A2R: An Asymmetric Two-Stage Reasoning Framework for Parallel Reasoning

Recent Large Reasoning Models have achieved significant improvements in complex task-solving capabilities by allocating more computation at the inference stage with a "thinking longer" paradigm. Even as the foundational reasoning capabilities of models advance rapidly, the persistent gap between a model's performance in a single attempt and its latent potential, often revealed only across multiple solution paths, starkly highlights the disparity between its realized and inherent capabilities. To address this, we present A2R, an Asymmetric Two-Stage Reasoning framework designed to explicitly bridge the gap between a model's potential and its actual performance. In this framework, an "explorer" model first generates potential solutions in parallel through repeated sampling. Subsequently,a "synthesizer" model integrates these references for a more refined, second stage of reasoning. This two-stage process allows computation to be scaled orthogonally to existing sequential methods. Our work makes two key innovations: First, we present A2R as a plug-and-play parallel reasoning framework that explicitly enhances a model's capabilities on complex questions. For example, using our framework, the Qwen3-8B-distill model achieves a 75% performance improvement compared to its self-consistency baseline. Second, through a systematic analysis of the explorer and synthesizer roles, we identify an effective asymmetric scaling paradigm. This insight leads to A2R-Efficient, a "small-to-big" variant that combines a Qwen3-4B explorer with a Qwen3-8B synthesizer. This configuration surpasses the average performance of a monolithic Qwen3-32B model at a nearly 30% lower cost. Collectively, these results show that A2R is not only a performance-boosting framework but also an efficient and practical solution for real-world applications.

Design and Control of a Perching Drone Inspired by the Prey-Capturing Mechanism of Venus Flytrap

The endurance and energy efficiency of drones remain critical challenges in their design and operation. To extend mission duration, numerous studies explored perching mechanisms that enable drones to conserve energy by temporarily suspending flight. This paper presents a new perching drone that utilizes an active flexible perching mechanism inspired by the rapid predation mechanism of the Venus flytrap, achieving perching in less than 100 ms. The proposed system is designed for high-speed adaptability to the perching targets. The overall drone design is outlined, followed by the development and validation of the biomimetic perching structure. To enhance the system stability, a cascade extended high-gain observer (EHGO) based control method is developed, which can estimate and compensate for the external disturbance in real time. The experimental results demonstrate the adaptability of the perching structure and the superiority of the cascaded EHGO in resisting wind and perching disturbances.

LoRA in LoRA: Towards Parameter-Efficient Architecture Expansion for Continual Visual Instruction Tuning

Continual Visual Instruction Tuning (CVIT) enables Multimodal Large Language Models (MLLMs) to incrementally learn new tasks over time. However, this process is challenged by catastrophic forgetting, where performance on previously learned tasks deteriorates as the model adapts to new ones. A common approach to mitigate forgetting is architecture expansion, which introduces task-specific modules to prevent interference. Yet, existing methods often expand entire layers for each task, leading to significant parameter overhead and poor scalability. To overcome these issues, we introduce LoRA in LoRA (LiLoRA), a highly efficient architecture expansion method tailored for CVIT in MLLMs. LiLoRA shares the LoRA matrix A across tasks to reduce redundancy, applies an additional low-rank decomposition to matrix B to minimize task-specific parameters, and incorporates a cosine-regularized stability loss to preserve consistency in shared representations over time. Extensive experiments on a diverse CVIT benchmark show that LiLoRA consistently achieves superior performance in sequential task learning while significantly improving parameter efficiency compared to existing approaches.

Can Mixture-of-Experts Surpass Dense LLMs Under Strictly Equal Resources?

Mixture-of-Experts (MoE) language models dramatically expand model capacity and achieve remarkable performance without increasing per-token compute. However, can MoEs surpass dense architectures under strictly equal resource constraints - that is, when the total parameter count, training compute, and data budget are identical? This question remains under-explored despite its significant practical value and potential. In this paper, we propose a novel perspective and methodological framework to study this question thoroughly. First, we comprehensively investigate the architecture of MoEs and achieve an optimal model design that maximizes the performance. Based on this, we subsequently find that an MoE model with activation rate in an optimal region is able to outperform its dense counterpart under the same total parameter, training compute and data resource. More importantly, this optimal region remains consistent across different model sizes. Although additional amount of data turns out to be a trade-off for the enhanced performance, we show that this can be resolved via reusing data. We validate our findings through extensive experiments, training nearly 200 language models at 2B scale and over 50 at 7B scale, cumulatively processing 50 trillion tokens. All models will be released publicly.

RM-R1: Reward Modeling as Reasoning

Reward modeling is essential for aligning large language models (LLMs) with human preferences, especially through reinforcement learning from human feedback (RLHF). To provide accurate reward signals, a reward model (RM) should stimulate deep thinking and conduct interpretable reasoning before assigning a score or a judgment. However, existing RMs either produce opaque scalar scores or directly generate the prediction of a preferred answer, making them struggle to integrate natural language critiques, thus lacking interpretability. Inspired by recent advances of long chain-of-thought (CoT) on reasoning-intensive tasks, we hypothesize and validate that integrating reasoning capabilities into reward modeling significantly enhances RM's interpretability and performance. In this work, we introduce a new class of generative reward models -- Reasoning Reward Models (ReasRMs) -- which formulate reward modeling as a reasoning task. We propose a reasoning-oriented training pipeline and train a family of ReasRMs, RM-R1. The training consists of two key stages: (1) distillation of high-quality reasoning chains and (2) reinforcement learning with verifiable rewards. RM-R1 improves LLM rollouts by self-generating reasoning traces or chat-specific rubrics and evaluating candidate responses against them. Empirically, our models achieve state-of-the-art or near state-of-the-art performance of generative RMs across multiple comprehensive reward model benchmarks, outperforming much larger open-weight models (e.g., Llama3.1-405B) and proprietary ones (e.g., GPT-4o) by up to 13.8%. Beyond final performance, we perform thorough empirical analysis to understand the key ingredients of successful ReasRM training. To facilitate future research, we release six ReasRM models along with code and data at https://github.com/RM-R1-UIUC/RM-R1.

