Small Generalizable Prompt Predictive Models Can Steer Efficient RL Post-Training of Large Reasoning Models

Reinforcement learning enhances the reasoning capabilities of large language models but often involves high computational costs due to rollout-intensive optimization. Online prompt selection presents a plausible solution by prioritizing informative prompts to improve training efficiency. However, current methods either depend on costly, exact evaluations or construct prompt-specific predictive models lacking generalization across prompts. This study introduces Generalizable Predictive Prompt Selection (GPS), which performs Bayesian inference towards prompt difficulty using a lightweight generative model trained on the shared optimization history. Intermediate-difficulty prioritization and history-anchored diversity are incorporated into the batch acquisition principle to select informative prompt batches. The small predictive model also generalizes at test-time for efficient computational allocation. Experiments across varied reasoning benchmarks indicate GPS's substantial improvements in training efficiency, final performance, and test-time efficiency over superior baseline methods.

Kimi K2.5: Visual Agentic Intelligence

We introduce Kimi K2.5, an open-source multimodal agentic model designed to advance general agentic intelligence. K2.5 emphasizes the joint optimization of text and vision so that two modalities enhance each other. This includes a series of techniques such as joint text-vision pre-training, zero-vision SFT, and joint text-vision reinforcement learning. Building on this multimodal foundation, K2.5 introduces Agent Swarm, a self-directed parallel agent orchestration framework that dynamically decomposes complex tasks into heterogeneous sub-problems and executes them concurrently. Extensive evaluations show that Kimi K2.5 achieves state-of-the-art results across various domains including coding, vision, reasoning, and agentic tasks. Agent Swarm also reduces latency by up to $4.5\times$ over single-agent baselines. We release the post-trained Kimi K2.5 model checkpoint to facilitate future research and real-world applications of agentic intelligence.

MoST: Mixing Speech and Text with Modality-Aware Mixture of Experts

We present MoST (Mixture of Speech and Text), a novel multimodal large language model that seamlessly integrates speech and text processing through our proposed Modality-Aware Mixture of Experts (MAMoE) architecture. While current multimodal models typically process diverse modality representations with identical parameters, disregarding their inherent representational differences, we introduce specialized routing pathways that direct tokens to modality-appropriate experts based on input type. MAMoE simultaneously enhances modality-specific learning and cross-modal understanding through two complementary components: modality-specific expert groups that capture domain-specific patterns and shared experts that facilitate information transfer between modalities. Building on this architecture, we develop an efficient transformation pipeline that adapts the pretrained MoE language model through strategic post-training on ASR and TTS datasets, followed by fine-tuning with a carefully curated speech-text instruction dataset. A key feature of this pipeline is that it relies exclusively on fully accessible, open-source datasets to achieve strong performance and data efficiency. Comprehensive evaluations across ASR, TTS, audio language modeling, and spoken question answering benchmarks show that MoST consistently outperforms existing models of comparable parameter counts. Our ablation studies confirm that the modality-specific routing mechanism and shared experts design significantly contribute to performance gains across all tested domains. To our knowledge, MoST represents the first fully open-source speech-text LLM built on a Mixture of Experts architecture. \footnote{We release MoST model, training code, inference code, and training data at https://github.com/NUS-HPC-AI-Lab/MoST

KDCM: Reducing Hallucination in LLMs through Explicit Reasoning Structures

To mitigate hallucinations in large language models (LLMs), we propose a framework that focuses on errors induced by prompts. Our method extends a chain-style knowledge distillation approach by incorporating a programmable module that guides knowledge graph exploration. This module is embedded as executable code within the reasoning prompt, allowing the model to leverage external structured knowledge during inference. Based on this design, we develop an enhanced distillation-based reasoning framework that explicitly regulates intermediate reasoning steps, resulting in more reliable predictions. We evaluate the proposed approach on multiple public benchmarks using GPT-4 and LLaMA-3.3. Experimental results show that code-guided reasoning significantly improves contextual modeling and reduces prompt-induced hallucinations. Specifically, HIT@1, HIT@3, and HIT@5 increase by 15.64%, 13.38%, and 13.28%, respectively, with scores exceeding 95% across several evaluation settings. These findings indicate that the proposed method effectively constrains erroneous reasoning while improving both accuracy and interpretability.

Mitigating Prompt-Induced Hallucinations in Large Language Models via Structured Reasoning

To address hallucination issues in large language models (LLMs), this paper proposes a method for mitigating prompt-induced hallucinations. Building on a knowledge distillation chain-style model, we introduce a code module to guide knowledge-graph exploration and incorporate code as part of the chain-of-thought prompt, forming an external knowledge input that provides more accurate and structured information to the model. Based on this design, we develop an improved knowledge distillation chain-style model and leverage it to analyze and constrain the reasoning process of LLMs, thereby improving inference accuracy. We empirically evaluate the proposed approach using GPT-4 and LLaMA-3.3 on multiple public datasets. Experimental results demonstrate that incorporating code modules significantly enhances the model's ability to capture contextual information and effectively mitigates prompt-induced hallucinations. Specifically, HIT@1, HIT@3, and HIT@5 improve by 15.64%, 13.38%, and 13.28%, respectively. Moreover, the proposed method achieves HIT@1, HIT@3, and HIT@5 scores exceeding 95% across several evaluation settings. These results indicate that the proposed approach substantially reduces hallucination behavior while improving the accuracy and verifiability of large language models.

