Jiaqi Zhang

Deli Huang, Cunguang Wang, Hongyin Tang et al.

Long-context language models now advertise context windows up to millions of tokens, yet evaluations typically report a single length or a narrow task family, masking two failure modes: performance can collapse as length grows, and strong retrieval need not transfer to downstream use. We present ATLAS, a benchmarking framework that redefines long-context evaluation as length-dependent capability profiling. ATLAS contributes three methodological principles:(i) a layered taxonomy separating foundational operations from application workloads so failures can be attributed, (ii) length-aware AUC scoring that integrates score-length curves over a fixed 8K-1M grid, replacing single-point metrics with full degradation profiles, and (iii) ATLAScore, a harmonic-mean aggregate over taxonomy categories that penalizes imbalanced profiles, with end-to-end uncertainty propagation from subset scores through the nonlinear final aggregate. We instantiate the framework across eight capability dimensions with nine auditable components and 6,438 instances, and evaluate 26 models. Gemini-3.1-Pro-Preview leads at 128K, Claude-Opus-4.6 leads at 1M. Rankings reshuffle substantially between ATLASscore@8K-128K and ATLASscore@8K-1M: 7 models move by at least two ranks, and the two taxonomy layers share only 61% of cross-model variance, with individual rank gaps up to 12 positions. These results support reporting long-context quality by capability and length, not by a single headline score.

Jiaqi Zhang, Keyou You, Tamer Başar

This paper proposes distributed discrete-time algorithms to cooperatively solve an additive cost optimization problem in multi-agent networks. The striking feature lies in the use of only the sign of relative state information between neighbors, which substantially differentiates our algorithms from others in the existing literature. We first interpret the proposed algorithms in terms of the penalty method in optimization theory and then perform non-asymptotic analysis to study convergence for static network graphs. Compared with the celebrated distributed subgradient algorithms, which however use the exact relative state information, the convergence speed is essentially not affected by the loss of information. We also study how introducing noise into the relative state information and randomly activated graphs affect the performance of our algorithms. Finally, we validate the theoretical results on a class of distributed quantile regression problems.

Jiaqi Zhang, Keyou You, Lihua Xie

This work studies the state estimation problem of a stochastic nonlinear system with unknown sensor measurement losses. If the estimator knows the sensor measurement losses of a linear Gaussian system, the minimum variance estimate is easily computed by the celebrated intermittent Kalman filter (IKF). However, this will no longer be the case when the measurement losses are unknown and/or the system is nonlinear or non-Gaussian. By exploiting the binary property of the measurement loss process and the IKF, we design three suboptimal filters for the state estimation, i.e., BKF-I, BKF-II and RBPF. The BKF-I is based on the MAP estimator of the measurement loss process and the BKF-II is derived by estimating the conditional loss probability. The RBPF is a particle filter based algorithm which marginalizes out the loss process to increase the efficiency of particles. All the proposed filters can be easily implemented in recursive forms. Finally, a linear system, a target tracking system and a quadrotor's path control problem are included to illustrate their effectiveness, and show the tradeoff between computational complexity and estimation accuracy of the proposed filters.

Yiqi Wang, Jiaqi Zhang, Taotao Cai et al.

Large language model (LLM)-based agents increasingly solve complex tasks by interacting with external tools, retrieval systems, memory modules, environments, and other agents. These capabilities expand agent autonomy, but also make agent behavior harder to verify, debug, and audit. Final-answer accuracy alone cannot explain how an output was produced, which evidence supported each claim, whether tool calls were justified, how memory influenced later decisions, or where execution failures originated. Evidence tracing and execution provenance address this gap by modeling how retrieved evidence, tool outputs, memory items, environment observations, intermediate claims, actions, and final answers are connected throughout agent execution. This survey provides a systematic review and conceptual framework for evidence tracing and execution provenance in LLM agents. We organize related work around a unified provenance perspective that connects retrieval grounding, claim support, tool-use safety, memory lineage, observability, debugging, audit, and recovery. We introduce a taxonomy covering trace sources, evidence and execution units, provenance relations, tracing granularity and timing, representation forms, and trust functions. We review key methodological directions, including provenance representation, evidence attribution, tool-use provenance, runtime guardrails, provenance-bearing memory, trace-based observability, and failure diagnosis. We also map existing benchmarks, datasets, and evaluation metrics to provenance-related capabilities, and discuss how evaluation can move from final-answer correctness toward process-level accountability. Finally, we outline open challenges, including unified trace schemas, claim-level and semantic provenance, provenance-aware safety mechanisms, realistic execution-trace benchmarks, recovery-oriented evaluation, and privacy-aware audit infrastructure.

Meituan LongCat Team, Bin Xiao, Chao Wang et al.

The prevailing Next-Token Prediction (NTP) paradigm has driven the success of large language models through discrete autoregressive modeling. However, contemporary multimodal systems remain language-centric, often treating non-linguistic modalities as external attachments, leading to fragmented architectures and suboptimal integration. To transcend this limitation, we introduce Discrete Native Autoregressive (DiNA), a unified framework that represents multimodal information within a shared discrete space, enabling a consistent and principled autoregressive modeling across modalities. A key innovation is the Discrete Native Any-resolution Visual Transformer (dNaViT), which performs tokenization and de-tokenization at arbitrary resolutions, transforming continuous visual signals into hierarchical discrete tokens. Building on this foundation, we develop LongCat-Next, a native multimodal model that processes text, vision, and audio under a single autoregressive objective with minimal modality-specific design. As an industrial-strength foundation model, it excels at seeing, painting, and talking within a single framework, achieving strong performance across a wide range of multimodal benchmarks. In particular, LongCat-Next addresses the long-standing performance ceiling of discrete vision modeling on understanding tasks and provides a unified approach to effectively reconcile the conflict between understanding and generation. As an attempt toward native multimodality, we open-source the LongCat-Next and its tokenizers, hoping to foster further research and development in the community. GitHub: https://github.com/meituan-longcat/LongCat-Next

Jiaqi Zhang, Ji Hou, Qing Sun et al.

Estimating hip muscle forces and joint moments during gait typically relies on musculoskeletal simulation, which is informative but time-consuming and difficult to apply in clinical settings. This study developed a deep learning framework to predict these hip dynamics parameters directly from lower-limb gait kinematics and compared three representative sequence models under a unified protocol. Gait data were collected from 60 healthy adults under three metronome-guided cadence conditions. Ten bilateral lower-limb joint angles were used as inputs, and OpenSim-derived hip muscle forces and hip joint moments were used as reference outputs. Three deep learning models of LSTM, Transformer, and Mamba were trained and evaluated using the same subject-level split, preprocessing pipeline, and metrics. The best model was then directly tested on an external cohort of 9 patients with osteonecrosis of the femoral head (ONFH) without retraining. In the healthy-subject benchmark, Transformer achieved the best subject-level mean performance for both hip muscle force prediction (RMSE = 1.33 N/kg, MAE = 0.57 N/kg, R2 = 0.819) and hip joint moment prediction (RMSE = 0.11 Nm/kg, MAE = 0.07 Nm/kg, R2 = 0.862), with similar advantages across walking cadences. In zero-shot external validation, Transformer retained moderate predictive ability in ONFH for hip muscle force prediction (RMSE = 1.51 N/kg, MAE = 0.70 N/kg, R2 = 0.537) and hip joint moment prediction (RMSE = 0.17 Nm/kg, MAE = 0.12 Nm/kg, R2 = 0.569). These findings support the feasibility of estimating hip dynamics from gait kinematics, identify Transformer as a strong baseline, and highlight the need for broader pathological validation and improved generalization before clinical application.

