Recent multimodal large language models (MLLMs) increasingly integrate multiple vision encoders to improve performance on various benchmarks, assuming that diverse pretraining objectives yield complementary visual signals. However, we show this assumption often fails in practice. Through systematic encoder masking across representative multi-encoder MLLMs, we find that performance typically degrades gracefully--and sometimes even improves--when selected encoders are masked, revealing pervasive encoder redundancy. To quantify this effect, we introduce two principled metrics: the Conditional Utilization Rate (CUR), which measures an encoder's marginal contribution in the presence of others, and the Information Gap (IG), which captures heterogeneity in encoder utility within a model. Using these tools, we observe: (i) strong specialization on tasks like OCR & Chart, where a single encoder can dominate with a CUR > 90%, (ii) high redundancy on general VQA and knowledge-based tasks, where encoders are largely interchangeable, (iii) instances of detrimental encoders with negative CUR. Notably, masking specific encoders can yield up to 16% higher accuracy on a specific task category and 3.6% overall performance boost compared to the full model. Furthermore, single-and dual-encoder variants recover over 90% of baseline on most non-OCR tasks. Our analysis challenges the "more encoders are better" heuristic in MLLMs and provides actionable diagnostics for developing more efficient and effective multimodal architectures. Multimodal large language models (MLLMs) have marked a major leap in artificial intelligence (AI), exhibiting remarkable prowess in integrating visual and textual information for complex generation and reasoning tasks (OpenAI, 2025a; DeepMind, 2025; Anthropic, 2024; Bai et al., 2025; Zhu et al., 2025). Their ability to interpret images (Luo et al., 2024), answer visual questions (Zhu et al., 2025; Li et al., 2025a), and perform visual reasoning (OpenAI, 2025b; Peng et al., 2025) has positioned them at the forefront of AI research.

We present a novel approach to compute three-dimensional Magnetohydrodynamic equilibria by parametrizing Fourier modes with artificial neural networks and compare it to equilibria computed by conventional solvers. The full nonlinear global force residual across the volume in real space is then minimized with first order optimizers. Already,we observe competitive computational cost to arrive at the same minimum residuals computed by existing codes. With increased computational cost,lower minima of the residual are achieved by the neural networks,establishing a new lower bound for the force residual. We use minimally complex neural networks,and we expect significant improvements for solving not only single equilibria with neural networks,but also for computing neural network models valid over continuous distributions of equilibria.

Recent multi-modal Large Language Models (LLMs) such as GPT-4o have demonstrated strong capabilities of direct speech interaction. However, the lack of specialized and comprehensive benchmarks for end-to-end speech LLM evaluation hinders optimizing the user experience of Audio LLMs in real-world applications. Existing evaluation methods often adapt text-based benchmarks, overlooking speech's unique characteristics and challenges, including prosody, homophones, stuttering, and differing user expectations. Here, we introduce the first comprehensive benchmark designed to systematically evaluate end-to-end speechLLMs in practical speech conversations. We systematically curate real-world chat data relevant to spoken scenarios, introduce diversity in speaker attributes and acoustic conditions, and augment the dataset with speech-specific phenomena. We further design a query-aware evaluation method to use customized evaluation checklists and prompts to enhance the accuracy of automatic evaluation. We conduct comprehensive testing and detailed analysis of various mainstream speech models, revealing significant differences in model performance across different speech scenarios. The use of query-aware evaluation further enables a finer-grained assessment under various speech-specific scenarios. Our benchmark can provide valuable insights for speech model development and evaluation.

Multimodal Knowledge Graphs (MMKGs), which represent explicit knowledge across multiple modalities, play a pivotal role by complementing the implicit knowledge of Multimodal Large Language Models (MLLMs) and enabling more grounded reasoning via Retrieval Augmented Generation (RAG). However, existing MMKGs are generally limited in scope: they are often constructed by augmenting pre-existing knowledge graphs, which restricts their knowledge, resulting in outdated or incomplete knowledge coverage, and they often support only a narrow range of modalities, such as text and visual information. These limitations restrict applicability to multimodal tasks, particularly as recent MLLMs adopt richer modalities like video and audio. Therefore, we propose the Visual-Audio-Text Knowledge Graph (VAT-KG), the first concept-centric and knowledge-intensive multimodal knowledge graph that covers visual, audio, and text information, where each triplet is linked to multimodal data and enriched with detailed descriptions of concepts. Specifically, our construction pipeline ensures cross-modal knowledge alignment between multimodal data and fine-grained semantics through a series of stringent filtering and alignment steps, enabling the automatic generation of MMKGs from any multimodal dataset. We further introduce a novel multimodal RAG framework that retrieves detailed concept-level knowledge in response to queries from arbitrary modalities. Experiments on question answering tasks across various modalities demonstrate the effectiveness of VAT-KG in supporting MLLMs, highlighting its practical value in unifying and leveraging multimodal knowledge.

