Thinking Broad, Acting Fast: Latent Reasoning Distillation from Multi-Perspective Chain-of-Thought for E-Commerce Relevance

Effective relevance modeling is crucial for e-commerce search, as it aligns search results with user intent and enhances customer experience. Recent work has leveraged large language models (LLMs) to address the limitations of traditional relevance models, especially for long-tail and ambiguous queries. By incorporating Chain-of-Thought (CoT) reasoning, these approaches improve both accuracy and interpretability through multi-step reasoning. However, two key limitations remain: (1) most existing approaches rely on single-perspective CoT reasoning, which fails to capture the multifaceted nature of e-commerce relevance (e.g., user intent vs. attribute-level matching vs. business-specific rules); and (2) although CoT-enhanced LLM's offer rich reasoning capabilities, their high inference latency necessitates knowledge distillation for real-time deployment, yet current distillation methods discard the CoT rationale structure at inference, using it as a transient auxiliary signal and forfeiting its reasoning utility. To address these challenges, we propose a novel framework that better exploits CoT semantics throughout the optimization pipeline. Specifically, the teacher model leverages Multi-Perspective CoT (MPCoT) to generate diverse rationales and combines Supervised Fine-Tuning (SFT) with Direct Preference Optimization (DPO) to construct a more robust reasoner. For distillation, we introduce Latent Reasoning Knowledge Distillation (LRKD), which endows a student model with a lightweight inference-time latent reasoning extractor, allowing efficient and low-latency internalization of the LLM's sophisticated reasoning capabilities. Evaluated in offline experiments and online A/B tests on an e-commerce search advertising platform serving tens of millions of users daily, our method delivers significant offline gains, showing clear benefits in both commercial performance and user experience.

Reinforced Rate Control for Neural Video Compression via Inter-Frame Rate-Distortion Awareness

Neural video compression (NVC) has demonstrated superior compression efficiency, yet effective rate control remains a significant challenge due to complex temporal dependencies. Existing rate control schemes typically leverage frame content to capture distortion interactions, overlooking inter-frame rate dependencies arising from shifts in per-frame coding parameters. This often leads to suboptimal bitrate allocation and cascading parameter decisions. To address this, we propose a reinforcement-learning (RL)-based rate control framework that formulates the task as a frame-by-frame sequential decision process. At each frame, an RL agent observes a spatiotemporal state and selects coding parameters to optimize a long-term reward that reflects rate-distortion (R-D) performance and bitrate adherence. Unlike prior methods, our approach jointly determines bitrate allocation and coding parameters in a single step, independent of group of pictures (GOP) structure. Extensive experiments across diverse NVC architectures show that our method reduces the average relative bitrate error to 1.20% and achieves up to 13.45% bitrate savings at typical GOP sizes, outperforming existing approaches. In addition, our framework demonstrates improved robustness to content variation and bandwidth fluctuations with lower coding overhead, making it highly suitable for practical deployment.

Glance and Focus Reinforcement for Pan-cancer Screening

Pan-cancer screening in large-scale CT scans remains challenging for existing AI methods, primarily due to the difficulty of localizing diverse types of tiny lesions in large CT volumes. The extreme foreground-background imbalance significantly hinders models from focusing on diseased regions, while redundant focus on healthy regions not only decreases the efficiency but also increases false positives. Inspired by radiologists' glance and focus diagnostic strategy, we introduce GF-Screen, a Glance and Focus reinforcement learning framework for pan-cancer screening. GF-Screen employs a Glance model to localize the diseased regions and a Focus model to precisely segment the lesions, where segmentation results of the Focus model are leveraged to reward the Glance model via Reinforcement Learning (RL). Specifically, the Glance model crops a group of sub-volumes from the entire CT volume and learns to select the sub-volumes with lesions for the Focus model to segment. Given that the selecting operation is non-differentiable for segmentation training, we propose to employ the segmentation results to reward the Glance model. To optimize the Glance model, we introduce a novel group relative learning paradigm, which employs group relative comparison to prioritize high-advantage predictions and discard low-advantage predictions within sub-volume groups, not only improving efficiency but also reducing false positives. In this way, for the first time, we effectively extend cutting-edge RL techniques to tackle the specific challenges in pan-cancer screening. Extensive experiments on 16 internal and 7 external datasets across 9 lesion types demonstrated the effectiveness of GF-Screen. Notably, GF-Screen leads the public validation leaderboard of MICCAI FLARE25 pan-cancer challenge, surpassing the FLARE24 champion solution by a large margin (+25.6% DSC and +28.2% NSD).