Jiaqi Zhang, Chandler Squires, Kristjan Greenewald et al.

Causal disentanglement aims to uncover a representation of data using latent variables that are interrelated through a causal model. Such a representation is identifiable if the latent model that explains the data is unique. In this paper, we focus on the scenario where unpaired observational and interventional data are available, with each intervention changing the mechanism of a latent variable. When the causal variables are fully observed, statistically consistent algorithms have been developed to identify the causal model under faithfulness assumptions. We here show that identifiability can still be achieved with unobserved causal variables, given a generalized notion of faithfulness. Our results guarantee that we can recover the latent causal model up to an equivalence class and predict the effect of unseen combinations of interventions, in the limit of infinite data. We implement our causal disentanglement framework by developing an autoencoding variational Bayes algorithm and apply it to the problem of predicting combinatorial perturbation effects in genomics.

Jiaqi Zhang, Louis Cammarata, Chandler Squires et al.

Sequential experimental design to discover interventions that achieve a desired outcome is a key problem in various domains including science, engineering and public policy. When the space of possible interventions is large, making an exhaustive search infeasible, experimental design strategies are needed. In this context, encoding the causal relationships between the variables, and thus the effect of interventions on the system, is critical for identifying desirable interventions more efficiently. Here, we develop a causal active learning strategy to identify interventions that are optimal, as measured by the discrepancy between the post-interventional mean of the distribution and a desired target mean. The approach employs a Bayesian update for the causal model and prioritizes interventions using a carefully designed, causally informed acquisition function. This acquisition function is evaluated in closed form, allowing for fast optimization. The resulting algorithms are theoretically grounded with information-theoretic bounds and provable consistency results for linear causal models with known causal graph. We apply our approach to both synthetic data and single-cell transcriptomic data from Perturb-CITE-seq experiments to identify optimal perturbations that induce a specific cell state transition. The causally informed acquisition function generally outperforms existing criteria allowing for optimal intervention design with fewer but carefully selected samples.

Hui Xiong, Congying Chu, Lingzhong Fan et al.

In recent years, advances in neuroscience and artificial intelligence have paved the way for unprecedented opportunities for understanding the complexity of the brain and its emulation by computational systems. Cutting-edge advancements in neuroscience research have revealed the intricate relationship between brain structure and function, while the success of artificial neural networks highlights the importance of network architecture. Now is the time to bring them together to better unravel how intelligence emerges from the brain's multiscale repositories. In this review, we propose the Digital Twin Brain (DTB) as a transformative platform that bridges the gap between biological and artificial intelligence. It consists of three core elements: the brain structure that is fundamental to the twinning process, bottom-layer models to generate brain functions, and its wide spectrum of applications. Crucially, brain atlases provide a vital constraint, preserving the brain's network organization within the DTB. Furthermore, we highlight open questions that invite joint efforts from interdisciplinary fields and emphasize the far-reaching implications of the DTB. The DTB can offer unprecedented insights into the emergence of intelligence and neurological disorders, which holds tremendous promise for advancing our understanding of both biological and artificial intelligence, and ultimately propelling the development of artificial general intelligence and facilitating precision mental healthcare.

Jinglong Luo, Yehong Zhang, Jiaqi Zhang et al.

Gaussian process regression (GPR) is a non-parametric model that has been used in many real-world applications that involve sensitive personal data (e.g., healthcare, finance, etc.) from multiple data owners. To fully and securely exploit the value of different data sources, this paper proposes a privacy-preserving GPR method based on secret sharing (SS), a secure multi-party computation (SMPC) technique. In contrast to existing studies that protect the data privacy of GPR via homomorphic encryption, differential privacy, or federated learning, our proposed method is more practical and can be used to preserve the data privacy of both the model inputs and outputs for various data-sharing scenarios (e.g., horizontally/vertically-partitioned data). However, it is non-trivial to directly apply SS on the conventional GPR algorithm, as it includes some operations whose accuracy and/or efficiency have not been well-enhanced in the current SMPC protocol. To address this issue, we derive a new SS-based exponentiation operation through the idea of 'confusion-correction' and construct an SS-based matrix inversion algorithm based on Cholesky decomposition. More importantly, we theoretically analyze the communication cost and the security of the proposed SS-based operations. Empirical results show that our proposed method can achieve reasonable accuracy and efficiency under the premise of preserving data privacy.

Xin Mu, Yu Wang, Yehong Zhang et al.

Understanding the life cycle of the machine learning (ML) model is an intriguing area of research (e.g., understanding where the model comes from, how it is trained, and how it is used). This paper focuses on a novel problem within this field, namely Model Provenance (MP), which concerns the relationship between a target model and its pre-training model and aims to determine whether a source model serves as the provenance for a target model. This is an important problem that has significant implications for ensuring the security and intellectual property of machine learning models but has not received much attention in the literature. To fill in this gap, we introduce a novel concept of Model DNA which represents the unique characteristics of a machine learning model. We utilize a data-driven and model-driven representation learning method to encode the model's training data and input-output information as a compact and comprehensive representation (i.e., DNA) of the model. Using this model DNA, we develop an efficient framework for model provenance identification, which enables us to identify whether a source model is a pre-training model of a target model. We conduct evaluations on both computer vision and natural language processing tasks using various models, datasets, and scenarios to demonstrate the effectiveness of our approach in accurately identifying model provenance.

Jiacheng Li, Jianchao Tan, Hongtao Xu et al.

The Muon optimizer has recently offered a promising alternative to AdamW for large language model training, leveraging matrix orthogonalization to produce geometry-aware updates. However, like all first-order methods, Muon can become trapped in sharp local minima. In this work, we present MONA, an optimizer that bridges Muon's orthogonalization framework with curvature-aware acceleration. MONA adds an acceleration term directly into Muon's gradient processing pipeline. This term is calculated from the exponential moving average of gradient differences. We provide a detailed convergence analysis for MONA, showing that the acceleration term enables escape from sharp minima while preserving Muon's spectral-norm regularization. Empirically, MONA achieves better convergence and downstream task performance compared to both Muon and AdamW across three scales of Mixture-of-Experts pretraining, spanning from 1B to 68B parameters, with the largest model trained on 1 trillion tokens. Furthermore, we conduct supervised fine-tuning on the MOE-68B-A3B model and evaluate it on general capability, mathematical reasoning, and code generation benchmarks, where MONA achieves SOTA performance.

Jiaqi Zhang, Joel Jennings, Agrin Hilmkil et al.

Foundation models have brought changes to the landscape of machine learning, demonstrating sparks of human-level intelligence across a diverse array of tasks. However, a gap persists in complex tasks such as causal inference, primarily due to challenges associated with intricate reasoning steps and high numerical precision requirements. In this work, we take a first step towards building causally-aware foundation models for treatment effect estimations. We propose a novel, theoretically justified method called Causal Inference with Attention (CInA), which utilizes multiple unlabeled datasets to perform self-supervised causal learning, and subsequently enables zero-shot causal inference on unseen tasks with new data. This is based on our theoretical results that demonstrate the primal-dual connection between optimal covariate balancing and self-attention, facilitating zero-shot causal inference through the final layer of a trained transformer-type architecture. We demonstrate empirically that CInA effectively generalizes to out-of-distribution datasets and various real-world datasets, matching or even surpassing traditional per-dataset methodologies. These results provide compelling evidence that our method has the potential to serve as a stepping stone for the development of causal foundation models.

