arXiv subjects

Yuxin Wu

Publications and source records attributed to Yuxin Wu.

At least 19 recordsLinked to original sources

arXiv2026

MetaRSI / RSI2: A Meta-Recursive Self-Improving System for Recursive Self-Improving Systems Themselves

Recursive self-improvement (RSI) lets a system improve the model-building machinery from its own failures, so every later model inherits the gain. Yet RSI has been validated almost exclusively on coding and formal benchmarks such as science QA and mathematics. This format bound limits RSI to improvement within a machine-checkable slice, not general capability where questions are open and correctness is settled by argument, replication, or measurement. We argue RSI must next operate across real, diverse scientific, engineering, and meta-scientific domains, not where formal evaluation is merely tractable. To that end we present MetaRSI-v1, where improvement is the scheduled composition of three typed operators over one unified paradigm. Data-RSI amplifies existing competence and marks its boundary; Harness-RSI edits a five-slot scaffold without touching weights; Model-RSI internalizes capability into parameters through bounded training. Sharing one loop kernel and artifact vocabulary, they make data, scaffold, and model changes composable rather than exclusive. A two-axis optimizer jointly decides operator order and each operator's proposal policy, while a meta-level policy revises the schedule across terms. We validate MetaRSI-v1 under the field's standard evaluations, on code and closed-form science, with no external teacher: the target model plays every role in its own loop. MetaRSI-v1 reframes self-improvement from a single-surface edit to a composition across the full model-production pipeline, opening two paths: a model route internalizing capability through training, and a harness route leaving weights untouched and thus extending self-improvement to any model reachable through an interface, with Data-RSI redefined as the shared substrate feeding both. The framework further yields refutable laws on where loops exist, how operators compose, and what supervision buys.

cs.LG↗

arXiv2026

Combating Instruction Conflict via Energy-Driven Latent Conflict Detection

Large Language Models (LLMs) are increasingly deployed with hierarchical instructions, yet they remain vulnerable to conflicts in which user directives override system-level constraints. Existing defense mechanisms predominantly focus on static input inspection and therefore fail to detect Response Drift, a phenomenon in which the model's final response violates system-level constraints despite seemingly compliant inputs. To bridge this gap, we introduce ELCD, a response-level latent conflict detector for post-generation, pre-delivery verification. Given the full generated output, ELCD constructs a composite hidden-state representation by concatenating the final-token embedding with the mean-pooled response embedding. It then optimizes a pairwise margin ranking objective to separate compliant and drifting responses in latent space. Extensive experiments across five mainstream LLMs ranging from 1.5B to 14B parameters demonstrate that ELCD significantly outperforms competitive baselines. Notably, it improves the PR-AUC on Llama-2-7B by approximately 30 percentage points and reduces the False Positive Rate at 95% TPR (FPR95) on Mistral-7B to 2.67%. These results suggest that ELCD provides a promising approach for latent instruction-conflict detection in open-weight or self-hosted LLM deployments.

cs.CL↗

arXiv2026

Open Rubric System: Scaling Reinforcement Learning with Pairwise Adaptive Rubric

Scalar reward models compress multi-dimensional human preferences into a single opaque score, creating an information bottleneck that often leads to brittleness and reward hacking in open-ended alignment. We argue that robust alignment for non-verifiable tasks is fundamentally a principle generalization problem: reward should not be a learned function internalized into a judge, but an explicit reasoning process executed under inspectable principles. To operationalize this view, we present the Open Rubric System (OpenRS), a plug-and-play, rubrics-based LLM-as-a-Judge framework built around Pairwise Adaptive Meta-Rubrics (PAMR) and lightweight Pointwise Verifiable Rubrics (PVRs), which provide both hard-constraint guardrails and verifiable reward components when ground-truth or programmatic checks are available. OpenRS uses an explicit meta-rubric -- a constitution-like specification that governs how rubrics are instantiated, weighted, and enforced -- and instantiates adaptive rubrics on the fly by conditioning on the semantic differences between two candidate responses. It then performs criterion-wise pairwise comparisons and aggregates criterion-level preferences externally, avoiding pointwise weighted scalarization while improving discriminability in open-ended settings. To keep principles consistent yet editable across various domains, we introduce a two-level meta-rubric refinement pipeline (automated evolutionary refinement for general principles and a reproducible human-in-the-loop procedure for domain principles), complemented with pointwise verifiable rubrics that act as both guardrails against degenerate behaviors and a source of verifiable reward for objective sub-tasks. Finally, we instantiate OpenRS as reward supervision in pairwise RL training.

cs.CL↗

arXiv2026

Wilf Equivalence for Length-Three Patterns and Flat POPs, and a Conjecture of Qiu and Remmel

It is well known that, for each classical pattern $τ$ of length 3, the number of $τ$-avoiding permutations of length $n$ is the $n$th Catalan number, and numerous bijections between different length-three avoidance classes have been constructed and studied. In this paper, we refine this classical problem by studying Wilf equivalence among permutations that simultaneously avoid a classical pattern of length three and a flat partially ordered pattern. Partially ordered patterns (POPs) provide a flexible framework for encoding families of classical permutation patterns. For $\ell\geq 3$ and $1\leq x\leq\ell$, let $P_{\ell,x}$ be the length-$\ell$ POP in which the entry at position $x$ is required to be smaller than all the other entries, while no relations are imposed among the remaining entries. Such POPs are called flat POPs. We classify the Wilf equivalences among all pairs $(τ,P_{\ell,x})$, where $τ$ is a classical pattern of length three. For every $\ell\geq4$, the resulting $6\ell$ pairs form exactly $2\ell-1$ Wilf equivalence classes, while the exceptional case $\ell=3$ gives four classes. Our proofs combine the derivation of explicit formulas and recurrence relations with the construction of bijections. Moreover, we introduce novel prime-divisor arguments to distinguish the remaining candidate classes, reducing the problem to showing that a certain Diophantine equation has no solutions for $\ell\ge 3{,}274$, where the bound $3{,}274$ is not claimed to be sharp. Finally, by extending our work on POPs, we resolve a conjecture of Qiu and Remmel concerning the distribution of quadrant marked mesh patterns on 132-avoiding permutations and correct an error in their paper that is crucial to the proof.

