Scalable Variational Bayesian Fine-Tuning of LLMs via Orthogonalized Low-Rank Adapters

When deploying large language models (LLMs) to safety-critical applications, uncertainty quantification (UQ) is of utmost importance to self-assess the reliability of the LLM-based decisions. However, such decisions typically suffer from overconfidence, particularly after parameter-efficient fine-tuning (PEFT) for downstream domain-specific tasks with limited data. Existing methods to alleviate this issue either rely on Laplace approximation based post-hoc framework, which may yield suboptimal calibration depending on the training trajectory, or variational Bayesian training that requires multiple complete forward passes through the entire LLM backbone at inference time for Monte Carlo estimation, posing scalability challenges for deployment. To address these limitations, we build on the Bayesian last layer (BLL) model, where the LLM-based deterministic feature extractor is followed by random last layer parameters for uncertainty reasoning. Since existing low-rank adapters (LoRA) for PEFT have limited expressiveness due to rank collapse, we address this with Polar-decomposed Low-rank Adapter Representation (PoLAR), an orthogonalized parameterization paired with Riemannian optimization to enable more stable and expressive adaptation. Building on this PoLAR-BLL model, we leverage the variational (V) inference framework to put forth a scalable Bayesian fine-tuning approach which jointly seeks the PoLAR parameters and approximate posterior of the last layer parameters via alternating optimization. The resulting PoLAR-VBLL is a flexible framework that nicely integrates architecture-enhanced optimization with scalable Bayesian inference to endow LLMs with well-calibrated UQ. Our empirical results verify the effectiveness of PoLAR-VBLL in terms of generalization and uncertainty estimation on both in-distribution and out-of-distribution data for various common-sense reasoning tasks.

Diffusion Policy with Bayesian Expert Selection for Active Multi-Target Tracking

Active multi-target tracking requires a mobile robot to balance exploration for undetected targets with exploitation of uncertain tracked ones. Diffusion policies have emerged as a powerful approach for capturing diverse behavioral strategies by learning action sequences from expert demonstrations. However, existing methods implicitly select among strategies through the denoising process, without uncertainty quantification over which strategy to execute. We formulate expert selection for diffusion policies as an offline contextual bandit problem and propose a Bayesian framework for pessimistic, uncertainty-aware strategy selection. A multi-head Variational Bayesian Last Layer (VBLL) model predicts the expected tracking performance of each expert strategy given the current belief state, providing both a point estimate and predictive uncertainty. Following the pessimism principle for offline decision-making, a Lower Confidence Bound (LCB) criterion then selects the expert whose worst-case predicted performance is best, avoiding overcommitment to experts with unreliable predictions. The selected expert conditions a diffusion policy to generate corresponding action sequences. Experiments on simulated indoor tracking scenarios demonstrate that our approach outperforms both the base diffusion policy and standard gating methods, including Mixture-of-Experts selection and deterministic regression baselines.

Approximation of Log-Partition Function in Policy Mirror Descent Induces Implicit Regularization for LLM Post-Training

Policy mirror descent (PMD) provides a principled framework for reinforcement learning (RL) by iteratively solving KL-regularized policy improvement subproblems. While this approach has been adopted in training advanced LLMs such as Kimi K1.5/K2, the ideal closed-form PMD updates require reliable partition function estimation, a significant challenge when working with limited rollouts in the vast action spaces of LLMs. We investigate a practical algorithm, termed PMD-mean, that approximates the log-partition term with the mean reward under the sampling policy and performs regression in log-policy space. Specifically, we characterize the population solution of PMD-mean and demonstrate that it implicitly optimizes mirror descent subproblems with an adaptive mixed KL--$χ^2$ regularizer. This additional $χ^2$ regularization constrains large probability changes, producing more conservative updates when expected rewards are low and enhancing robustness against finite-sample estimation errors. Experiments on math reasoning tasks show that PMD-mean achieves superior performance with improved stability and time efficiency. These findings deepen our understanding of PMD-mean and illuminate pathways toward principled improvements in RL algorithms for LLMs. Code is available at https://github.com/horizon-rl/OpenKimi.