Mingyang Yu, Jiaqi Zhang, Haorui Yang et al.

High-dimensional portfolio optimization faces significant computational challenges under complex constraints, with traditional optimization methods struggling to balance convergence speed and global exploration capability. To address this, firstly, we introduce an enhanced Sharpe ratio-based model that incorporates all constraints into the objective function using adaptive penalty terms, transforming the original constrained problem into an unconstrained single-objective formulation. This approach preserves financial interpretability while simplifying algorithmic implementation. To efficiently solve the resulting high-dimensional optimization problem, we propose a Quantum Hybrid Differential Evolution (QHDE) algorithm, which integrates Quantum-inspired probabilistic behavior into the standard DE framework. QHDE employs a Schrodinger-inspired probabilistic mechanism for population evolution, enabling more flexible and diversified solution updates. To further enhance performance, a good point set-chaos reverse learning strategy is adopted to generate a well-dispersed initial population, and a dynamic elite pool combined with Cauchy-Gaussian hybrid perturbations strengthens global exploration and mitigates premature convergence. Experimental validation on CEC benchmarks and real-world portfolios involving 20 to 80 assets demonstrates that QHDE's performance improves by up to 73.4%. It attains faster convergence, higher solution precision, and greater robustness than seven state-of-the-art counterparts, thereby confirming its suitability for complex, high-dimensional portfolio optimization and advancing quantum-inspired evolutionary research in computational finance.

Kirankumar Shiragur, Jiaqi Zhang, Caroline Uhler

Learning causal structures from interventional data is a fundamental problem with broad applications across various fields. While many previous works have focused on recovering the entire causal graph, in practice, there are scenarios where learning only part of the causal graph suffices. This is called $targeted$ causal discovery. In our work, we focus on two such well-motivated problems: subset search and causal matching. We aim to minimize the number of interventions in both cases. Towards this, we introduce the $Meek~separator$, which is a subset of vertices that, when intervened, decomposes the remaining unoriented edges into smaller connected components. We then present an efficient algorithm to find Meek separators that are of small sizes. Such a procedure is helpful in designing various divide-and-conquer-based approaches. In particular, we propose two randomized algorithms that achieve logarithmic approximation for subset search and causal matching, respectively. Our results provide the first known average-case provable guarantees for both problems. We believe that this opens up possibilities to design near-optimal methods for many other targeted causal structure learning problems arising from various applications.

Jiayi Wang, Lichun Zhang, Xiaoqi Zhuang et al.

Video technology is advancing toward Ultra High Definition (UHD) and High Dynamic Range (HDR), which intensifies the need for higher compression efficiency for these high-specification videos. Beyond advances in traditional codecs, neural video codecs (NVCs) have attracted significant research attention and have evolved rapidly over the past few years. The coding artifacts of NVCs often exhibit content-varying and generative characteristics, which differ from those of conventional codecs and are challenging for traditional video quality assessment (VQA) methods to capture. Therefore, VQA metrics are required to generalize across different codecs, content types, and dynamic ranges to better support video codec research and evaluation. In this paper, we propose FDIM, a feature-distance-based generic video quality metric for both traditional and neural video codecs across SDR and HDR formats. FDIM employs a hybrid architecture that integrates deep and hand-crafted features. The deep feature component learns multi-scale representations to capture distortions ranging from structural and textural fidelity degradation to high-level semantic deviations, while the hand-crafted feature component provides stable complementary cues to improve overall generalization. We trained FDIM on a large-scale subjective quality assessment dataset (DCVQA) consisting of over 16k video sequences encoded by traditional block-based hybrid video codecs and end-to-end perceptually optimized neural video codecs. Extensive experiments on ten SDR/HDR VQA datasets containing diverse, previously unseen codecs demonstrate that FDIM achieves strong generalization and high correlation with subjective assessment. The source code for FDIM and the DCVQA validation set will be released at https://github.com/MCL-ZJU/FDIM.

Meituan LongCat Team, Bayan, Bei Li et al.

We introduce LongCat-Flash, a 560-billion-parameter Mixture-of-Experts (MoE) language model designed for both computational efficiency and advanced agentic capabilities. Stemming from the need for scalable efficiency, LongCat-Flash adopts two novel designs: (a) Zero-computation Experts, which enables dynamic computational budget allocation and activates 18.6B-31.3B (27B on average) per token depending on contextual demands, optimizing resource usage. (b) Shortcut-connected MoE, which enlarges the computation-communication overlap window, demonstrating notable gains in inference efficiency and throughput compared to models of a comparable scale. We develop a comprehensive scaling framework for large models that combines hyperparameter transfer, model-growth initialization, a multi-pronged stability suite, and deterministic computation to achieve stable and reproducible training. Notably, leveraging the synergy among scalable architectural design and infrastructure efforts, we complete model training on more than 20 trillion tokens within 30 days, while achieving over 100 tokens per second (TPS) for inference at a cost of \$0.70 per million output tokens. To cultivate LongCat-Flash towards agentic intelligence, we conduct a large-scale pre-training on optimized mixtures, followed by targeted mid- and post-training on reasoning, code, and instructions, with further augmentation from synthetic data and tool use tasks. Comprehensive evaluations demonstrate that, as a non-thinking foundation model, LongCat-Flash delivers highly competitive performance among other leading models, with exceptional strengths in agentic tasks. The model checkpoint of LongCat-Flash is open-sourced to foster community research. LongCat Chat: https://longcat.ai Hugging Face: https://huggingface.co/meituan-longcat GitHub: https://github.com/meituan-longcat

Jiaqi Zhang, Chen Gao, Liyuan Zhang et al.

Recent advances in embodied agents with multimodal perception and reasoning capabilities based on large vision-language models (LVLMs), excel in autonomously interacting either real or cyber worlds, helping people make intelligent decisions in complex environments. However, the current works are normally optimized by golden action trajectories or ideal task-oriented solutions toward a definitive goal. This paradigm considers limited user-oriented factors, which could be the reason for their performance reduction in a wide range of personal assistant applications. To address this, we propose Chain-of-User-Thought (COUT), a novel embodied reasoning paradigm that takes a chain of thought from basic action thinking to explicit and implicit personalized preference thought to incorporate personalized factors into autonomous agent learning. To target COUT, we introduce SmartAgent, an agent framework perceiving cyber environments and reasoning personalized requirements as 1) interacting with GUI to access an item pool, 2) generating users' explicit requirements implied by previous actions, and 3) recommending items to fulfill users' implicit requirements. To demonstrate SmartAgent's capabilities, we also create a brand-new dataset SmartSpot that offers a full-stage personalized action-involved environment. To our best knowledge, our work is the first to formulate the COUT process, serving as a preliminary attempt towards embodied personalized agent learning. Our extensive experiments on SmartSpot illuminate SmartAgent's functionality among a series of embodied and personalized sub-tasks. We will release code and data upon paper notification at https://github.com/tsinghua-fib-lab/SmartAgent.

Chao Wang, Bei Li, Jiaqi Zhang et al.