PMPBench: A Paired Multi-Modal Pan-Cancer Benchmark for Medical Image Synthesis

Contrast medium plays a pivotal role in radiological imaging, as it amplifies lesion conspicuity and improves detection for the diagnosis of tumor-related diseases. However, depending on the patient's health condition or the medical resources available, the use of contrast medium is not always feasible. Recent work has explored AI-based image translation to synthesize contrast-enhanced images directly from non-contrast scans, aims to reduce side effects and streamlines clinical workflows. Progress in this direction has been constrained by data limitations: (1) existing public datasets focus almost exclusively on brain-related paired MR modalities; (2) other collections include partially paired data but suffer from missing modalities/timestamps and imperfect spatial alignment; (3) explicit labeling of CT vs. CTC or DCE phases is often absent; (4) substantial resources remain private. To bridge this gap, we introduce the first public, fully paired, pan-cancer medical imaging dataset spanning 11 human organs. The MR data include complete dynamic contrast-enhanced (DCE) sequences covering all three phases (DCE1-DCE3), while the CT data provide paired non-contrast and contrast-enhanced acquisitions (CTC). The dataset is curated for anatomical correspondence, enabling rigorous evaluation of 1-to-1, N-to-1, and N-to-N translation settings (e.g., predicting DCE phases from non-contrast inputs). Built upon this resource, we establish a comprehensive benchmark. We report results from representative baselines of contemporary image-to-image translation. We release the dataset and benchmark to catalyze research on safe, effective contrast synthesis, with direct relevance to multi-organ oncology imaging workflows. Our code and dataset are publicly available at https://github.com/YifanChen02/PMPBench.

S2DiT: Sandwich Diffusion Transformer for Mobile Streaming Video Generation

Diffusion Transformers (DiTs) have recently improved video generation quality. However, their heavy computational cost makes real-time or on-device generation infeasible. In this work, we introduce S2DiT, a Streaming Sandwich Diffusion Transformer designed for efficient, high-fidelity, and streaming video generation on mobile hardware. S2DiT generates more tokens but maintains efficiency with novel efficient attentions: a mixture of LinConv Hybrid Attention (LCHA) and Stride Self-Attention (SSA). Based on this, we uncover the sandwich design via a budget-aware dynamic programming search, achieving superior quality and efficiency. We further propose a 2-in-1 distillation framework that transfers the capacity of large teacher models (e.g., Wan 2.2-14B) to the compact few-step sandwich model. Together, S2DiT achieves quality on par with state-of-the-art server video models, while streaming at over 10 FPS on an iPhone.

Integrated Sensing and Semantic Communication with Adaptive Source-Channel Coding

Semantic communication has emerged as a new paradigm to facilitate the performance of integrated sensing and communication systems in 6G. However, most of the existing works mainly focus on sensing data compression to reduce the subsequent communication overheads, without considering the integrated transmission framework for both the SemCom and sensing tasks. This paper proposes an adaptive source-channel coding and beamforming design framework for integrated sensing and SemCom systems by jointly optimizing the coding rate for SemCom task and the transmit beamforming for both the SemCom and sensing tasks. Specifically, an end-to-end semantic distortion function is approximated by deriving an upper bound composing of source and channel coding induced components, and then a hybrid Cramér-Rao bound (HCRB) is also derived for target position under imperfect time synchronization. To facilitate the joint optimization, a distortion minimization problem is formulated by considering the HCRB threshold, channel uses, and power budget. Subsequently, an alternative optimization algorithm composed of successive convex approximation and fractional programming is proposed to address this problem by decoupling it into two subproblems for coding rate and beamforming designs, respectively. Simulation results demonstrate that our proposed scheme outperforms the conventional deep joint source-channel coding -water filling-zero forcing benchmark.