Ask a Strong LLM Judge when Your Reward Model is Uncertain

Reward model (RM) plays a pivotal role in reinforcement learning with human feedback (RLHF) for aligning large language models (LLMs). However, classical RMs trained on human preferences are vulnerable to reward hacking and generalize poorly to out-of-distribution (OOD) inputs. By contrast, strong LLM judges equipped with reasoning capabilities demonstrate superior generalization, even without additional training, but incur significantly higher inference costs, limiting their applicability in online RLHF. In this work, we propose an uncertainty-based routing framework that efficiently complements a fast RM with a strong but costly LLM judge. Our approach formulates advantage estimation in policy gradient (PG) methods as pairwise preference classification, enabling principled uncertainty quantification to guide routing. Uncertain pairs are forwarded to the LLM judge, while confident ones are evaluated by the RM. Experiments on RM benchmarks demonstrate that our uncertainty-based routing strategy significantly outperforms random judge calling at the same cost, and downstream alignment results showcase its effectiveness in improving online RLHF.

Fine-tuning LLMs with variational Bayesian last layer for high-dimensional Bayesian optimzation

A plethora of applications entail solving black-box optimization problems with high evaluation costs, including drug discovery, material design, as well as hyperparameter tuning. Toward finding the global optimum of such black-box optimization problems with sample efficiency, Bayesian optimization (BO) is a theoretically elegant framework that relies on a probabilistic surrogate model so as to iteratively select the query point with well-balanced exploration-exploitation tradeoffs. The Gaussian process (GP), as the de-facto choice for surrogate modeling, has achieved compelling performances for vanilla BO with low-dimensional continuous variables. However, GPs fall short in coping with high-dimensional counterparts with {\it irregular} variables (e.g., categorical, ordinal, etc.). To alleviate this, neural network-based surrogates have been explored. Inspired by the powerful capabilities of LLMs, we adopt the LLM as the surrogate to model the mapping from the high-dimensional input variables to the objective function. To adapt to the current problem, we leverage the low-rank adaptation (LoRA) to fine-tune the LLM parameters together with the posterior of a linear regression head via the variational Bayesian last layer (VBLL) framework. The resulting LoRA-VBLL is not only computationally light compared to existing alternatives, but also admits recursive updates. To automate the critical selection of the LoRA rank as well as other hyperparameters, a weighted ensemble (ENS) of LoRA-VBLL surrogates has been devised, which further accommodates continual update of the per-model weight and individual LoRA-VBLL parameters via recursive Bayes. Extensive experimental results demonstrate the compelling performance of the proposed (ENS-)LoRA-VBLL approaches on various high-dimensional benchmarks and the real-world molecular optimization tasks.

Deploying AI for Signal Processing education: Selected challenges and intriguing opportunities

Powerful artificial intelligence (AI) tools that have emerged in recent years -- including large language models, automated coding assistants, and advanced image and speech generation technologies -- are the result of monumental human achievements. These breakthroughs reflect mastery across multiple technical disciplines and the resolution of significant technological challenges. However, some of the most profound challenges may still lie ahead. These challenges are not purely technical but pertain to the fair and responsible use of AI in ways that genuinely improve the global human condition. This article explores one promising application aligned with that vision: the use of AI tools to facilitate and enhance education, with a specific focus on signal processing (SP). It presents two interrelated perspectives: identifying and addressing technical limitations, and applying AI tools in practice to improve educational experiences. Primers are provided on several core technical issues that arise when using AI in educational settings, including how to ensure fairness and inclusivity, handle hallucinated outputs, and achieve efficient use of resources. These and other considerations -- such as transparency, explainability, and trustworthiness -- are illustrated through the development of an immersive, structured, and reliable "smart textbook." The article serves as a resource for researchers and educators seeking to advance AI's role in engineering education.

WebAgent-R1: Training Web Agents via End-to-End Multi-Turn Reinforcement Learning

While reinforcement learning (RL) has demonstrated remarkable success in enhancing large language models (LLMs), it has primarily focused on single-turn tasks such as solving math problems. Training effective web agents for multi-turn interactions remains challenging due to the complexity of long-horizon decision-making across dynamic web interfaces. In this work, we present WebAgent-R1, a simple yet effective end-to-end multi-turn RL framework for training web agents. It learns directly from online interactions with web environments by asynchronously generating diverse trajectories, entirely guided by binary rewards depending on task success. Experiments on the WebArena-Lite benchmark demonstrate the effectiveness of WebAgent-R1, boosting the task success rate of Qwen-2.5-3B from 6.1% to 33.9% and Llama-3.1-8B from 8.5% to 44.8%, significantly outperforming existing state-of-the-art methods and strong proprietary models such as OpenAI o3. In-depth analyses reveal the effectiveness of the thinking-based prompting strategy and test-time scaling through increased interactions for web tasks. We further investigate different RL initialization policies by introducing two variants, namely WebAgent-R1-Zero and WebAgent-R1-CoT, which highlight the importance of the warm-up training stage (i.e., behavior cloning) and provide insights on incorporating long chain-of-thought (CoT) reasoning in web agents.