The success of Large Language Models (LLMs) hinges on the stable training of deep Transformer architectures. A critical design choice is the placement of normalization layers, leading to a fundamental trade-off: the ``PreNorm'' architecture ensures training stability at the cost of potential performance degradation in deep models, while the ``PostNorm'' architecture offers strong performance but suffers from severe training instability. In this work, we propose SpanNorm, a novel technique designed to resolve this dilemma by integrating the strengths of both paradigms. Structurally, SpanNorm establishes a clean residual connection that spans the entire transformer block to stabilize signal propagation, while employing a PostNorm-style computation that normalizes the aggregated output to enhance model performance. We provide a theoretical analysis demonstrating that SpanNorm, combined with a principled scaling strategy, maintains bounded signal variance throughout the network, preventing the gradient issues that plague PostNorm models, and also alleviating the representation collapse of PreNorm. Empirically, SpanNorm consistently outperforms standard normalization schemes in both dense and Mixture-of-Experts (MoE) scenarios, paving the way for more powerful and stable Transformer architectures.

Haotian Zhang, Feiyue Long, Yixin Yu et al.

Multi-view image compression (MIC) aims to achieve high compression efficiency by exploiting inter-image correlations, playing a crucial role in 3D applications. As a subfield of MIC, distributed multi-view image compression (DMIC) offers performance comparable to MIC while eliminating the need for inter-view information at the encoder side. However, existing methods in DMIC typically treat all images equally, overlooking the varying degrees of correlation between different views during decoding, which leads to suboptimal coding performance. To address this limitation, we propose a novel $\textbf{OmniParallax Attention Mechanism}$ (OPAM), which is a general mechanism for explicitly modeling correlations and aligned features between arbitrary pairs of information sources. Building upon OPAM, we propose a Parallax Multi Information Fusion Module (PMIFM) to adaptively integrate information from different sources. PMIFM is incorporated into both the joint decoder and the entropy model to construct our end-to-end DMIC framework, $\textbf{ParaHydra}$. Extensive experiments demonstrate that $\textbf{ParaHydra}$ is $\textbf{the first DMIC method}$ to significantly surpass state-of-the-art MIC codecs, while maintaining low computational overhead. Performance gains become more pronounced as the number of input views increases. Compared with LDMIC, $\textbf{ParaHydra}$ achieves bitrate savings of $\textbf{19.72%}$ on WildTrack(3) and up to $\textbf{24.18%}$ on WildTrack(6), while significantly improving coding efficiency (as much as $\textbf{65}\times$ in decoding and $\textbf{34}\times$ in encoding).

Yuxiang Guo, Lu Yin, Bo Jiang et al.

Aligning large language models (LLMs) with human intent is critical for enhancing their performance across a variety of tasks. Standard alignment techniques, such as Direct Preference Optimization (DPO), often rely on the binary Bradley-Terry (BT) model, which can struggle to capture the complexities of human preferences -- particularly in the presence of noisy or inconsistent labels and frequent ties. To address these limitations, we introduce the Tie-rank Oriented Bradley-Terry model (TOBT), an extension of the BT model that explicitly incorporates ties, enabling more nuanced preference representation. Building on this, we propose Tie-rank Oriented Direct Preference Optimization (TODO), a novel alignment algorithm that leverages TOBT's ternary ranking system to improve preference alignment. In evaluations on Mistral-7B and Llama 3-8B models, TODO consistently outperforms DPO in modeling preferences across both in-distribution and out-of-distribution datasets. Additional assessments using MT Bench and benchmarks such as Piqa, ARC-c, and MMLU further demonstrate TODO's superior alignment performance. Notably, TODO also shows strong results in binary preference alignment, highlighting its versatility and potential for broader integration into LLM alignment. The implementation details can be found in https://github.com/XXares/TODO.

Mingyu Cao, Gen Li, Jie Ji et al.

Mixture-of-Experts (MoE) has garnered significant attention for its ability to scale up neural networks while utilizing the same or even fewer active parameters. However, MoE does not alleviate the massive memory requirements of networks, which limits their practicality in real-world applications, especially in the era of large language models (LLMs). While recent work explores the possibility of removing entire layers of MoE to reduce memory, the performance degradation is still notable. In this paper, we propose ConDense-MoE (CD-MoE), which, instead of dropping the entire MoE layer, condenses the large, sparse MoE layer into a smaller, denser layer with only a few experts activated for all tokens, while maintaining hardware friendliness. Our approach is specifically designed for fine-grained MoE with shared experts, where Feed-Forward Networks are split into many small experts, with certain experts isolated to serve as shared experts that are always activated, such as DeepSeekMoE and QwenMoE. We demonstrate the effectiveness of our method. Specifically, for the DeepSeekMoE-16B model, our approach maintains 90% of the average accuracy while reducing memory usage by 27.5% and increasing inference speed by 1.26 times. Moreover, we show that by applying lightweight expert fine-tuning -- only to the condensed layers -- and using 5 hours on a single 80G A100 GPU, we can successfully recover 98% of the original performance. Our code is available at: https://github.com/duterscmy/CD-MoE/tree/main.

Pingwei Sun, Yuxuan Hu, Jianchao Tan et al.

Linear attention mechanisms have emerged as promising alternatives to softmax attention, offering linear-time complexity during inference. Recent advances such as Gated DeltaNet (GDN) and Kimi Delta Attention (KDA) have demonstrated that the delta rule, an online gradient descent update, enables superior associative recall compared to simple additive updates. While KDA refined the coarse head-wise decay gate into channel-wise decay, the learning rate $β_t$ in the delta update remains a scalar, limiting the model's capacity for dimension-specific adaptation. We introduce FG$^2$-GDN, which replaces the scalar $β_t$ with a channel-wise vector analogous to the transition from SGD to per-coordinate adaptive optimizers such as AdaGrad and Adam. We further propose FG$^2$-GDN+, which decouples the scaling for keys and values, enabling independent control of erasure strength and write strength. Experiments on synthetic and real-world benchmarks show that FG$^2$-GDN and its variant improve associative recall and long-context understanding over GDN and KDA, with comparable computational efficiency.

Jiaqi Zhang, Zhuodong Liu, Kejian Yu

Accurate identification of agricultural pests is essential for crop protection but remains challenging due to the large intra-class variance and fine-grained differences among pest species. While deep learning has advanced pest detection, most existing approaches rely solely on low-level visual features and lack effective multi-modal integration, leading to limited accuracy and poor interpretability. Moreover, the scarcity of high-quality multi-modal agricultural datasets further restricts progress in this field. To address these issues, we construct two novel multi-modal benchmarks-CTIP102 and STIP102-based on the widely-used IP102 dataset, and introduce a Multi-scale Cross-Modal Fusion Network (MSFNet-CPD) for robust pest detection. Our approach enhances visual quality via a super-resolution reconstruction module, and feeds both the original and reconstructed images into the network to improve clarity and detection performance. To better exploit semantic cues, we propose an Image-Text Fusion (ITF) module for joint modeling of visual and textual features, and an Image-Text Converter (ITC) that reconstructs fine-grained details across multiple scales to handle challenging backgrounds. Furthermore, we introduce an Arbitrary Combination Image Enhancement (ACIE) strategy to generate a more complex and diverse pest detection dataset, MTIP102, improving the model's generalization to real-world scenarios. Extensive experiments demonstrate that MSFNet-CPD consistently outperforms state-of-the-art methods on multiple pest detection benchmarks. All code and datasets will be made publicly available at: https://github.com/Healer-ML/MSFNet-CPD.

