In this paper, we address the challenging task of multimodal reasoning by incorporating the notion of "slow thinking" into multimodal large language models (MLLMs). Our core idea is that models can learn to adaptively use different levels of reasoning to tackle questions of varying complexity. We propose a novel paradigm of Self-structured Chain of Thought (SCoT), which consists of minimal semantic atomic steps. Unlike existing methods that rely on structured templates or free-form paradigms, our method not only generates flexible CoT structures for various complex tasks but also mitigates the phenomenon of overthinking for easier tasks. To introduce structured reasoning into visual cognition, we design a novel AtomThink framework with four key modules: (i) a data engine to generate high-quality multimodal reasoning paths; (ii) a supervised fine-tuning (SFT) process with serialized inference data; (iii) a policy-guided multi-turn inference method; and (iv) an atomic capability metric to evaluate the single-step utilization rate. Extensive experiments demonstrate that the proposed AtomThink significantly improves the performance of baseline MLLMs, achieving more than 10% average accuracy gains on MathVista and MathVerse. Compared to state-of-the-art structured CoT approaches, our method not only achieves higher accuracy but also improves data utilization by 5 × and boosts inference efficiency by 85.3%.

GPT-4o, an omni-modal model that enables vocal conversations with diverse emotions and tones, marks a milestone for omni-modal foundation models. However, empowering Large Language Models to perceive and generate images, texts, and speeches end-to-end with publicly available data remains challenging for the open-source community. Existing vision-language models rely on external tools for speech processing, while speech-language models still suffer from limited or totally without vision-understanding capabilities. To address this gap, we propose the EMOVA (EMotionally Omni-present Voice Assistant), to enable Large Language Models with end-to-end speech abilities while maintaining the leading vision-language performance. With a semantic-acoustic disentangled speech tokenizer, we surprisingly notice that omni-modal alignment can further enhance vision-language and speech abilities compared with the bi-modal aligned counterparts. Moreover, a lightweight style module is introduced for the flexible speech style controls including emotions and pitches. For the first time, EMOVA achieves state-of-the-art performance on both the vision-language and speech benchmarks, and meanwhile, supporting omni-modal spoken dialogue with vivid emotions.

We present SeePhys, a large-scale multimodal benchmark for LLM reasoning grounded in physics questions ranging from middle school to PhD qualifying exams. The benchmark covers 7 fundamental domains spanning the physics discipline, incorporating 21 categories of highly heterogeneous diagrams. In contrast to prior works where visual elements mainly serve auxiliary purposes, our benchmark features a substantial proportion of vision-essential problems (75%) that mandate visual information extraction for correct solutions. Through extensive evaluation, we observe that even the most advanced visual reasoning models (e.g., Gemini-2.5-pro and o4-mini) achieve sub-60% accuracy on our benchmark. These results reveal fundamental challenges in current large language models' visual understanding capabilities, particularly in: (i) establishing rigorous coupling between diagram interpretation and physics reasoning, and (ii) overcoming their persistent reliance on textual cues as cognitive shortcuts.Project Page: github.com/SeePhys/seephys-projectHugging Face: huggingface.co/datasets/SeePhys/SeePhys

Recently, numerous deep unfolding networks (DUN) based compressive sensing (CS) methods have been proposed to combine the neural network with conventional optimization algorithms and implemented by a stack of recurrent trainable stages. To obtain high-quality recovered images, it is common practice to employ identical iteration stages for all images. However, in the case of straightforward images with limited high-complexity information, a low number of iterations already yields satisfactory quality. Extending the iteration stages does not notably enhance the quality but results in increased reconstruction time. In this paper, we propose an adaptive stage activated unfolding network (ASAUN-CS), mainly including a measurement assessment module in the compressed domain and a stage selection module in the feature domain. By the proposed method, the images with different recovered complexity can adaptively select the stage number in the unfolding subnet, which reduces the computational time significantly. Comprehensive experimental results demonstrate that the proposed framework obtains satisfying reconstructed performance while using fewer parameters and faster processing speed.

