Direct Preference Optimization (DPO) is a powerful paradigm to align language models with human preferences using pairwise comparisons. However, its binary win-or-loss supervision often proves insufficient for training small models with limited capacity. Prior works attempt to distill information from large teacher models using behavior cloning or KL divergence. These methods often focus on mimicking current behavior and overlook distilling reward modeling. To address this issue, we propose \textit{Teacher Value-based Knowledge Distillation} (TVKD), which introduces an auxiliary reward from the value function of the teacher model to provide a soft guide. This auxiliary reward is formulated to satisfy potential-based reward shaping, ensuring that the global reward structure and optimal policy of DPO are preserved. TVKD can be integrated into the standard DPO training framework and does not require additional rollouts. Our experimental results show that TVKD consistently improves performance across various benchmarks and model sizes.

Federated learning (FL) enables distributed devices to collaboratively train machine learning models while maintaining data privacy. However, the heterogeneous hardware capabilities of devices often result in significant training delays, as straggler clients with limited resources prolong the aggregation process. Existing solutions such as client selection, asynchronous FL, and partial training partially address these challenges but encounter issues such as reduced accuracy, stale updates, and compromised model performance due to inconsistent training contributions. To overcome these limitations, we propose FedEL, a federated elastic learning framework that enhances training efficiency while maintaining model accuracy. FedEL introduces a novel window-based training process, sliding the window to locate the training part of the model and dynamically selecting important tensors for training within a coordinated runtime budget. This approach ensures progressive and balanced training across all clients, including stragglers. Additionally, FedEL employs a tensor importance adjustment module, harmonizing local and global tensor importance to mitigate biases caused by data heterogeneity. The experiment results show that FedEL achieves up to 3.87Πimprovement in time-to-accuracy compared to baselines while maintaining or exceeding final test accuracy.

Contrastive learning is a powerful technique for discovering meaningful data representations by optimizing objectives based on $\textit{contrastive information}$, often given as a set of weighted triplets $\{(x_i, y_i^+, z_{i}^-)\}_{i = 1}^m$ indicating that an "anchor" $x_i$ is more similar to a "positive" example $y_i$ than to a "negative" example $z_i$. The goal is to find representations (e.g., embeddings in $\mathbb{R}^d$ or a tree metric) where anchors are placed closer to positive than to negative examples. While finding $\textit{global}$ optima of contrastive objectives is $\mathsf{NP}$-hard, the complexity of finding $\textit{local}$ optima -- representations that do not improve by local search algorithms such as gradient-based methods -- remains open. Our work settles the complexity of finding local optima in various contrastive learning problems by proving $\mathsf{PLS}$-hardness in discrete settings (e.g., maximize satisfied triplets) and $\mathsf{CLS}$-hardness in continuous settings (e.g., minimize Triplet Loss), where $\mathsf{PLS}$ (Polynomial Local Search) and $\mathsf{CLS}$ (Continuous Local Search) are well-studied complexity classes capturing local search dynamics in discrete and continuous optimization, respectively. Our results imply that no polynomial time algorithm (local search or otherwise) can find a local optimum for various contrastive learning problems, unless $\mathsf{PLS}\subseteq\mathsf{P}$ (or $\mathsf{CLS}\subseteq \mathsf{P}$ for continuous problems). Even in the unlikely scenario that $\mathsf{PLS}\subseteq\mathsf{P}$ (or $\mathsf{CLS}\subseteq \mathsf{P}$), our reductions imply that there exist instances where local search algorithms need exponential time to reach a local optimum, even for $d=1$ (embeddings on a line).