Think-RM: Enabling Long-Horizon Reasoning in Generative Reward Models

Reinforcement learning from human feedback (RLHF) has become a powerful post-training paradigm for aligning large language models with human preferences. A core challenge in RLHF is constructing accurate reward signals, where the conventional Bradley-Terry reward models (BT RMs) often suffer from sensitivity to data size and coverage, as well as vulnerability to reward hacking. Generative reward models (GenRMs) offer a more robust alternative by generating chain-of-thought (CoT) rationales followed by a final reward. However, existing GenRMs rely on shallow, vertically scaled reasoning, limiting their capacity to handle nuanced or complex (e.g., reasoning-intensive) tasks. Moreover, their pairwise preference outputs are incompatible with standard RLHF algorithms that require pointwise reward signals. In this work, we introduce Think-RM, a training framework that enables long-horizon reasoning in GenRMs by modeling an internal thinking process. Rather than producing structured, externally provided rationales, Think-RM generates flexible, self-guided reasoning traces that support advanced capabilities such as self-reflection, hypothetical reasoning, and divergent reasoning. To elicit these reasoning abilities, we first warm-up the models by supervised fine-tuning (SFT) over long CoT data. We then further improve the model's long-horizon abilities by rule-based reinforcement learning (RL). In addition, we propose a novel pairwise RLHF pipeline that directly optimizes policies using pairwise preference rewards, eliminating the need for pointwise reward conversion and enabling more effective use of Think-RM outputs. Experiments show that Think-RM achieves state-of-the-art results on RM-Bench, outperforming both BT RM and vertically scaled GenRM by 8%. When combined with our pairwise RLHF pipeline, it demonstrates superior end-policy performance compared to traditional approaches.

Bayesian Optimization for Robust Identification of Ornstein-Uhlenbeck Model

This paper deals with the identification of the stochastic Ornstein-Uhlenbeck (OU) process error model, which is characterized by an inverse time constant, and the unknown variances of the process and observation noises. Although the availability of the explicit expression of the log-likelihood function allows one to obtain the maximum likelihood estimator (MLE), this entails evaluating the nontrivial gradient and also often struggles with local optima. To address these limitations, we put forth a sample-efficient global optimization approach based on the Bayesian optimization (BO) framework, which relies on a Gaussian process (GP) surrogate model for the objective function that effectively balances exploration and exploitation to select the query points. Specifically, each evaluation of the objective is implemented efficiently through the Kalman filter (KF) recursion. Comprehensive experiments on various parameter settings and sampling intervals corroborate that BO-based estimator consistently outperforms MLE implemented by the steady-state KF approximation and the expectation-maximization algorithm (whose derivation is a side contribution) in terms of root mean-square error (RMSE) and statistical consistency, confirming the effectiveness and robustness of the BO for identification of the stochastic OU process. Notably, the RMSE values produced by the BO-based estimator are smaller than the classical Cram\'{e}r-Rao lower bound, especially for the inverse time constant, estimating which has been a long-standing challenge. This seemingly counterintuitive result can be explained by the data-driven prior for the learning parameters indirectly injected by BO through the GP prior over the objective function.

Online scalable Gaussian processes with conformal prediction for guaranteed coverage

The Gaussian process (GP) is a Bayesian nonparametric paradigm that is widely adopted for uncertainty quantification (UQ) in a number of safety-critical applications, including robotics, healthcare, as well as surveillance. The consistency of the resulting uncertainty values however, hinges on the premise that the learning function conforms to the properties specified by the GP model, such as smoothness, periodicity and more, which may not be satisfied in practice, especially with data arriving on the fly. To combat against such model mis-specification, we propose to wed the GP with the prevailing conformal prediction (CP), a distribution-free post-processing framework that produces it prediction sets with a provably valid coverage under the sole assumption of data exchangeability. However, this assumption is usually violated in the online setting, where a prediction set is sought before revealing the true label. To ensure long-term coverage guarantee, we will adaptively set the key threshold parameter based on the feedback whether the true label falls inside the prediction set. Numerical results demonstrate the merits of the online GP-CP approach relative to existing alternatives in the long-term coverage performance.