Hong Liu, Jiaqi Zhang, Chao Wang et al.

While Mixture-of-Experts (MoE) architectures have become the standard for sparsity scaling in large language models, they increasingly face diminishing returns and system-level bottlenecks. In this work, we explore embedding scaling as a potent, orthogonal dimension for scaling sparsity. Through a comprehensive analysis and experiments, we identify specific regimes where embedding scaling achieves a superior Pareto frontier compared to expert scaling. We systematically characterize the critical architectural factors governing this efficacy -- ranging from parameter budgeting to the interplay with model width and depth. Moreover, by integrating tailored system optimizations and speculative decoding, we effectively convert this sparsity into tangible inference speedups. Guided by these insights, we introduce LongCat-Flash-Lite, a 68.5B parameter model with ~3B activated trained from scratch. Despite allocating over 30B parameters to embeddings, LongCat-Flash-Lite not only surpasses parameter-equivalent MoE baselines but also exhibits exceptional competitiveness against existing models of comparable scale, particularly in agentic and coding domains.

Meituan LongCat Team, Anchun Gui, Bei Li et al.

We present LongCat-Flash-Thinking, an efficient 560-billion-parameter open-source Mixture-of-Experts (MoE) reasoning model. Its advanced capabilities are cultivated through a meticulously crafted training process, beginning with long Chain-of-Thought (CoT) data cold-start and culminating in large-scale Reinforcement Learning (RL). We first employ a well-designed cold-start training strategy, which significantly enhances the reasoning potential and equips the model with specialized skills in both formal and agentic reasoning. Then, a core innovation is our domain-parallel training scheme, which decouples optimization across distinct domains (e.g., STEM, Code, Agentic) and subsequently fuses the resulting expert models into a single, nearly Pareto-optimal model. This entire process is powered by our Dynamic ORchestration for Asynchronous rollout (DORA) system, a large-scale RL framework that delivers a greater than threefold training speedup over synchronous methods on tens of thousands of accelerators. As a result, LongCat-Flash-Thinking achieves state-of-the-art performance among open-source models on a suite of complex reasoning tasks. The model exhibits exceptional efficiency in agentic reasoning, reducing average token consumption by 64.5% (from 19, 653 to 6, 965) on AIME-25, without degrading task accuracy. We release LongCat-Flash-Thinking to promote further advances in reasoning systems and agentic AI research.

Wenquan Lu, Jiaqi Zhang, Hugues Van Assel et al.

Self-Supervised Learning (SSL) has become a powerful solution to extract rich representations from unlabeled data. Yet, SSL research is mostly focused on clean, curated and high-quality datasets. As a result, applying SSL on noisy data remains a challenge, despite being crucial to applications such as astrophysics, medical imaging, geophysics or finance. In this work, we present a fully self-supervised framework that enables noise-robust representation learning without requiring a denoiser at inference or downstream fine-tuning. Our method first trains an SSL denoiser on noisy data, then uses it to construct a denoised-to-noisy data curriculum (i.e., training first on denoised, then noisy samples) for pretraining a SSL backbone (e.g., DINOv2), combined with a teacher-guided regularization that anchors noisy embeddings to their denoised counterparts. This process encourages the model to internalize noise robustness. Notably, the denoiser can be discarded after pretraining, simplifying deployment. On ImageNet-1k with ViT-B under extreme Gaussian noise ($σ=255$, SNR = 0.72 dB), our method improves linear probing accuracy by 4.8% over DINOv2, demonstrating that denoiser-free robustness can emerge from noise-aware pretraining. The code is available at https://github.com/wenquanlu/noisy_dinov2.

Marc Franquesa Monés, Jiaqi Zhang, Caroline Uhler

Learning causal relations from observational data is a fundamental problem with wide-ranging applications across many fields. Constraint-based methods infer the underlying causal structure by performing conditional independence tests. However, existing algorithms such as the prominent PC algorithm need to perform a large number of independence tests, which in the worst case is exponential in the maximum degree of the causal graph. Despite extensive research, it remains unclear if there exist algorithms with better complexity without additional assumptions. Here, we establish an algorithm that achieves a better complexity of $p^{\mathcal{O}(s)}$ tests, where $p$ is the number of nodes in the graph and $s$ denotes the maximum undirected clique size of the underlying essential graph. Complementing this result, we prove that any constraint-based algorithm must perform at least $2^{Ω(s)}$ conditional independence tests, establishing that our proposed algorithm achieves exponent-optimality up to a logarithmic factor in terms of the number of conditional independence tests needed. Finally, we validate our theoretical findings through simulations, on semi-synthetic gene-expression data, and real-world data, demonstrating the efficiency of our algorithm compared to existing methods in terms of number of conditional independence tests needed.

Tekin Gunasar, Alexandra Rekesh, Atul Nair et al.

Electromyography signals can be used as training data by machine learning models to classify various gestures. We seek to produce a model that can classify six different hand gestures with a limited number of samples that generalizes well to a wider audience while comparing the effect of our feature extraction results on model accuracy to other more conventional methods such as the use of AR parameters on a sliding window across the channels of a signal. We appeal to a set of more elementary methods such as the use of random bounds on a signal, but desire to show the power these methods can carry in an online setting where EMG classification is being conducted, as opposed to more complicated methods such as the use of the Fourier Transform. To augment our limited training data, we used a standard technique, known as jitter, where random noise is added to each observation in a channel wise manner. Once all datasets were produced using the above methods, we performed a grid search with Random Forest and XGBoost to ultimately create a high accuracy model. For human computer interface purposes, high accuracy classification of EMG signals is of particular importance to their functioning and given the difficulty and cost of amassing any sort of biomedical data in a high volume, it is valuable to have techniques that can work with a low amount of high-quality samples with less expensive feature extraction methods that can reliably be carried out in an online application.

Jiaqi Zhang, Juntuo Wang, Zhixin Sun et al.

Large-scale vision foundation models such as DINOv2 boast impressive performances by leveraging massive architectures and training datasets. But numerous scenarios require practitioners to reproduce those pre-training solutions, such as on private data, new modalities, or simply for scientific questioning--which is currently extremely demanding computation-wise. We thus propose a novel pre-training strategy for DINOv2 that simultaneously accelerates convergence--and strengthens robustness to common corruptions as a by-product. Our approach involves a frequency filtering curriculum--low-frequency being seen first--and the Gaussian noise patching augmentation. Applied to a ViT-B/16 backbone trained on ImageNet-1K, while pre-training time and FLOPs are reduced by 1.6x and 2.25x, our method still achieves matching robustness in corruption benchmarks (ImageNet-C) and maintains competitive linear probing performance compared with baseline. This dual benefit of efficiency and robustness makes large-scale self-supervised foundation modeling more attainable, while opening the door to novel exploration around data curriculum and augmentation as means to improve self-supervised learning models robustness. The code is available at https://github.com/KevinZ0217/fast_dinov2

Jiewen Deng, Renhe Jiang, Jiaqi Zhang et al.