F1Tenth is a widely adopted reduced-scale platform for developing and testing autonomous racing algorithms, hosting annual competitions worldwide. With high operating speeds, dynamic environments, and head-to-head interactions, autonomous racing requires algorithms that diverge from those in classical autonomous driving. Training such algorithms is particularly challenging: the need for rapid decision-making at high speeds severely limits model capacity. To address this, we propose End2Race, a novel end-to-end imitation learning algorithm designed for head-to-head autonomous racing. End2Race leverages a Gated Recurrent Unit (GRU) architecture to capture continuous temporal dependencies, enabling both short-term responsiveness and long-term strategic planning. We also adopt a sigmoid-based normalization function that transforms raw LiDAR scans into spatial pressure tokens, facilitating effective model training and convergence. The algorithm is extremely efficient, achieving an inference time of less than 0.5 milliseconds on a consumer-class GPU. Experiments in the F1Tenth simulator demonstrate that End2Race achieves a 94.2% safety rate across 2,400 overtaking scenarios, each with an 8-second time limit, and successfully completes overtakes in 59.2% of cases. This surpasses previous methods and establishes ours as a leading solution for the F1Tenth racing testbed. Code is available at https://github.com/michigan-traffic-lab/End2Race.

Mixture-of-Experts (MoE) models offer immense capacity via sparsely gated expert subnetworks, yet adapting them to multiple domains without catastrophic forgetting remains an open challenge. Existing approaches either incur prohibitive computation, suffer cross-domain interference, or require separate runs per domain. We propose DES-MoE, a dynamic expert specialization framework for multi-domain adaptation of Mixture-of-Experts models. DES-MoE addresses catastrophic forgetting through three innovations: (1) an adaptive router balancing pre-trained knowledge retention and task-specific updates via distillation, (2) real-time expert-domain correlation mapping to isolate domain-specific gradients, and (3) a three-phase adaptive fine-tuning schedule that progressively freezes non-specialized parameters. Evaluated on six domains (math, code, law, etc.), DES-MoE matches single-domain ESFT performance while training one unified model, reduces forgetting by 89% compared to full fine-tuning as domains scale from 2 to 6, and achieves 68% faster convergence than conventional methods. Our work establishes dynamic expert isolation as a scalable paradigm for multi-task MoE adaptation.

Most existing efforts on APA typically adopt segmental-level features as inputs and predict pronunciation scores at different granularities via hierarchical (or parallel) pronunciation modeling. This, however, inevitably causes assessments across linguistic levels (e.g., phone, word, and utterance) to rely solely on phoneme-level pronunciation features, nearly sidelining supra-segmental pronunciation cues. T o address this limitation, we introduce multi-task pre-training (MTP) for APA, a simple yet effective strategy that attempts to capture long-term temporal pronunciation cues while strengthening the intrinsic structures within an utterance via the objective of reconstructing input features. Specifically, for a phoneme-level encoder of an APA model, the proposed MTP strategy randomly masks segmental-level pronunciation features and reconstructs the masked ones based on their surrounding pronunciation context. Furthermore, current APA systems lack integration with automated speaking assessment (ASA), limiting holistic proficiency evaluation. Drawing on empirical studies and prior knowledge in ASA, our framework bridges this gap by incorporating handcrafted features (HCFs), such as fluency (speech rate, silence duration) and stress (pitch accent strength), derived from human-designed formulas via regressors to generate interpretable proficiency scores. Experiments on speechocean762 show improved pronunciation scoring and ASA proficiency correlation, enabling targeted training and comprehensive proficiency assessment. Index T erms--computer-assisted language learning, automatic pronunciation assessment, automated speaking assessment, multi-task learning.

Virtual brainstorming sessions have become a central component of collaborative problem solving, yet the large volume and uneven distribution of ideas often make it difficult to extract valuable insights efficiently. Manual coding of ideas is time-consuming and subjective, underscoring the need for automated approaches to support the evaluation of group creativity. In this study, we propose a semantic-driven topic modeling framework that integrates four modular components: transformer-based embeddings (Sentence-BERT), dimensionality reduction (UMAP), clustering (HDBSCAN), and topic extraction with refinement. We evaluate our approach on structured Zoom brainstorming sessions involving student groups tasked with improving their university. Results demonstrate that our model achieves higher topic coherence compared to established methods such as LDA, ETM, and BERTopic, with an average coherence score of 0.687 (CV), outperforming baselines by a significant margin. Beyond improved performance, the model provides interpretable insights into the depth and diversity of topics explored, supporting both convergent and divergent dimensions of group creativity. This work highlights the potential of embedding-based topic modeling for analyzing collaborative ideation and contributes an efficient and scalable framework for studying creativity in synchronous virtual meetings. Introduction Digital communication has become central to modern collaboration, with virtual meetings and conversational agents such as chatbots increasingly shaping how teams interact across geographical and cultural boundaries [1].