CausalSpatial: A Benchmark for Object-Centric Causal Spatial Reasoning

Humans can look at a static scene and instantly predict what happens next -- will moving this object cause a collision? We call this ability Causal Spatial Reasoning. However, current multimodal large language models (MLLMs) cannot do this, as they remain largely restricted to static spatial perception, struggling to answer "what-if" questions in a 3D scene. We introduce CausalSpatial, a diagnostic benchmark evaluating whether models can anticipate consequences of object motions across four tasks: Collision, Compatibility, Occlusion, and Trajectory. Results expose a severe gap: humans score 84% while GPT-5 achieves only 54%. Why do MLLMs fail? Our analysis uncovers a fundamental deficiency: models over-rely on textual chain-of-thought reasoning that drifts from visual evidence, producing fluent but spatially ungrounded hallucinations. To address this, we propose the Causal Object World model (COW), a framework that externalizes the simulation process by generating videos of hypothetical dynamics. With explicit visual cues of causality, COW enables models to ground their reasoning in physical reality rather than linguistic priors. We make the dataset and code publicly available here: https://github.com/CausalSpatial/CausalSpatial

TRIM: Hybrid Inference via Targeted Stepwise Routing in Multi-Step Reasoning Tasks

Multi-step reasoning tasks like mathematical problem solving are vulnerable to cascading failures, where a single incorrect step leads to complete solution breakdown. Current LLM routing methods assign entire queries to one model, treating all reasoning steps as equal. We propose TRIM (Targeted routing in multi-step reasoning tasks), which routes only critical steps$\unicode{x2013}$those likely to derail the solution$\unicode{x2013}$to larger models while letting smaller models handle routine continuations. Our key insight is that targeted step-level interventions can fundamentally transform inference efficiency by confining expensive calls to precisely those steps where stronger models prevent cascading errors. TRIM operates at the step-level: it uses process reward models to identify erroneous steps and makes routing decisions based on step-level uncertainty and budget constraints. We develop several routing strategies within TRIM, ranging from a simple threshold-based policy to more expressive policies that reason about long-horizon accuracy-cost trade-offs and uncertainty in step-level correctness estimates. On MATH-500, even the simplest thresholding strategy surpasses prior routing methods with 5x higher cost efficiency, while more advanced policies match the strong, expensive model's performance using 80% fewer expensive model tokens. On harder benchmarks such as AIME, TRIM achieves up to 6x higher cost efficiency. All methods generalize effectively across math reasoning tasks, demonstrating that step-level difficulty represents fundamental characteristics of reasoning.

Training-Trajectory-Aware Token Selection

Efficient distillation is a key pathway for converting expensive reasoning capability into deployable efficiency, yet in the frontier regime where the student already has strong reasoning ability, naive continual distillation often yields limited gains or even degradation. We observe a characteristic training phenomenon: even as loss decreases monotonically, all performance metrics can drop sharply at almost the same bottleneck, before gradually recovering. We further uncover a token-level mechanism: confidence bifurcates into steadily increasing Imitation-Anchor Tokens that quickly anchor optimization and other yet-to-learn tokens whose confidence is suppressed until after the bottleneck. And the characteristic that these two types of tokens cannot coexist is the root cause of the failure in continual distillation. To this end, we propose Training-Trajectory-Aware Token Selection (T3S) to reconstruct the training objective at the token level, clearing the optimization path for yet-to-learn tokens. T3 yields consistent gains in both AR and dLLM settings: with only hundreds of examples, Qwen3-8B surpasses DeepSeek-R1 on competitive reasoning benchmarks, Qwen3-32B approaches Qwen3-235B, and T3-trained LLaDA-2.0-Mini exceeds its AR baseline, achieving state-of-the-art performance among all of 16B-scale no-think models.

High-fidelity lunar topographic reconstruction across diverse terrain and illumination environments using deep learning

Topographic models are essential for characterizing planetary surfaces and for inferring underlying geological processes. Nevertheless, meter-scale topographic data remain limited, which constrains detailed planetary investigations, even for the Moon, where extensive high-resolution orbital images are available. Recent advances in deep learning (DL) exploit single-view imagery, constrained by low-resolution topography, for fast and flexible reconstruction of fine-scale topography. However, their robustness and general applicability across diverse lunar landforms and illumination conditions remain insufficiently explored. In this study, we build upon our previously proposed DL framework by incorporating a more robust scale recovery scheme and extending the model to polar regions under low solar illumination conditions. We demonstrate that, compared with single-view shape-from-shading methods, the proposed DL approach exhibits greater robustness to varying illumination and achieves more consistent and accurate topographic reconstructions. Furthermore, it reliably reconstructs topography across lunar features of diverse scales, morphologies, and geological ages. High-quality topographic models are also produced for the lunar south polar areas, including permanently shadowed regions, demonstrating the method's capability in reconstructing complex and low-illumination terrain. These findings suggest that DL-based approaches have the potential to leverage extensive lunar datasets to support advanced exploration missions and enable investigations of the Moon at unprecedented topographic resolution.