Multi-modality spatio-temporal (MoST) data extends spatio-temporal (ST) data by incorporating multiple modalities, which is prevalent in monitoring systems, encompassing diverse traffic demands and air quality assessments. Despite significant strides in ST modeling in recent years, there remains a need to emphasize harnessing the potential of information from different modalities. Robust MoST forecasting is more challenging because it possesses (i) high-dimensional and complex internal structures and (ii) dynamic heterogeneity caused by temporal, spatial, and modality variations. In this study, we propose a novel MoST learning framework via Self-Supervised Learning, namely MoSSL, which aims to uncover latent patterns from temporal, spatial, and modality perspectives while quantifying dynamic heterogeneity. Experiment results on two real-world MoST datasets verify the superiority of our approach compared with the state-of-the-art baselines. Model implementation is available at https://github.com/beginner-sketch/MoSSL.

Zhenyi Liu, Minghao Shen, Jiaqi Zhang et al.

We describe an open-source simulator that creates sensor irradiance and sensor images of typical automotive scenes in urban settings. The purpose of the system is to support camera design and testing for automotive applications. The user can specify scene parameters (e.g., scene type, road type, traffic density, time of day) to assemble a large number of random scenes from graphics assets stored in a database. The sensor irradiance is generated using quantitative computer graphics methods, and the sensor images are created using image systems sensor simulation. The synthetic sensor images have pixel level annotations; hence, they can be used to train and evaluate neural networks for imaging tasks, such as object detection and classification. The end-to-end simulation system supports quantitative assessment, from scene to camera to network accuracy, for automotive applications.

Yuxuan Hu, Jianchao Tan, Jiaqi Zhang et al.

In this work, we conduct a systematic analysis of Native Sparse Attention (NSA) and propose targeted improvements that enhance long-context modeling. A key insight is that alternating between local (sliding-window) and global (compression, selective) attention across layers, rather than using fixed patterns, enables more effective propagation of long-range dependencies and substantially boosts performance on long-sequence tasks. Meanwhile, we further refine NSA's branches with Latent Attention that the sliding-window branch is enhanced with Multi-head Latent Attention (MLA) while compression and selective branches adopt Group-head Latent Attention (GLA). These changes reduce KV-cache memory by 50\% versus NSA while improving the model's common-sense reasoning and long-text understanding capabilities. Experiments on models from 340M to 1.3B parameters (trained on 15B and 100B tokens) show our method matches or exceeds full attention and native sparse attention in both common-sense reasoning and long-context understanding tasks.

Shiyi Chen, Mingye Yang, Haiyan Mao et al.

Efficiently training quadruped robot navigation in densely cluttered environments remains a significant challenge. Existing methods are either limited by a lack of safety and agility in simple obstacle distributions or suffer from slow locomotion in complex environments, often requiring excessively long training phases. To this end, we propose SEA-Nav (Safe, Efficient, and Agile Navigation), a reinforcement learning framework for quadruped navigation. Within diverse and dense obstacle environments, a differentiable control barrier function (CBF)-based shield constraints the navigation policy to output safe velocity commands. An adaptive collision replay mechanism and hazardous exploration rewards are introduced to increase the probability of learning from critical experiences, guiding efficient exploration and exploitation. Finally, kinematic action constraints are incorporated to ensure safe velocity commands, facilitating successful physical deployment. To the best of our knowledge, this is the first approach that achieves highly challenging quadruped navigation in the real world with minute-level training time.

Caroline Uhler, Jiaqi Zhang

Massive data collection holds the promise of a better understanding of complex phenomena and, ultimately, better decisions. Representation learning has become a key driver of deep learning applications, as it allows learning latent spaces that capture important properties of the data without requiring any supervised annotations. Although representation learning has been hugely successful in predictive tasks, it can fail miserably in causal tasks including predicting the effect of a perturbation/intervention. This calls for a marriage between representation learning and causal inference. An exciting opportunity in this regard stems from the growing availability of multi-modal data (observational and perturbational, imaging-based and sequencing-based, at the single-cell level, tissue-level, and organism-level). We outline a statistical and computational framework for causal structure and representation learning motivated by fundamental biomedical questions: how to effectively use observational and perturbational data to perform causal discovery on observed causal variables; how to use multi-modal views of the system to learn causal variables; and how to design optimal perturbations.

Haotian Zhang, Xu Mo, Yixin Yu et al.

Free-Viewpoint Video (FVV) has emerged as a cornerstone of next-generation immersive media systems and attracted widespread attention. Previous methods primarily focus on short video sequences and suffer from significant performance degradation when processing long-horizon free-viewpoint video (LFVV). Motivated by bit allocation theory, we analyze dynamic-anchor-based volumetric video representation within a rate-distortion optimization framework and propose \textbf{SoLAR}, which is the first error-resilient streamable FVV framework that maintains stable reconstruction quality on long sequences without requiring group-of-pictures partitioning. We propose the Anchor Activation Dynamics (AAD), which enables dynamic anchors to model non-rigid transformations by dynamically activating informative anchors and suppressing redundant ones. Furthermore, we introduce Latent Discrepancy Aware Recalibration (LaDAR), which is a mechanism to identify discrepancies between latent representations and recalibrate the correspondences encoded in the network, effectively mitigating error propagation in LFVV without compromising real-time performance or storage compactness. Extensive experiments demonstrate that \textbf{SoLAR} achieves state-of-the-art reconstruction performance while maintaining minimum storage overhead, which provides a new direction for LFVV reconstruction and advances the practical deployment of immersive systems. Demo free-viewpoint videos are provided in the supplementary material.

Zecheng Tang, Jiaye Fu, Qiankun Gao et al.

Feed-forward 3D reconstruction models based on Vision Transformers can directly estimate scene geometry and camera poses from a small set of input images, but scaling them to video inputs with hundreds or thousands of frames remains challenging due to the quadratic cost of global attention layers. Recent token-merging methods accelerate these models by compressing the token sequence within the global attention layers, but they apply a uniform reduction to query tokens and key-value tokens, ignoring their functionally distinct roles in 3D reconstruction. In this work, we identify a key property of feed-forward 3D reconstruction models: query tokens encode view-specific geometric requests and are sensitive to compression, while key-value tokens represent shared scene context and tolerate aggressive compression. Guided by this insight, we propose Spark3R, a training-free acceleration framework that decouples the compression of query tokens and key-value tokens by assigning distinct reduction factors, with intra-group token merging applied to query tokens and lightweight token pruning to key-value tokens. Additionally, Spark3R adaptively adjusts the key-value reduction factor across layers, further improving the quality-efficiency trade-off. As a plug-and-play framework requiring no retraining, Spark3R integrates directly into multiple pretrained feed-forward 3D reconstruction models, including VGGT, $π^3$, and Depth-Anything-3, and achieves up to $28\times$ speedup on 1,000-frame inputs while maintaining competitive reconstruction quality.

Yuxuan Hu, Jianchao Tan, Jiaqi Zhang et al.

Long-context inference in LLMs faces the dual challenges of quadratic attention complexity and prohibitive KV cache memory. While token-level sparse attention offers superior accuracy, its indexing overhead is costly; block-level methods improve efficiency but sacrifice precision. We propose AsyncTLS, a hierarchical sparse attention system that combines coarse-grained block filtering with fine-grained token selection to balance accuracy and efficiency, coupled with an asynchronous offloading engine that overlaps KV cache transfers with computation via temporal locality exploitation. Evaluated on Qwen3 and GLM-4.7-Flash across GQA, and MLA architectures, AsyncTLS achieves accuracy comparable to full attention while delivering 1.2x - 10.0x operator speedups and 1.3x - 4.7x end-to-end throughput improvements on 48k - 96k contexts.

Dakuan Lu, Jiaqi Zhang, Cheng Yuan et al.