We present video-SALMONN 2, a family of audio-visual large language models that set new state-of-the-art (SOTA) results in video description and question answering (QA). Our core contribution is multi-round direct preference optimisation (MrDPO), paired with a caption-quality objective that jointly rewards completeness and factual accuracy. Unlike standard DPO with a fixed reference policy, MrDPO periodically refreshes the reference by bootstrapping from a newly re-initialised lightweight adapter trained on the latest preferences, avoiding reference staleness and enabling continual improvement. This strategy produces captions that are consistently more detailed and accurate than those from proprietary systems such as GPT-4o and Gemini-1.5 Pro. We further distil these gains by using our model to generate a high-quality video-caption corpus for supervised fine-tuning of new models, transferring benefits beyond captioning to strong performance on complex video-QA tasks. Across widely used audio-visual and visual-only understanding benchmarks (including Video-MME, WorldSense, AVUT, Video-Holmes, DailyOmni, MLVU, and LVBench), our 3B and 7B models achieve SOTA results at comparable scales, while the 72B model surpasses all other open-source systems. Our source code, models, and data are released at \href{https://github.com/bytedance/video-SALMONN-2}{https://github.com/bytedance/video-SALMONN-2}.

Recent advances in large language models (LLMs), particularly those enhanced through reinforced post-training, have demonstrated impressive reasoning capabilities, as exemplified by models such as OpenAI o1 and DeepSeek-R1. However, these capabilities are predominantly benchmarked on domains like mathematical problem solving and code generation, leaving open the question of whether such reasoning skills generalize to complex real-world scenarios. In this paper, we introduce LocationReasoner, a benchmark designed to evaluate LLMs' reasoning abilities in the context of real-world site selection, where models must identify feasible locations by reasoning over diverse and complicated spatial, environmental, and logistic constraints. The benchmark covers carefully crafted queries of varying difficulty levels and is supported by a sandbox environment with in-house tools for constraint-based location search. Automated verification further guarantees the scalability of the benchmark, enabling the addition of arbitrary number of queries. Extensive evaluations on real-world site selection data from Boston, New York, and Tampa reveal that state-of-the-art reasoning models offer limited improvement over their non-reasoning predecessors in real-world contexts, with even the latest OpenAI o4 model failing on 30% of site selection tasks. Moreover, agentic strategies such as ReAct and Reflexion often suffer from over-reasoning, leading to worse outcomes than direct prompting. With key limitations of LLMs in holistic and non-linear reasoning highlighted, we release LocationReasoner to foster the development of LLMs and agents capable of robust, grounded reasoning in real-world decision-making tasks. Codes and data for our benchmark are available at https://github.com/miho-koda/LocationReasoner.

Training large language models typically demands extensive GPU memory and substantial financial investment, which poses a barrier for many small- to medium-sized teams. In this paper, we propose a full-parameter pre-training and fine-tuning framework based on block coordinate descent (BCD), enhanced with engineering optimizations, to enable efficient training of large-scale models on cost-effective RTX 4090, A100 and A800 GPU clusters. Under identical hardware configurations, we reduce the training cost of a 7B model to 33% on A100/A800 and only 2.6% on RTX 4090, compared to standard full-parameter training. It also enables large models previously restricted to A100 clusters to be trained on RTX 4090 without degrading performance. BCD achieves comparable or better accuracy than full-parameter and fine-tuning methods at most cases, with lower GPU consumption and improved hardware utilization.

