Merge-of-Thought Distillation

Efficient reasoning distillation for long chain-of-thought (CoT) models is increasingly constrained by the assumption of a single oracle teacher, despite practical availability of multiple candidate teachers and growing CoT corpora. We revisit teacher selection and observe that different students have different "best teachers," and even for the same student the best teacher can vary across datasets. Therefore, to unify multiple teachers' reasoning abilities into student with overcoming conflicts among various teachers' supervision, we propose Merge-of-Thought Distillation (MoT), a lightweight framework that alternates between teacher-specific supervised fine-tuning branches and weight-space merging of the resulting student variants. On competition math benchmarks, using only about 200 high-quality CoT samples, applying MoT to a Qwen3-14B student surpasses strong models including DEEPSEEK-R1, QWEN3-30B-A3B, QWEN3-32B, and OPENAI-O1, demonstrating substantial gains. Besides, MoT consistently outperforms the best single-teacher distillation and the naive multi-teacher union, raises the performance ceiling while mitigating overfitting, and shows robustness to distribution-shifted and peer-level teachers. Moreover, MoT reduces catastrophic forgetting, improves general reasoning beyond mathematics and even cultivates a better teacher, indicating that consensus-filtered reasoning features transfer broadly. These results position MoT as a simple, scalable route to efficiently distilling long CoT capabilities from diverse teachers into compact students.

Multimodal Contrastive Pretraining of CBCT and IOS for Enhanced Tooth Segmentation

Digital dentistry represents a transformative shift in modern dental practice. The foundational step in this transformation is the accurate digital representation of the patient's dentition, which is obtained from segmented Cone-Beam Computed Tomography (CBCT) and Intraoral Scans (IOS). Despite the growing interest in digital dental technologies, existing segmentation methodologies frequently lack rigorous validation and demonstrate limited performance and clinical applicability. To the best of our knowledge, this is the first work to introduce a multimodal pretraining framework for tooth segmentation. We present ToothMCL, a Tooth Multimodal Contrastive Learning for pretraining that integrates volumetric (CBCT) and surface-based (IOS) modalities. By capturing modality-invariant representations through multimodal contrastive learning, our approach effectively models fine-grained anatomical features, enabling precise multi-class segmentation and accurate identification of F\'ed\'eration Dentaire Internationale (FDI) tooth numbering. Along with the framework, we curated CBCT-IOS3.8K, the largest paired CBCT and IOS dataset to date, comprising 3,867 patients. We then evaluated ToothMCL on a comprehensive collection of independent datasets, representing the largest and most diverse evaluation to date. Our method achieves state-of-the-art performance in both internal and external testing, with an increase of 12\% for CBCT segmentation and 8\% for IOS segmentation in the Dice Similarity Coefficient (DSC). Furthermore, ToothMCL consistently surpasses existing approaches in tooth groups and demonstrates robust generalizability across varying imaging conditions and clinical scenarios.

Hybrid Matrix Factorization Based Graph Contrastive Learning for Recommendation System

In recent years, methods that combine contrastive learning with graph neural networks have emerged to address the challenges of recommendation systems, demonstrating powerful performance and playing a significant role in this domain. Contrastive learning primarily tackles the issue of data sparsity by employing data augmentation strategies, effectively alleviating this problem and showing promising results. Although existing research has achieved favorable outcomes, most current graph contrastive learning methods are based on two types of data augmentation strategies: the first involves perturbing the graph structure, such as by randomly adding or removing edges; and the second applies clustering techniques. We believe that the interactive information obtained through these two strategies does not fully capture the user-item interactions. In this paper, we propose a novel method called HMFGCL (Hybrid Matrix Factorization Based Graph Contrastive Learning), which integrates two distinct matrix factorization techniques-low-rank matrix factorization (MF) and singular value decomposition (SVD)-to complementarily acquire global collaborative information, thereby constructing enhanced views. Experimental results on multiple public datasets demonstrate that our model outperforms existing baselines, particularly on small-scale datasets.