Recent advances in large language models (LLMs) have been largely driven by scaling laws for individual models, which predict performance improvements as model parameters and data volume increase. However, the capabilities of any single LLM are inherently bounded. One solution originates from intricate interactions among multiple LLMs, rendering their collective performance surpasses that of any constituent model. Despite the rapid proliferation of multi-model integration techniques such as model routing and post-hoc ensembling, a unifying theoretical framework of performance scaling for multi-model collaboration remains absent. In this work, we propose the Law of Multi-model Collaboration, a scaling law that predicts the performance limits of LLM ensembles based on their aggregated parameter budget. To quantify the intrinsic upper bound of multi-model collaboration, we adopt a method-agnostic formulation and assume an idealized integration oracle where the total cross-entropy loss of each sample is determined by the minimum loss of any model in the model pool. Experimental results reveal that multi-model systems follow a power-law scaling with respect to the total parameter count, exhibiting a more significant improvement trend and a lower theoretical loss floor compared to single model scaling. Moreover, ensembles of heterogeneous model families achieve better performance scaling than those formed within a single model family, indicating that model diversity is a primary driver of collaboration gains. These findings suggest that model collaboration represents a critical axis for extending the intelligence frontier of LLMs.

Shiyi Chen, Haiyi Liu, Mingye Yang et al.

Designing an open-world quadrupedal loco-manipulation system is highly challenging. Traditional reinforcement learning frameworks utilizing exteroception often suffer from extreme sample inefficiency and massive sim-to-real gaps. Furthermore, the inherent latency of visual tracking fundamentally conflicts with the high-frequency demands of precise floating-base control. Consequently, existing systems lean heavily on expensive external motion capture and off-board computation. To eliminate these dependencies, we present SigLoMa, a fully onboard, ego-centric vision-based pick-and-place framework. At the core of SigLoMa is the introduction of Sigma Points, a lightweight geometric representation for exteroception that guarantees high scalability and native sim-to-real alignment. To bridge the frequency divide between slow perception and fast control, we design an ego-centric Kalman Filter to provide robust, high-rate state estimation. On the learning front, we alleviate sample inefficiency via an Active Sampling Curriculum guided by Hint Poses, and tackle the robot's structural visual blind spots using temporal encoding coupled with simulated random-walk drift. Real-world experiments validate that, relying solely on a 5Hz (200 ms latency) open-vocabulary detector, SigLoMa successfully executes dynamic loco-manipulation across multiple tasks, achieving performance comparable to expert human teleoperation.

Jinglong Luo, Yehong Zhang, Zhuo Zhang et al.

With the growing use of Transformer models hosted on cloud platforms to offer inference services, privacy concerns are escalating, especially concerning sensitive data like investment plans and bank account details. Secure Multi-Party Computing (SMPC) emerges as a promising solution to protect the privacy of inference data and model parameters. However, the application of SMPC in Privacy-Preserving Inference (PPI) for Transformer models often leads to considerable slowdowns or declines in performance. This is largely due to the multitude of nonlinear operations in the Transformer architecture, which are not well-suited to SMPC and difficult to circumvent or optimize effectively. To address this concern, we introduce a comprehensive PPI framework called SecFormer to achieve fast and accurate PPI for Transformer models. We successfully eliminate the high-cost exponential and maximum operations in PPI without sacrificing model performance and develop a suite of efficient SMPC protocols by employing suitable numerical computation methods to boost other complex nonlinear functions in PPI, including GeLU, LayerNorm, and a redesigned Softmax. Our extensive experiments reveal that SecFormer outperforms MPCFormer in performance, showing improvements of $3.4\%$ and $24.7\%$ for BERT$_{\text{BASE}}$ and BERT$_{\text{LARGE}}$, respectively. In terms of efficiency, SecFormer is 3.57 and 3.58 times faster than PUMA for BERT$_{\text{BASE}}$ and BERT$_{\text{LARGE}}$, demonstrating its effectiveness and speed.

Xinchen Luo, Jiangxia Cao, Tianyu Sun et al.

In recent years, with the significant evolution of multi-modal large models, many recommender researchers realized the potential of multi-modal information for user interest modeling. In industry, a wide-used modeling architecture is a cascading paradigm: (1) first pre-training a multi-modal model to provide omnipotent representations for downstream services; (2) The downstream recommendation model takes the multi-modal representation as additional input to fit real user-item behaviours. Although such paradigm achieves remarkable improvements, however, there still exist two problems that limit model performance: (1) Representation Unmatching: The pre-trained multi-modal model is always supervised by the classic NLP/CV tasks, while the recommendation models are supervised by real user-item interaction. As a result, the two fundamentally different tasks' goals were relatively separate, and there was a lack of consistent objective on their representations; (2) Representation Unlearning: The generated multi-modal representations are always stored in cache store and serve as extra fixed input of recommendation model, thus could not be updated by recommendation model gradient, further unfriendly for downstream training. Inspired by the two difficulties challenges in downstream tasks usage, we introduce a quantitative multi-modal framework to customize the specialized and trainable multi-modal information for different downstream models.

Xupeng Chen, Binbin Shi, Chenqian Le et al.

Medical retrieval-augmented generation (RAG) systems typically operate on text chunks extracted from biomedical literature, discarding the rich visual content (tables, figures, structured layouts) of original document pages. We propose MED-VRAG, an iterative multimodal RAG framework that retrieves and reasons over PMC document page images instead of OCR'd text. The system pairs ColQwen2.5 patch-level page embeddings with a sharded MapReduce LLM filter, scaling to ~350K pages while keeping Stage-1 retrieval under 30 ms via an offline coarse-to-fine index (C=8 centroids per page, ANN over centroids, exact two-way scoring on the top-R shortlist). A vision-language model (VLM) then iteratively refines its query and accumulates evidence in a memory bank across up to 3 reasoning rounds, with a single iteration costing ~15.9 s and the full three-round pipeline ~47.8 s on 4xA100. Across four medical QA benchmarks (MedQA, MedMCQA, PubMedQA, MMLU-Med), MEDVRAG reaches 78.6% average accuracy. Under controlled comparison with the same Qwen2.5-VL-32B backbone, retrieval contributes a +5.8 point gain over the no-retrieval baseline; we also note a +1.8 point edge over MedRAG + GPT-4 (76.8%), with the caveat that this is a cross-paper rather than head-to-head comparison. Ablations isolate +1.0 from page-image vs text-chunk retrieval, +1.5 from iteration, and +1.0 from the memory bank.

Ryan Welch, Jiaqi Zhang, Caroline Uhler

Causal disentanglement aims to learn about latent causal factors behind data, holding the promise to augment existing representation learning methods in terms of interpretability and extrapolation. Recent advances establish identifiability results assuming that interventions on (single) latent factors are available; however, it remains debatable whether such assumptions are reasonable due to the inherent nature of intervening on latent variables. Accordingly, we reconsider the fundamentals and ask what can be learned using just observational data. We provide a precise characterization of latent factors that can be identified in nonlinear causal models with additive Gaussian noise and linear mixing, without any interventions or graphical restrictions. In particular, we show that the causal variables can be identified up to a layer-wise transformation and that further disentanglement is not possible. We transform these theoretical results into a practical algorithm consisting of solving a quadratic program over the score estimation of the observed data. We provide simulation results to support our theoretical guarantees and demonstrate that our algorithm can derive meaningful causal representations from purely observational data.

