially enhances LM performance, particularly on multi-step numerical and symbolic reasoning tasks, enabling a 7B model to outperform OpenAI o1 on table understanding. This highlights the potential of formula-driven RL to advance symbolic table reasoning in LMs.

Reinforcement learning (RL) has driven significant progress in robotics, but its complexity and long training times remain major bottlenecks. In this report, we introduce FastTD3, a simple, fast, and capable RL algorithm that significantly speeds up training for humanoid robots in popular suites such as HumanoidBench, IsaacLab, and MuJoCo Playground. Our recipe is remarkably simple: we train an off-policy TD3 agent with several modifications -- parallel simulation, large-batch updates, a distributional critic, and carefully tuned hyperparameters. FastTD3 solves a range of HumanoidBench tasks in under 3 hours on a single A100 GPU, while remaining stable during training. We also provide a lightweight and easy-to-use implementation of FastTD3 to accelerate RL research in robotics.

Reinforcement learning (RL) has significantly advanced the reasoning capabilities of vision-language models (VLMs). However, the use of RL beyond reasoning tasks remains largely unexplored, especially for perceptionintensive tasks like object detection and grounding. We propose V-Triune, a Visual Triple Unified Reinforcement Learning system that enables VLMs to jointly learn visual reasoning and perception tasks within a single training pipeline. V-Triune comprises triple complementary components: Sample-Level Data Formatting (to unify diverse task inputs), Verifier-Level Reward Computation (to deliver custom rewards via specialized verifiers) , and Source-Level Metric Monitoring (to diagnose problems at the data-source level). We further introduce a novel Dynamic IoU reward, which provides adaptive, progressive, and definite feedback for perception tasks handled by V-Triune. Our approach is instantiated within off-the-shelf RL training framework using open-source 7B and 32B backbone models. The resulting model, dubbed Orsta (One RL to See Them All), demonstrates consistent improvements across both reasoning and perception tasks. This broad capability is significantly shaped by its training on a diverse dataset, constructed around four representative visual reasoning tasks (Math, Puzzle, Chart, and Science) and four visual perception tasks (Grounding, Detection, Counting, and OCR). Subsequently, Orsta achieves substantial gains on MEGA-Bench Core, with improvements ranging from +2.1 to an impressive +14.1 across its various 7B and 32B model variants, with performance benefits extending to a wide range of downstream tasks. These results highlight the effectiveness and scalability of our unified RL approach for VLMs. The V-Triune system, along with the Orsta models, is publicly available at https://github.com/MiniMax-AI.

Large language models are quickly becoming the foundation for intelligent agents that are capable of using tools. However, training such agents is challenging because it requires human creation and annotation of a diverse set of tasks, tools, and evaluation criteria. In this paper, we propose the Self-Challenging framework for training an agent on high-quality tasks that are generated by itself. The agent first plays the role of challenger and generates a task after interacting with the given tools. The tasks take the form of a novel general class of problems termed Code-as-Task, which are defined by an instruction, a verification function and solution and failure cases which serve as tests, allowing to filter only for high-quality tasks. The agent then takes an executor role and trains on those tasks with reinforcement learning using the evaluation feedback as a reward. Evaluation on two existing multi-turn tool-use agent benchmarks, M3ToolEval and TauBench, shows the Self-Challenging framework achieves over a two-fold improvement in Llama-3.1-8B-Instruct, despite using only self-generated training data.

The Soft Actor-Critic (SAC) algorithm, a state-of-the-art method in maximum entropy reinforcement learning, traditionally relies on minimizing reverse Kullback-Leibler (KL) divergence for policy updates. However, this approach leads to an intractable optimal projection policy, necessitating gradient-based approximations that can suffer from instability and poor sample efficiency. This paper investigates the alternative use of forward KL divergence within SAC. We demonstrate that for Gaussian policies, forward KL divergence yields an explicit optimal projection policy -- corresponding to the mean and variance of the target Boltzmann distribution's action marginals. Building on the distinct advantages of both KL directions, we propose Bidirectional SAC, an algorithm that first initializes the policy using the explicit forward KL projection and then refines it by optimizing the reverse KL divergence. Comprehensive experiments on continuous control benchmarks show that Bidirectional SAC significantly outperforms standard SAC and other baselines, achieving up to a $30\%$ increase in episodic rewards, alongside enhanced sample efficiency.