SemiTooth: a Generalizable Semi-supervised Framework for Multi-Source Tooth Segmentation

With the rapid advancement of artificial intelligence, intelligent dentistry for clinical diagnosis and treatment has become increasingly promising. As the primary clinical dentistry task, tooth structure segmentation for Cone-Beam Computed Tomography (CBCT) has made significant progress in recent years. However, challenges arise from the obtainment difficulty of full-annotated data, and the acquisition variability of multi-source data across different institutions, which have caused low-quality utilization, voxel-level inconsistency, and domain-specific disparity in CBCT slices. Thus, the rational and efficient utilization of multi-source and unlabeled data represents a pivotal problem. In this paper, we propose SemiTooth, a generalizable semi-supervised framework for multi-source tooth segmentation. Specifically, we first compile MS3Toothset, Multi-Source Semi-Supervised Tooth DataSet for clinical dental CBCT, which contains data from three sources with different-level annotations. Then, we design a multi-teacher and multi-student framework, i.e., SemiTooth, which promotes semi-supervised learning for multi-source data. SemiTooth employs distinct student networks that learn from unlabeled data with different sources, supervised by its respective teachers. Furthermore, a Stricter Weighted-Confidence Constraint is introduced for multiple teachers to improve the multi-source accuracy.Extensive experiments are conducted on MS3Toothset to verify the feasibility and superiority of the SemiTooth framework, which achieves SOTA performance on the semi-supervised and multi-source tooth segmentation scenario.

Unifying Language-Action Understanding and Generation for Autonomous Driving

Vision-Language-Action (VLA) models are emerging as a promising paradigm for end-to-end autonomous driving, valued for their potential to leverage world knowledge and reason about complex driving scenes. However, existing methods suffer from two critical limitations: a persistent misalignment between language instructions and action outputs, and the inherent inefficiency of typical auto-regressive action generation. In this paper, we introduce LinkVLA, a novel architecture that directly addresses these challenges to enhance both alignment and efficiency. First, we establish a structural link by unifying language and action tokens into a shared discrete codebook, processed within a single multi-modal model. This structurally enforces cross-modal consistency from the ground up. Second, to create a deep semantic link, we introduce an auxiliary action understanding objective that trains the model to generate descriptive captions from trajectories, fostering a bidirectional language-action mapping. Finally, we replace the slow, step-by-step generation with a two-step coarse-to-fine generation method C2F that efficiently decodes the action sequence, saving 86% inference time. Experiments on closed-loop driving benchmarks show consistent gains in instruction following accuracy and driving performance, alongside reduced inference latency.

BaziQA-Benchmark: Evaluating Symbolic and Temporally Compositional Reasoning in Large Language Models

We present BaziQA-Benchmark, a standardized benchmark for evaluating symbolic and temporally compositional reasoning in large language models. The benchmark is derived from 200 professionally curated, multiple-choice problems from the Global Fortune-teller Competition (2021--2025), where each instance requires structured inference over a fixed symbolic chart and interacting temporal conditions. Unlike anecdotal or prompt-driven evaluations, BaziQA-Benchmark enables objective scoring and controlled comparison across years, domains, and model families. We evaluate contemporary language models under a multi-turn setting and analyze performance variation across temporal difficulty, reasoning domains, and inference protocols.To further probe reasoning behavior, we introduce a lightweight Structured Reasoning Protocol that constrains inference order without adding domain knowledge. Results show that models consistently outperform chance but remain far from saturation, exhibiting pronounced sensitivity to temporal composition and reasoning order, as well as systematic failures on precise temporal localization and multi-condition symbolic judgments.

Message Passing based Parameter Estimation in Cooperative MIMO-OFDM ISAC Systems

In integrated sensing and communication (ISAC) networks, multiple base stations (BSs) collaboratively sense a common target, leveraging diversity from multiple observation perspectives and joint signal processing to enhance sensing performance. This paper introduces a novel message-passing (MP)-based parameter estimation framework for collaborative MIMO-OFDM ISAC systems, which jointly estimates the target's position and velocity. First, a signal propagation model is established based on geometric relationships, and a factor graph is constructed to represent the unknown parameters. The sum-product algorithm (SPA) is then applied to this factor graph to jointly estimate the multi-dimensional parameter vector. To reduce communication overhead and computational complexity, we employ a hierarchical message-passing scheme with Gaussian approximation. By adopting parameterized message distributions and layered processing, the proposed method significantly reduces both computational complexity and inter-BS communication overhead. Simulation results demonstrate the effectiveness of the proposed MP-based parameter estimation algorithm and highlight the benefits of multi-perspective observations and joint signal processing for cooperative sensing in MIMO-OFDM ISAC systems.

The Optimal Token Baseline: Variance Reduction for Long-Horizon LLM-RL

Reinforcement Learning (RL) for Large Language Models (LLMs) often suffers from training collapse in long-horizon tasks due to exploding gradient variance. To mitigate this, a baseline is commonly introduced for advantage computation; however, traditional value models remain difficult to optimize, and standard group-based baselines overlook sequence heterogeneity. Although classic optimal baseline theory can achieve global variance reduction, it neglects token heterogeneity and requires prohibitive gradient-based computation. In this work, we derive the Optimal Token Baseline (OTB) from first principles, proving that gradient updates should be weighted inversely to their cumulative gradient norm. To ensure efficiency, we propose the Logit-Gradient Proxy that approximates the gradient norm using only forward-pass probabilities. Our method achieves training stability and matches the performance of large group sizes ($N=32$) with only $N=4$, reducing token consumption by over 65% across single-turn and tool-integrated reasoning tasks.