Ziyao Zhou, Tiancheng Cao, Chen Shen et al.

Accurate time synchronization is essential for Internet of Things (IoT) systems, where multiple distributed nodes must share a common time base for coordinated sensing and data fusion. However, conventional synchronization approaches suffer from nondeterministic transmission latency, limited precision, or restricted bidirectional functionality. This paper presents a protocol-independent wireless timer synchronization method that exploits radio timeslots to transmit precisely timestamped beacons in a proprietary radio mode. By decoupling synchronization from upper-layer packet retransmissions and leveraging hardware-timed radio events, the proposed approach significantly reduces scheduling uncertainty and achieves nanosecond-level synchronization accuracy. Comprehensive experiments evaluate the impacts of synchronization frequency, RSSI, BLE connection interval, and throughput on synchronization performance. The results demonstrate that an optimal synchronization frequency of 1000 Hz yields an approximately 20 ns delay in the absence of communication stack activity while maintaining sub-500 ns accuracy under most realistic BLE traffic conditions. Furthermore, larger connection intervals, lower application throughput, and higher RSSI consistently improve synchronization quality by reducing radio resource contention and packet loss. The proposed scheme provides a general and high-precision synchronization solution suitable for resource-constrained IoT systems.

Xiaohan Lu, Jiaye Fu, Jiaqi Zhang et al.

Feed-forward 3D Gaussian Splatting (3DGS) has recently demonstrated promising results for novel view synthesis (NVS) from sparse input views, particularly under narrow-baseline conditions. However, its performance significantly degrades in wide-baseline scenarios due to limited texture details and geometric inconsistencies across views. To address these challenges, in this paper, we propose ProSplat, a two-stage feed-forward framework designed for high-fidelity rendering under wide-baseline conditions. The first stage involves generating 3D Gaussian primitives via a 3DGS generator. In the second stage, rendered views from these primitives are enhanced through an improvement model. Specifically, this improvement model is based on a one-step diffusion model, further optimized by our proposed Maximum Overlap Reference view Injection (MORI) and Distance-Weighted Epipolar Attention (DWEA). MORI supplements missing texture and color by strategically selecting a reference view with maximum viewpoint overlap, while DWEA enforces geometric consistency using epipolar constraints. Additionally, we introduce a divide-and-conquer training strategy that aligns data distributions between the two stages through joint optimization. We evaluate ProSplat on the RealEstate10K and DL3DV-10K datasets under wide-baseline settings. Experimental results demonstrate that ProSplat achieves an average improvement of 1 dB in PSNR compared to recent SOTA methods.

Wentao Song, He Huang, Youqiang Sun et al.

Advanced plant phenotyping technologies play a crucial role in targeted trait improvement and accelerating intelligent breeding. Due to the species diversity of plants, existing methods heavily rely on large-scale high-precision manually annotated data. For self-occluded objects at the grain level, unsupervised methods often prove ineffective. This study proposes IPENS, an interactive unsupervised multi-target point cloud extraction method. The method utilizes radiance field information to lift 2D masks, which are segmented by SAM2 (Segment Anything Model 2), into 3D space for target point cloud extraction. A multi-target collaborative optimization strategy is designed to effectively resolve the single-interaction multi-target segmentation challenge. Experimental validation demonstrates that IPENS achieves a grain-level segmentation accuracy (mIoU) of 63.72% on a rice dataset, with strong phenotypic estimation capabilities: grain volume prediction yields R2 = 0.7697 (RMSE = 0.0025), leaf surface area R2 = 0.84 (RMSE = 18.93), and leaf length and width predictions achieve R2 = 0.97 and 0.87 (RMSE = 1.49 and 0.21). On a wheat dataset,IPENS further improves segmentation accuracy to 89.68% (mIoU), with equally outstanding phenotypic estimation performance: spike volume prediction achieves R2 = 0.9956 (RMSE = 0.0055), leaf surface area R2 = 1.00 (RMSE = 0.67), and leaf length and width predictions reach R2 = 0.99 and 0.92 (RMSE = 0.23 and 0.15). This method provides a non-invasive, high-quality phenotyping extraction solution for rice and wheat. Without requiring annotated data, it rapidly extracts grain-level point clouds within 3 minutes through simple single-round interactions on images for multiple targets, demonstrating significant potential to accelerate intelligent breeding efficiency.

Bijan Mazaheri, Jiaqi Zhang, Caroline Uhler

Constraint-based causal discovery algorithms utilize many statistical tests for conditional independence to uncover networks of causal dependencies. These approaches to causal discovery rely on an assumed correspondence between the graphical properties of a causal structure and the conditional independence properties of observed variables, known as the causal Markov condition and faithfulness. Finite data yields an empirical distribution that is "close" to the actual distribution. Across these many possible empirical distributions, the correspondence to the graphical properties can break down for different conditional independencies, and multiple violations can occur at the same time. We study this "meta-dependence" between conditional independence properties using the following geometric intuition: each conditional independence property constrains the space of possible joint distributions to a manifold. The "meta-dependence" between conditional independences is informed by the position of these manifolds relative to the true probability distribution. We provide a simple-to-compute measure of this meta-dependence using information projections and consolidate our findings empirically using both synthetic and real-world data.

Jinbo Wang, Binghui Li, Zhanpeng Zhou et al.

Batch size scheduling (BSS) plays a critical role in large-scale deep learning training, influencing both optimization dynamics and computational efficiency. Yet, its theoretical foundations remain poorly understood. In this work, we show that the functional scaling law (FSL) framework introduced in Li et al. (2025a) provides a principled lens for analyzing BSS. Specifically, we characterize the optimal BSS under a fixed data budget and show that its structure depends sharply on task difficulty. For easy tasks, optimal schedules keep increasing batch size throughout. In contrast, for hard tasks, the optimal schedule maintains small batch sizes for most of training and switches to large batches only in a late stage. To explain the emergence of late switching, we uncover a dynamical mechanism -- the fast catch-up effect -- which also manifests in large language model (LLM) pretraining. After switching from small to large batches, the loss rapidly aligns with the constant large-batch trajectory. Using FSL, we show that this effect stems from rapid forgetting of accumulated gradient noise, with the catch-up speed determined by task difficulty. Crucially, this effect implies that large batches can be safely deferred to late training without sacrificing performance, while substantially reducing data consumption. Finally, extensive LLM pretraining experiments -- covering both Dense and MoE architectures with up to 1.1B parameters and 1T tokens -- validate our theoretical predictions. Across all settings, late-switch schedules consistently outperform constant-batch and early-switch baselines.

Xiao Li, Jiaqi Zhang, Shuxiang Zhang et al.

We propose an Authentic Discrete Diffusion (ADD) framework that fundamentally redefines prior pseudo-discrete approaches by preserving core diffusion characteristics directly in the one-hot space through a suite of coordinated mechanisms. Unlike conventional "pseudo" discrete diffusion (PDD) methods, ADD reformulates the diffusion input by directly using float-encoded one-hot class data, without relying on diffusing in the continuous latent spaces or masking policies. At its core, a timestep-conditioned cross-entropy loss is introduced between the diffusion model's outputs and the original one-hot labels. This synergistic design establishes a bridge between discriminative and generative learning. Our experiments demonstrate that ADD not only achieves superior performance on classification tasks compared to the baseline, but also exhibits excellent text generation capabilities on Image captioning. Extensive ablations validate the measurable gains of each component.