Unmanned aerial vehicles (UAVs) show significant potential in enhancing communication services within the mobile edge computing (MEC) system by taking their advantages on the flexible mobility and reliable line-of-sight links. However, in the scenarios with multiple UAV-MECs (UMs) operating concurrently, potential conflicts in their trajectories need to be mitigated. Thus, the 3-D trajectory needs to be properly designed in a highly reliable manner. Besides, such an infrastructure-free communication paradigm also exposes a potential risk of misuse by malicious parties, which allows them to eavesdrop on private communications, posing a threat to the security and privacy. Therefore, we consider a multi-UAV-assisted MEC communication system, where a UAV maliciously eavesdrops on the data transmission from the user devices (UDs) while a jammer is deployed on the ground to interfere with the eavesdropping channel. In specific, our objective is to minimize the energy consumption and latency while incorporating fairness metrics by optimizing the 3-D trajectories of UMs, transmission power of UDs, and the offloading strategies under the constraints of ensuring communication security and load fairness. Given the complexity of this mixed-integer nonconvex programming problem, we decompose the formulated problem into three subproblems. Specifically, at each time slot, we optimize the transmit power and offloading strategies using theoretical derivation and mathematical analysis, respectively. Additionally, a multiagent deep deterministic policy gradient (MADDPG) algorithm is employed to optimize the trajectories of UMs. Simulation results demonstrate that our proposed joint optimization algorithm successfully minimizes the system energy consumption and delay as compared to benchmarking schemes.

As the COVID-19 pandemic puts pressure on healthcare systems worldwide, the computed tomography image based AI diagnostic system has become a sustainable solution for early diagnosis. However, the model-wise vulnerability under adversarial perturbation hinders its deployment in practical situation. The existing adversarial training strategies are difficult to generalized into medical imaging field challenged by complex medical texture features. To overcome this challenge, we propose a Contour Attention Preserving (CAP) method based on lung cavity edge extraction. The contour prior features are injected to attention layer via a parameter regularization and we optimize the robust empirical risk with hybrid distance metric. We then introduce a new cross-nation CT scan dataset to evaluate the generalization capability of the adversarial robustness under distribution shift. Experimental results indicate that the proposed method achieves state-of-the-art performance in multiple adversarial defense and generalization tasks. The code and dataset are available at https://github.com/Quinn777/CAP.

As an emerging technology, mobile edge computing (MEC) can provide users with higher quality of service (Qos) such as reducing tasks computing latency and energy consumption of user equipments. Unmanned Aerial Vehicle (UAV) -assisted MEC can apply this technology to more scenarios. In this paper, we design a joint optimization algorithm to optimize the user's task offloading strategy and the trajectory of the UAV. When the MEC server interacts with multiple users at the same time, we adopt the differential evolution (DE) algorithm to obtain the offloading policy of each user in the current time slot based on the user location and UAV location. Aiming at the trajectory optimization problem of the UAV, we adopt the optimistic actor-critic (OAC) algorithm, which can minimize the weighted sum of energy consumption and delay of the system, and derive the optimal path through training. Simulation results show that the proposed algorithm is superior to other algorithms in terms of energy consumption and convergence performance.

Satisfiable data quality is the basic guarantee for data-based research, decision-making, and service. Today, new trends in the creation, collection, and utilization of data are constantly emerging. With the usage of massive data, the problem of data quality is highlighted. Several studies on the measurement, evaluation, and management of big data quality have been proposed, and the data quality problem in the big data environment has received attention. The big data characteristics Vs model describes the dimensions and attributes information of data sources in detail, which can be implemented in big data quality measurement. In this paper, a novel rigorous big data quality measurement architecture is proposed for automatically and parallelly quantifying the value of six big data Vs, which are Volume, Variety, Velocity, Veracity, Validity, and Vincularity according to the developed algorithms in every big data process step and time phase of the big data pipeline. Thresholds for the six big data Vs are provided correspondingly for analyzing the result values. The hierarchical measurement model is constructed with multiple-based measures, derived measures, and indicators. The model is verified by comparative experiments and experiments results indicate that the designed architecture can improve the outcomes of data source implementation.

In online health-care community (OHC), patients can choose which physician would provide the expert consultation service. However, the success of registered service depends on many factors including online contributions by physicians in the OHC. This study investigates the moderating effect of physicians' endorsement on their online workload. With data from a leading OHC in China, we designed an empirical study involving 140156 records. The results show that for the physician's contribution behavior, the standardized coefficient was 0.317, and the standardized coefficient of the physician's popularity was 1.588. All estimates were statistically significantly at a 0.1% level. Our study discovered that the popularity of physicians plays a main role in gaining registered service, and their online contributions also increase the volume of this service. Moreover, the physicians from tertiary hospitals, with more thank letters and from larger cities were more potential to gain OHC service, comparing with physicians with higher position, with longer profile words, awards and longer relationship time.