EntroPIC: Towards Stable Long-Term Training of LLMs via Entropy Stabilization with Proportional-Integral Control

Long-term training of large language models (LLMs) requires maintaining stable exploration to prevent the model from collapsing into sub-optimal behaviors. Entropy is crucial in this context, as it controls exploration and helps avoid premature convergence to sub-optimal solutions. However, existing reinforcement learning methods struggle to maintain an appropriate level of entropy, as the training process involves a mix of positive and negative samples, each affecting entropy in different ways across steps. To address this, we propose Entropy stablilization via Proportional-Integral Control (EntroPIC), a novel method that adaptively adjusts the influence of positive and negative samples by dynamically tuning their loss coefficients. This approach stabilizes entropy throughout training, ensuring efficient exploration and steady progress. We provide a comprehensive theoretical analysis for both on-policy and off-policy learning settings, demonstrating that EntroPIC is effective at controlling entropy in large-scale LLM training. Experimental results show that our method successfully maintains desired entropy levels, enabling stable and optimal RL training for LLMs.

RoS-Guard: Robust and Scalable Online Change Detection with Delay-Optimal Guarantees

Online change detection (OCD) aims to rapidly identify change points in streaming data and is critical in applications such as power system monitoring, wireless network sensing, and financial anomaly detection. Existing OCD methods typically assume precise system knowledge, which is unrealistic due to estimation errors and environmental variations. Moreover, existing OCD methods often struggle with efficiency in large-scale systems. To overcome these challenges, we propose RoS-Guard, a robust and optimal OCD algorithm tailored for linear systems with uncertainty. Through a tight relaxation and reformulation of the OCD optimization problem, RoS-Guard employs neural unrolling to enable efficient parallel computation via GPU acceleration. The algorithm provides theoretical guarantees on performance, including expected false alarm rate and worst-case average detection delay. Extensive experiments validate the effectiveness of RoS-Guard and demonstrate significant computational speedup in large-scale system scenarios.

Contributions to Robust and Efficient Methods for Analysis of High Dimensional Data

A ubiquitous feature of data of our era is their extra-large sizes and dimensions. Analyzing such high-dimensional data poses significant challenges, since the feature dimension is often much larger than the sample size. This thesis introduces robust and computationally efficient methods to address several common challenges associated with high-dimensional data. In my first manuscript, I propose a coherent approach to variable screening that accommodates nonlinear associations. I develop a novel variable screening method that transcends traditional linear assumptions by leveraging mutual information, with an intended application in neuroimaging data. This approach allows for accurate identification of important variables by capturing nonlinear as well as linear relationships between the outcome and covariates. Building on this foundation, I develop new optimization methods for sparse estimation using nonconvex penalties in my second manuscript. These methods address notable challenges in current statistical computing practices, facilitating computationally efficient and robust analyses of complex datasets. The proposed method can be applied to a general class of optimization problems. In my third manuscript, I contribute to robust modeling of high-dimensional correlated observations by developing a mixed-effects model based on Tsallis power-law entropy maximization and discussed the theoretical properties of such distribution. This model surpasses the constraints of conventional Gaussian models by accommodating a broader class of distributions with enhanced robustness to outliers. Additionally, I develop a proximal nonlinear conjugate gradient algorithm that accelerates convergence while maintaining numerical stability, along with rigorous statistical properties for the proposed framework.

Graph Federated Learning for Personalized Privacy Recommendation

Federated recommendation systems (FedRecs) have gained significant attention for providing privacy-preserving recommendation services. However, existing FedRecs assume that all users have the same requirements for privacy protection, i.e., they do not upload any data to the server. The approaches overlook the potential to enhance the recommendation service by utilizing publicly available user data. In real-world applications, users can choose to be private or public. Private users' interaction data is not shared, while public users' interaction data can be shared. Inspired by the issue, this paper proposes a novel Graph Federated Learning for Personalized Privacy Recommendation (GFed-PP) that adapts to different privacy requirements while improving recommendation performance. GFed-PP incorporates the interaction data of public users to build a user-item interaction graph, which is then used to form a user relationship graph. A lightweight graph convolutional network (GCN) is employed to learn each user's user-specific personalized item embedding. To protect user privacy, each client learns the user embedding and the scoring function locally. Additionally, GFed-PP achieves optimization of the federated recommendation framework through the initialization of item embedding on clients and the aggregation of the user relationship graph on the server. Experimental results demonstrate that GFed-PP significantly outperforms existing methods for five datasets, offering superior recommendation accuracy without compromising privacy. This framework provides a practical solution for accommodating varying privacy preferences in federated recommendation systems.

Stable Preference Optimization for LLMs: A Bilevel Approach Beyond Direct Preference Optimization

Direct Preference Optimization (DPO) has emerged as a popular and efficient alternative to reward modeling and reinforcement learning for aligning language models with human preferences. Despite its empirical success, the theoretical properties and intrinsic limitations of DPO remain underexplored. In this work, we first present a comprehensive analysis of DPO's dynamics from a probability evolution perspective. Our analysis reveals that DPO is highly sensitive to initialization. It also tends to misallocate probability mass, which can inadvertently shift probability toward irrelevant or undesired responses. This misallocation may unintentionally reinforce model bias, thereby compromising both the stability of model alignment and the consistency with intended preferences. Motivated by these theoretical findings, we propose a theoretically grounded bilevel optimization framework that tightly integrate supervised fine-tuning with an enhanced DPO objective a.k.a. stable preference optimization. Our approach introduces a principled regularization scheme to explicitly encourage absolute probability improvement for preferred outputs, while maintaining stable optimization dynamics. Experiments on challenging reasoning and summarization benchmarks elucidate that our method consistently improves reasoning accuracy and better aligns output distributions with intended preferences, outperforming standard DPO. Stable preference optimization provides new insights into the design of preference-based alignment objectives and opens up new avenues towards more reliable and interpretable language model alignment.