TalkToAgent: A Human-centric Explanation of Reinforcement Learning Agents with Large Language Models

Explainable Reinforcement Learning (XRL) has emerged as a promising approach in improving the transparency of Reinforcement Learning (RL) agents. However, there remains a gap between complex RL policies and domain experts, due to the limited comprehensibility of XRL results and isolated coverage of current XRL approaches that leave users uncertain about which tools to employ. To address these challenges, we introduce TalkToAgent, a multi-agent Large Language Models (LLM) framework that delivers interactive, natural language explanations for RL policies. The architecture with five specialized LLM agents (Coordinator, Explainer, Coder, Evaluator, and Debugger) enables TalkToAgent to automatically map user queries to relevant XRL tools and clarify an agent's actions in terms of either key state variables, expected outcomes, or counterfactual explanations. Moreover, our approach extends previous counterfactual explanations by deriving alternative scenarios from qualitative behavioral descriptions, or even new rule-based policies. We validated TalkToAgent on quadruple-tank process control problem, a well-known nonlinear control benchmark. Results demonstrated that TalkToAgent successfully mapped user queries into XRL tasks with high accuracy, and coder-debugger interactions minimized failures in counterfactual generation. Furthermore, qualitative evaluation confirmed that TalkToAgent effectively interpreted agent's actions and contextualized their meaning within the problem domain.

Beyond Correctness: Harmonizing Process and Outcome Rewards through RL Training

Reinforcement learning with verifiable rewards (RLVR) has emerged to be a predominant paradigm for mathematical reasoning tasks, offering stable improvements in reasoning ability. However, Outcome Reward Models (ORMs) in RLVR are too coarse-grained to distinguish flawed reasoning within correct answers or valid reasoning within incorrect answers. This lack of granularity introduces noisy and misleading gradients significantly and hinders further progress in reasoning process quality. While Process Reward Models (PRMs) offer fine-grained guidance for intermediate steps, they frequently suffer from inaccuracies and are susceptible to reward hacking. To resolve this dilemma, we introduce PRocess cOnsistency Filter (PROF), an effective data process curation method that harmonizes noisy, fine-grained process rewards with accurate, coarse-grained outcome rewards. Rather than naively blending PRM and ORM in the objective function (arXiv:archive/2506.18896), PROF leverages their complementary strengths through consistency-driven sample selection. Our approach retains correct responses with higher averaged process values and incorrect responses with lower averaged process values, while maintaining positive/negative training sample balance. Extensive experiments demonstrate that our method not only consistently improves the final accuracy over $4\%$ compared to the blending approaches, but also strengthens the quality of intermediate reasoning steps. Codes and training recipes are available at https://github.com/Chenluye99/PROF.

Towards a Real-World Aligned Benchmark for Unlearning in Recommender Systems

Modern recommender systems heavily leverage user interaction data to deliver personalized experiences. However, relying on personal data presents challenges in adhering to privacy regulations, such as the GDPR's "right to be forgotten". Machine unlearning (MU) aims to address these challenges by enabling the efficient removal of specific training data from models post-training, without compromising model utility or leaving residual information. However, current benchmarks for unlearning in recommender systems -- most notably CURE4Rec -- fail to reflect real-world operational demands. They focus narrowly on collaborative filtering, overlook tasks like session-based and next-basket recommendation, simulate unrealistically large unlearning requests, and ignore critical efficiency constraints. In this paper, we propose a set of design desiderata and research questions to guide the development of a more realistic benchmark for unlearning in recommender systems, with the goal of gathering feedback from the research community. Our benchmark proposal spans multiple recommendation tasks, includes domain-specific unlearning scenarios, and several unlearning algorithms -- including ones adapted from a recent NeurIPS unlearning competition. Furthermore, we argue for an unlearning setup that reflects the sequential, time-sensitive nature of real-world deletion requests. We also present a preliminary experiment in a next-basket recommendation setting based on our proposed desiderata and find that unlearning also works for sequential recommendation models, exposed to many small unlearning requests. In this case, we observe that a modification of a custom-designed unlearning algorithm for recommender systems outperforms general unlearning algorithms significantly, and that unlearning can be executed with a latency of only several seconds.

