Contextual bandits are a core technology for personalized mobile health interventions, where decision-making requires adapting to complex, non-linear user behaviors. While Thompson Sampling (TS) is a preferred strategy for these problems, its performance hinges on the quality of the underlying reward model. Standard linear models suffer from high bias, while neural network approaches are often brittle and difficult to tune in online settings. Conversely, tree ensembles dominate tabular data prediction but typically rely on heuristic uncertainty quantification, lacking a principled probabilistic basis for TS. We propose Bayesian Forest Thompson Sampling (BFTS), the first contextual bandit algorithm to integrate Bayesian Additive Regression Trees (BART), a fully probabilistic sum-of-trees model, directly into the exploration loop. We prove that BFTS is theoretically sound, deriving an information-theoretic Bayesian regret bound of $\tilde{O}(\sqrt{T})$. As a complementary result, we establish frequentist minimax optimality for a "feel-good" variant, confirming the structural suitability of BART priors for non-parametric bandits. Empirically, BFTS achieves state-of-the-art regret on tabular benchmarks with near-nominal uncertainty calibration. Furthermore, in an offline policy evaluation on the Drink Less micro-randomized trial, BFTS improves engagement rates by over 30% compared to the deployed policy, demonstrating its practical effectiveness for behavioral interventions.

AStackelberg game[30, 31] is a strategic interaction between two utility-maximizing players in which one player (theleader) is able to commit to a (possibly mixed) strategy before the other player (thefollower)takesanaction.

Effective post-training is essential to align Large Language Models (LLMs) with specialized biomedical knowledge to accelerate life science research. However, current approaches face significant limitations. First, biomedical reasoning involves intricate mechanisms often represented by sparse textual data. Standard Supervised Fine-Tuning (SFT) tends to overfit to surface-level instruction patterns without effectively internalizing this fragmented scientific knowledge. Second, Reinforcement Learning (RL) is impractical for this domain, as defining meaningful rewards often necessitates prohibitive experimental validation (e.g., wet-lab verification of drug responses), rendering real-time feedback unfeasible. We propose Balanced Fine-Tuning (BFT), an efficient post-training method designed to learn complex reasoning from sparse data without external reward signals. BFT operates through a two-layer weighting mechanism: 1. At the token level, it scales loss via prediction probabilities to stabilize gradients and prevent overfitting; 2. At the sample level, it uses "minimum group confidence" to adaptively enhance the learning of hard samples. Experiments demonstrate that BFT significantly outperforms SFT. In medical tasks, it enables LLMs to acquire knowledge that SFT misses. In biological tasks, BFT-based LLMs surpass GeneAgent (an accurate agent for biology analysis) in biological process reasoning. Moreover, the text embeddings generated by BFT can be directly applied to downstream tasks, such as gene interaction and single-cell perturbation response prediction. These results indicate that BFT facilitates broad applications of LLMs in biomedical research.

Text-to-image diffusion models can emit copyrighted, unsafe, or private content. Safety alignment aims to suppress specific concepts, yet evaluations seldom test whether safety persists under benign downstream fine-tuning routinely applied after deployment (e.g., LoRA personalization, style/domain adapters). We study the stability of current safety methods under benign fine-tuning and observe frequent breakdowns. As true safety alignment must withstand even benign post-deployment adaptations, we introduce the SPQR benchmark (Safety-Prompt adherence-Quality-Robustness). SPQR is a single-scored metric that provides a standardized and reproducible framework to evaluate how well safety-aligned diffusion models preserve safety, utility, and robustness under benign fine-tuning, by reporting a single leaderboard score to facilitate comparisons. We conduct multilingual, domain-specific, and out-of-distribution analyses, along with category-wise breakdowns, to identify when safety alignment fails after benign fine-tuning, ultimately showcasing SPQR as a concise yet comprehensive benchmark for T2I safety alignment techniques for T2I models.

Fidelity quantum kernels have shown promise in classification tasks, particularly when a group structure in the data can be identified and exploited through a covariant feature map. In fact, there exist classification problems on which covariant kernels provide a provable advantage, thus establishing a separation between quantum and classical learners. However, their practical application poses two challenges: on one side, the group structure may be unknown and approximate in real-world data, and on the other side, scaling to the `utility' regime (above 100 qubits) is affected by exponential concentration. In this work, we address said challenges by applying fidelity kernels to real-world data with unknown structure, related to the scheduling of a fleet of electric vehicles, and to synthetic data generated from the union of subspaces, which is then close to many relevant real-world datasets. Furthermore, we propose a novel error mitigation strategy specifically tailored for fidelity kernels, called Bit Flip Tolerance (BFT), to alleviate the exponential concentration in our utility-scale experiments. Our multiclass classification reaches accuracies comparable to classical SVCs up to 156 qubits, thus constituting the largest experimental demonstration of quantum machine learning on IBM devices to date. For the real-world data experiments, the effect of the proposed BFT becomes manifest on 40+ qubits, where mitigated accuracies reach 80%, in line with classical, compared to 33% without BFT. Through the union-of-subspace synthetic dataset with 156 qubits, we demonstrate a mitigated accuracy of 80%, compared to 83% of classical models, and 37% of unmitigated quantum, using a test set of limited size.