Consistent high-quality nursing care is essential for patient safety, yet current nursing education depends on subjective, time-intensive instructor feedback in training future nurses, which limits scalability and efficiency in their training, and thus hampers nursing competency when they enter the workforce. In this paper, we introduce a video-language model (VLM) based framework to develop the AI capability of automated procedural assessment and feedback for nursing skills training, with the potential of being integrated into existing training programs. Mimicking human skill acquisition, the framework follows a curriculum-inspired progression, advancing from high-level action recognition, fine-grained subaction decomposition, and ultimately to procedural reasoning. This design supports scalable evaluation by reducing instructor workload while preserving assessment quality. The system provides three core capabilities: 1) diagnosing errors by identifying missing or incorrect subactions in nursing skill instruction videos, 2) generating explainable feedback by clarifying why a step is out of order or omitted, and 3) enabling objective, consistent formative evaluation of procedures. Validation on synthesized videos demonstrates reliable error detection and temporal localization, confirming its potential to handle real-world training variability. By addressing workflow bottlenecks and supporting large-scale, standardized evaluation, this work advances AI applications in nursing education, contributing to stronger workforce development and ultimately safer patient care.

In recent years, training methods centered on Reinforcement Learning (RL) have markedly enhanced the reasoning and alignment performance of Large Language Models (LLMs), particularly in understanding human intents, following user instructions, and bolstering inferential strength. Although existing surveys offer overviews of RL augmented LLMs, their scope is often limited, failing to provide a comprehensive summary of how RL operates across the full lifecycle of LLMs. We systematically review the theoretical and practical advancements whereby RL empowers LLMs, especially Reinforcement Learning with Verifiable Rewards (RLVR). First, we briefly introduce the basic theory of RL. Second, we thoroughly detail application strategies for RL across various phases of the LLM lifecycle, including pre-training, alignment fine-tuning, and reinforced reasoning. In particular, we emphasize that RL methods in the reinforced reasoning phase serve as a pivotal driving force for advancing model reasoning to its limits. Next, we collate existing datasets and evaluation benchmarks currently used for RL fine-tuning, spanning human-annotated datasets, AI-assisted preference data, and program-verification-style corpora. Subsequently, we review the mainstream open-source tools and training frameworks available, providing clear practical references for subsequent research. Finally, we analyse the future challenges and trends in the field of RL-enhanced LLMs. This survey aims to present researchers and practitioners with the latest developments and frontier trends at the intersection of RL and LLMs, with the goal of fostering the evolution of LLMs that are more intelligent, generalizable, and secure.

We deploy VMS on the Unitree G1 humanoid robot, demonstrating its capability to perform a broad category of motion skills with strong stability and generalization. The repertoire includes (a) walking and running, (b) ball throwing and racket swinging, (c) dancing, (d) diverse kicking, (e) Kung Fu and (f) long sequences of martial arts and dance. Abstract-- Learning versatile whole-body skills by tracking various human motions is a fundamental step toward general-purpose humanoid robots. This task is particularly challenging because a single policy must master a broad repertoire of motion skills while ensuring stability over long-horizon sequences. T o this end, we present VMS, a unified whole-body controller that enables humanoid robots to learn diverse and dynamic behaviors within a single policy. Our framework integrates a hybrid tracking objective that balances local motion fidelity with global trajectory consistency, and an Orthogonal Mixture-of-Experts (OMoE) architecture that encourages skill specialization while enhancing generalization across motions. A segment-level tracking reward is further introduced to relax rigid step-wise matching, enhancing robustness when handling global displacements and transient inaccuracies.