Tinker: Diffusion's Gift to 3D--Multi-View Consistent Editing From Sparse Inputs without Per-Scene Optimization

We introduce Tinker, a versatile framework for high-fidelity 3D editing that operates in both one-shot and few-shot regimes without any per-scene finetuning. Unlike prior techniques that demand extensive per-scene optimization to ensure multi-view consistency or to produce dozens of consistent edited input views, Tinker delivers robust, multi-view consistent edits from as few as one or two images. This capability stems from repurposing pretrained diffusion models, which unlocks their latent 3D awareness. To drive research in this space, we curate the first large-scale multi-view editing dataset and data pipeline, spanning diverse scenes and styles. Building on this dataset, we develop our framework capable of generating multi-view consistent edited views without per-scene training, which consists of two novel components: (1) Referring multi-view editor: Enables precise, reference-driven edits that remain coherent across all viewpoints. (2) Any-view-to-video synthesizer: Leverages spatial-temporal priors from video diffusion to perform high-quality scene completion and novel-view generation even from sparse inputs. Through extensive experiments, Tinker significantly reduces the barrier to generalizable 3D content creation, achieving state-of-the-art performance on editing, novel-view synthesis, and rendering enhancement tasks. We believe that Tinker represents a key step towards truly scalable, zero-shot 3D editing. Project webpage: https://aim-uofa.github.io/Tinker

Time Is a Feature: Exploiting Temporal Dynamics in Diffusion Language Models

Diffusion large language models (dLLMs) generate text through iterative denoising, yet current decoding strategies discard rich intermediate predictions in favor of the final output. Our work here reveals a critical phenomenon, temporal oscillation, where correct answers often emerge in the middle process, but are overwritten in later denoising steps. To address this issue, we introduce two complementary methods that exploit temporal consistency: 1) Temporal Self-Consistency Voting, a training-free, test-time decoding strategy that aggregates predictions across denoising steps to select the most consistent output; and 2) a post-training method termed Temporal Consistency Reinforcement, which uses Temporal Semantic Entropy (TSE), a measure of semantic stability across intermediate predictions, as a reward signal to encourage stable generations. Empirical results across multiple benchmarks demonstrate the effectiveness of our approach. Using the negative TSE reward alone, we observe a remarkable average improvement of 24.7% on the Countdown dataset over an existing dLLM. Combined with the accuracy reward, we achieve absolute gains of 2.0% on GSM8K, 4.3% on MATH500, 6.6% on SVAMP, and 25.3% on Countdown, respectively. Our findings underscore the untapped potential of temporal dynamics in dLLMs and offer two simple yet effective tools to harness them.

VQ-VAE Based Digital Semantic Communication with Importance-Aware OFDM Transmission

Semantic communication (SemCom) significantly reduces redundant data and improves transmission efficiency by extracting the latent features of information. However, most of the conventional deep learning-based SemCom systems focus on analog transmission and lack in compatibility with practical digital communications. This paper proposes a vector quantized-variational autoencoder (VQ-VAE) based digital SemCom system that directly transmits the semantic features and incorporates the importance-aware orthogonal frequency division multiplexing (OFDM) transmission to enhance the SemCom performance, where the VQ-VAE generates a discrete codebook shared between the transmitter and receiver. At transmitter, the latent semantic features are firstly extracted by VQ-VAE, and then the shared codebook is adopted to match these features, which are subsequently transformed into a discrete version to adapt the digital transmission. To protect the semantic information, an importance-aware OFDM transmission strategy is proposed to allocate the key features near the OFDM reference signals, where the feature importance is derived from the gradient-based method. At the receiver, the features are rematched with the shared codebook to further correct errors. Finally, experimental results demonstrate that our proposed scheme outperforms the conventional DeepSC and achieves better reconstruction performance under low SNR region.

Learned Regularization for Microwave Tomography

Microwave Tomography (MWT) aims to reconstruct the dielectric properties of tissues from measured scattered electromagnetic fields. This inverse problem is highly nonlinear and ill-posed, posing significant challenges for conventional optimization-based methods, which, despite being grounded in physical models, often fail to recover fine structural details. Recent deep learning strategies, including end-to-end and post-processing networks, have improved reconstruction quality but typically require large paired training datasets and may struggle to generalize. To overcome these limitations, we propose a physics-informed hybrid framework that integrates diffusion models as learned regularization within a data-consistency-driven variational scheme. Specifically, we introduce Single-Step Diffusion Regularization (SSD-Reg), a novel approach that embeds diffusion priors into the iterative reconstruction process, enabling the recovery of complex anatomical structures without the need for paired data. SSD-Reg maintains fidelity to both the governing physics and learned structural distributions, improving accuracy, stability, and robustness. Extensive experiments demonstrate that SSD-Reg, implemented as a Plug-and-Play (PnP) module, provides a flexible and effective solution for tackling the ill-posedness inherent in functional image reconstruction.