Dr. Kernel: Reinforcement Learning Done Right for Triton Kernel Generations

High-quality kernel is critical for scalable AI systems, and enabling LLMs to generate such code would advance AI development. However, training LLMs for this task requires sufficient data, a robust environment, and the process is often vulnerable to reward hacking and lazy optimization. In these cases, models may hack training rewards and prioritize trivial correctness over meaningful speedup. In this paper, we systematically study reinforcement learning (RL) for kernel generation. We first design KernelGYM, a robust distributed GPU environment that supports reward hacking check, data collection from multi-turn interactions and long-term RL training. Building on KernelGYM, we investigate effective multi-turn RL methods and identify a biased policy gradient issue caused by self-inclusion in GRPO. To solve this, we propose Turn-level Reinforce-Leave-One-Out (TRLOO) to provide unbiased advantage estimation for multi-turn RL. To alleviate lazy optimization, we incorporate mismatch correction for training stability and introduce Profiling-based Rewards (PR) and Profiling-based Rejection Sampling (PRS) to overcome the issue. The trained model, Dr.Kernel-14B, reaches performance competitive with Claude-4.5-Sonnet in Kernelbench. Finally, we study sequential test-time scaling for Dr.Kernel-14B. On the KernelBench Level-2 subset, 31.6% of the generated kernels achieve at least a 1.2x speedup over the Torch reference, surpassing Claude-4.5-Sonnet (26.7%) and GPT-5 (28.6%). When selecting the best candidate across all turns, this 1.2x speedup rate further increases to 47.8%. All resources, including environment, training code, models, and dataset, are included in https://www.github.com/hkust-nlp/KernelGYM.

Proxy Compression for Language Modeling

Modern language models are trained almost exclusively on token sequences produced by a fixed tokenizer, an external lossless compressor often over UTF-8 byte sequences, thereby coupling the model to that compressor. This work introduces proxy compression, an alternative training scheme that preserves the efficiency benefits of compressed inputs while providing an end-to-end, raw-byte interface at inference time. During training, one language model is jointly trained on raw byte sequences and compressed views generated by external compressors; through the process, the model learns to internally align compressed sequences and raw bytes. This alignment enables strong transfer between the two formats, even when training predominantly on compressed inputs which are discarded at inference. Extensive experiments on code language modeling demonstrate that proxy compression substantially improves training efficiency and significantly outperforms pure byte-level baselines given fixed compute budgets. As model scale increases, these gains become more pronounced, and proxy-trained models eventually match or rival tokenizer approaches, all while operating solely on raw bytes and retaining the inherent robustness of byte-level modeling.

Beyond Precision: Training-Inference Mismatch is an Optimization Problem and Simple LR Scheduling Fixes It

Reinforcement Learning (RL) for training Large Language Models is notoriously unstable. While recent studies attribute this to "training inference mismatch stemming" from inconsistent hybrid engines, standard remedies, such as Importance Sampling, might fail during extended training runs. In this work, we analyze this instability through the lens of optimization, demonstrating that gradient noise and training-inference mismatch escalate in tandem as training progresses. Meanwhile, we find that the mismatch can be effectively suppressed by shrinking the update size. Taken together, we deduce that the mismatch is not merely a static numerical discrepancy, but a dynamic failure coupled with the model's optimization. Based on this insight, we propose a simple yet effective solution: a specialized Learning Rate (LR) scheduler. Instead of pre-defined decay schedule in traditional LR scheduler, our method dynamically triggers LR decay based on response length, which we identify as a reliable early-warning signal for impending instability. Empirical evidence suggests that by reducing the learning rate as gradient noise rises, we can consistently stabilize RL training and keep the training-inference mismatch at a safe level.

Taming the Tail: Stable LLM Reinforcement Learning via Dynamic Vocabulary Pruning

Reinforcement learning for large language models (LLMs) faces a fundamental tension: high-throughput inference engines and numerically-precise training systems produce different probability distributions from the same parameters, creating a training-inference mismatch. We prove this mismatch has an asymmetric effect: the bound on log-probability mismatch scales as $(1-p)$ where $p$ is the token probability. For high-probability tokens, this bound vanishes, contributing negligibly to sequence-level mismatch. For low-probability tokens in the tail, the bound remains large, and moreover, when sampled, these tokens exhibit systematically biased mismatches that accumulate over sequences, destabilizing gradient estimation. Rather than applying post-hoc corrections, we propose constraining the RL objective to a dynamically-pruned ``safe'' vocabulary that excludes the extreme tail. By pruning such tokens, we trade large, systematically biased mismatches for a small, bounded optimization bias. Empirically, our method achieves stable training; theoretically, we bound the optimization bias introduced by vocabulary pruning.

LLM-based Behaviour Driven Development for Hardware Design

Test and verification are essential activities in hardware and system design, but their complexity grows significantly with increasing system sizes. While Behavior Driven Development (BDD) has proven effective in software engineering, it is not yet well established in hardware design, and its practical use remains limited. One contributing factor is the manual effort required to derive precise behavioral scenarios from textual specifications. Recent advances in Large Language Models (LLMs) offer new opportunities to automate this step. In this paper, we investigate the use of LLM-based techniques to support BDD in the context of hardware design.