Sensor-based local inference at IoT devices faces severe computational limitations, often requiring data transmission over noisy wireless channels for server-side processing. To address this, split-network Deep Neural Network (DNN) based Joint Source-Channel Coding (JSCC) schemes are used to extract and transmit relevant features instead of raw data. However, most existing methods rely on fixed network splits and static configurations, lacking adaptability to varying computational budgets and channel conditions. In this paper, we propose a novel SNR- and computation-adaptive distributed CNN framework for wireless image classification across IoT devices and edge servers. We introduce a learning-assisted intelligent Genetic Algorithm (LAIGA) that efficiently explores the CNN hyperparameter space to optimize network configuration under given FLOPs constraints and given SNR. LAIGA intelligently discards the infeasible network configurations that exceed computational budget at IoT device. It also benefits from the Random Forests based learning assistance to avoid a thorough exploration of hyperparameter space and to induce application specific bias in candidate optimal configurations. Experimental results demonstrate that the proposed framework outperforms fixed-split architectures and existing SNR-adaptive methods, especially under low SNR and limited computational resources. We achieve a 10\% increase in classification accuracy as compared to existing JSCC based SNR-adaptive multilayer framework at an SNR as low as -10dB across a range of available computational budget (1M to 70M FLOPs) at IoT device.

The "Reason-Then-Respond" paradigm, enhanced by Reinforcement Learning, has shown great promise in advancing Multimodal Large Language Models. However, its application to the video domain has led to specialized models that excel at either question answering (QA) or captioning tasks, but struggle to master both. Naively combining reward signals from these tasks results in mutual performance degradation, which we attribute to a conflict between their opposing task natures. To address this challenge, we propose a novel training framework built upon two intermediate proxy tasks: DarkEventInfer, which presents videos with masked event segments, requiring models to infer the obscured content based on contextual video cues; and MixVidQA, which presents interleaved video sequences composed of two distinct clips, challenging models to isolate and reason about one while disregarding the other. These proxy tasks compel the model to simultaneously develop both holistic, divergent understanding and precise, convergent reasoning capabilities. Extensive experiments show that VidBridge-R1 achieves significant performance gains on both QA and captioning within one model, demonstrating the efficacy of our approach in fostering more generaliz-able and powerful video understanding models. The release of OpenAI o1/o3 (Jaech et al., 2024) and DeepSeek-R1 (Guo et al., 2025) has introduced a novel Reason-Then-Respond paradigm to the development of large language models (LLMs), which significantly enhances model performance through test-time scaling. Inspired by this approach, a growing body of research (Team et al., 2025; Chen et al., 2025a; Shen et al., 2025; Deng et al., 2025; Xia et al., 2025; Y ao et al., 2025) has extended this paradigm to multimodal large language models (MLLMs). By leveraging reinforcement learning (RL), particularly the Group Relative Policy Optimization (GRPO) algorithm (Shao et al., 2024), these studies have achieved promising results in image-based reasoning tasks. Recently, several studies (Feng et al., 2025; Zhang et al., 2025b; Chen et al., 2025e;f) have begun to explore the application of the Reason-Then-Respond paradigm in the video modality. Some efforts focus on enhancing question answering (QA) capabilities in general or reasoning scenarios (Li et al., 2025b; Dang et al., 2025), while some other works concentrate solely on improving video captioning performance (Li et al., 2025a; Meng et al., 2025a). This work was conducted during the author's internship at Kling Team, Kuaishou Technology A key advantage of MLLMs lies in their versatility, enabling strong performance across diverse tasks simultaneously. It is therefore undesirable to enhance reasoning capabilities at the expense of generalizability by over-specializing the model in a single task. To preserve generality in both QA and captioning tasks, an intuitive approach is to combine the reward signals from them during training.

Aircraft trajectory modeling plays a crucial role in air traffic management (ATM) and is important for various downstream tasks, including conflict detection and landing time prediction. Dataset augmentation by adding synthetically generated trajectory data is necessary to develop a more robust aircraft trajectory model and ensure that the trajectory dataset is sufficient and balanced. We propose a novel framework called ATRADA for aircraft trajectory dataset augmentation. In the proposed framework, a Transformer encoder learns the underlying patterns in the original trajectory dataset and converts each data point into a context vector in the learned latent space. The converted dataset is projected to reduced dimensions using principal component analysis (PCA), and a Gaussian mixture model (GMM) is applied to fit the probability distribution of the data points in the reduced-dimensional space. Finally, new samples are drawn from the fitted GMM, the dimension of the samples is reverted to the original dimension, and the samples are decoded with a multi-layer perceptron (MLP). Several experiments demonstrate that the framework effectively generates new, high-quality synthetic aircraft trajectory data, which were compared to the results of several baselines.