Embodied Neuromorphic Control Applied on a 7-DOF Robotic Manipulator

The development of artificial intelligence towards real-time interaction with the environment is a key aspect of embodied intelligence and robotics. Inverse dynamics is a fundamental robotics problem, which maps from joint space to torque space of robotic systems. Traditional methods for solving it rely on direct physical modeling of robots which is difficult or even impossible due to nonlinearity and external disturbance. Recently, data-based model-learning algorithms are adopted to address this issue. However, they often require manual parameter tuning and high computational costs. Neuromorphic computing is inherently suitable to process spatiotemporal features in robot motion control at extremely low costs. However, current research is still in its infancy: existing works control only low-degree-of-freedom systems and lack performance quantification and comparison. In this paper, we propose a neuromorphic control framework to control 7 degree-of-freedom robotic manipulators. We use Spiking Neural Network to leverage the spatiotemporal continuity of the motion data to improve control accuracy, and eliminate manual parameters tuning. We validated the algorithm on two robotic platforms, which reduces torque prediction error by at least 60% and performs a target position tracking task successfully. This work advances embodied neuromorphic control by one step forward from proof of concept to applications in complex real-world tasks.

The Tenth NTIRE 2025 Image Denoising Challenge Report

This paper presents an overview of the NTIRE 2025 Image Denoising Challenge ({\sigma} = 50), highlighting the proposed methodologies and corresponding results. The primary objective is to develop a network architecture capable of achieving high-quality denoising performance, quantitatively evaluated using PSNR, without constraints on computational complexity or model size. The task assumes independent additive white Gaussian noise (AWGN) with a fixed noise level of 50. A total of 290 participants registered for the challenge, with 20 teams successfully submitting valid results, providing insights into the current state-of-the-art in image denoising.

MMAIF: Multi-task and Multi-degradation All-in-One for Image Fusion with Language Guidance

Image fusion, a fundamental low-level vision task, aims to integrate multiple image sequences into a single output while preserving as much information as possible from the input. However, existing methods face several significant limitations: 1) requiring task- or dataset-specific models; 2) neglecting real-world image degradations (\textit{e.g.}, noise), which causes failure when processing degraded inputs; 3) operating in pixel space, where attention mechanisms are computationally expensive; and 4) lacking user interaction capabilities. To address these challenges, we propose a unified framework for multi-task, multi-degradation, and language-guided image fusion. Our framework includes two key components: 1) a practical degradation pipeline that simulates real-world image degradations and generates interactive prompts to guide the model; 2) an all-in-one Diffusion Transformer (DiT) operating in latent space, which fuses a clean image conditioned on both the degraded inputs and the generated prompts. Furthermore, we introduce principled modifications to the original DiT architecture to better suit the fusion task. Based on this framework, we develop two versions of the model: Regression-based and Flow Matching-based variants. Extensive qualitative and quantitative experiments demonstrate that our approach effectively addresses the aforementioned limitations and outperforms previous restoration+fusion and all-in-one pipelines. Codes are available at https://github.com/294coder/MMAIF.

MobiVital: Self-supervised Time-series Quality Estimation for Contactless Respiration Monitoring Using UWB Radar

Respiration waveforms are increasingly recognized as important biomarkers, offering insights beyond simple respiration rates, such as detecting breathing irregularities for disease diagnosis or monitoring breath patterns to guide rehabilitation training. Previous works in wireless respiration monitoring have primarily focused on estimating respiration rate, where the breath waveforms are often generated as a by-product. As a result, issues such as waveform deformation and inversion have largely been overlooked, reducing the signal's utility for applications requiring breathing waveforms. To address this problem, we present a novel approach, MobiVital, that improves the quality of respiration waveforms obtained from ultra-wideband (UWB) radar data. MobiVital combines a self-supervised autoregressive model for breathing waveform extraction with a biology-informed algorithm to detect and correct waveform inversions. To encourage reproducible research efforts for developing wireless vital signal monitoring systems, we also release a 12-person, 24-hour UWB radar vital signal dataset, with time-synchronized ground truth obtained from wearable sensors. Our results show that the respiration waveforms produced by our system exhibit a 7-34% increase in fidelity to the ground truth compared to the baselines and can benefit downstream tasks such as respiration rate estimation.

Entailment-Preserving First-order Logic Representations in Natural Language Entailment

First-order logic (FOL) can represent the logical entailment semantics of natural language (NL) sentences, but determining natural language entailment using FOL remains a challenge. To address this, we propose the Entailment-Preserving FOL representations (EPF) task and introduce reference-free evaluation metrics for EPF, the Entailment-Preserving Rate (EPR) family. In EPF, one should generate FOL representations from multi-premise natural language entailment data (e.g. EntailmentBank) so that the automatic prover's result preserves the entailment labels. Experiments show that existing methods for NL-to-FOL translation struggle in EPF. To this extent, we propose a training method specialized for the task, iterative learning-to-rank, which directly optimizes the model's EPR score through a novel scoring function and a learning-to-rank objective. Our method achieves a 1.8-2.7% improvement in EPR and a 17.4-20.6% increase in EPR@16 compared to diverse baselines in three datasets. Further analyses reveal that iterative learning-to-rank effectively suppresses the arbitrariness of FOL representation by reducing the diversity of predicate signatures, and maintains strong performance across diverse inference types and out-of-domain data.