math.CO↗

arXiv2026

Long-term marine acoustic and seismic monitoring using distributed acoustic sensing and deep learning

The ocean remains one of the least instrumented parts of Earth, and many geophysical, biological, and anthropogenic signals go undetected for lack of instrumentation. Distributed acoustic sensing (DAS) can transform submarine fiber-optic cables into dense seafloor sensor arrays, but extracting diverse signals from massive DAS recordings remains challenging. Here we present DASNet, a deep learning framework that detects, classifies, and picks arrival times of diverse marine signals in continuous DAS data. Applied to nearly four years of Seafloor Fiber-Optic Array in Monterey Bay recordings, DASNet identifies more than 620,000 events. These detections reveal local earthquakes; distant earthquake- and volcanic-eruption-generated T-waves from the southwestern Pacific and mid-ocean ridge systems; more than 510,000 blue and fin whale calls with seasonal and interannual variability consistent with hydrophone records; and vessel traffic near the cable. Together, these results show that submarine fiber-optic cables combined with deep learning enable scalable, high-resolution ocean monitoring.

physics.geo-ph↗

arXiv2026

Uno-Orchestra: Parsimonious Agent Routing via Selective Delegation

Large language model (LLM) multi-agent systems typically rely on rigid orchestration, committing either to flat per-query routing or to hand-engineered task decomposition, so decomposition depth, worker choice, and inference budget are not jointly optimized under one objective. We introduce Uno-Orchestra, a unified orchestration policy that selectively decomposes a task and dispatches each subtask to an admissible (model, primitive) pair, with both decisions learned together from curated RL trajectories grounded in real worker interactions. Against 22 baselines on a 13-benchmark suite spanning math, code, knowledge, long-context, and agentic tool-use, Uno-Orchestra reaches 77.0% macro pass@1, roughly 16% above the strongest workflow baseline, at roughly an order of magnitude lower per-query cost, advancing the accuracy-efficiency frontier of selective delegation.

cs.AI↗

arXiv2026

Attention Residuals

Residual connections with PreNorm are standard in modern LLMs, yet they accumulate all layer outputs with fixed unit weights. This uniform aggregation causes uncontrolled hidden-state growth with depth, progressively diluting each layer's contribution. We propose Attention Residuals (AttnRes), which replaces this fixed accumulation with softmax attention over preceding layer outputs, allowing each layer to selectively aggregate earlier representations with learned, input-dependent weights. To address the memory and communication overhead of attending over all preceding layer outputs for large-scale model training, we introduce Block AttnRes, which partitions layers into blocks and attends over block-level representations, reducing the memory footprint while preserving most of the gains of full AttnRes. Combined with cache-based pipeline communication and a two-phase computation strategy, Block AttnRes becomes a practical drop-in replacement for standard residual connections with minimal overhead. Scaling law experiments confirm that the improvement is consistent across model sizes, and ablations validate the benefit of content-dependent depth-wise selection. We further integrate AttnRes into the Kimi Linear architecture (48B total / 3B activated parameters) and pre-train on 1.4T tokens, where AttnRes mitigates PreNorm dilution, yielding more uniform output magnitudes and gradient distribution across depth, and improves downstream performance across all evaluated tasks.

cs.CL↗

arXiv2026

Large Language Model Compression with Global Rank and Sparsity Optimization

Low-rank and sparse composite approximation is a natural idea to compress Large Language Models (LLMs). However, such an idea faces two primary challenges that adversely affect the performance of existing methods. The first challenge relates to the interaction and cooperation between low-rank and sparse matrices, while the second involves determining weight allocation across different layers, as redundancy varies considerably among them. To address these challenges, we propose a novel two-stage LLM compression method with the capability of global resource allocation for rank and sparsity. It is noteworthy that the overall optimization space is vast, making comprehensive optimization computationally prohibitive. Therefore, to reduce the optimization space, our first stage utilizes robust principal component analysis to decompose the weight matrices of LLMs into low-rank and sparse components, which span the low dimensional and sparse spaces containing the resultant low-rank and sparse matrices, respectively. In the second stage, we propose a probabilistic global allocation strategy to jointly identify the low-rank and sparse structures within the above two spaces. The appealing feature of our approach is its ability to automatically detect the redundancy across different layers and to manage the interaction between the sparse and low-rank components. Extensive experimental results indicate that our method significantly surpasses state-of-the-art techniques for sparsification and composite approximation.

cs.LG↗

arXiv2026

Surrogate Ensemble in Expensive Multi-Objective Optimization via Deep Q-Learning

Surrogate-assisted Evolutionary Algorithms~(SAEAs) have shown promising robustness in solving expensive optimization problems. A key aspect that impacts SAEAs' effectiveness is surrogate model selection, which in existing works is predominantly decided by human developer. Such human-made design choice introduces strong bias into SAEAs and may hurt their expected performance on out-of-scope tasks. In this paper, we propose a reinforcement learning-assisted ensemble framework, termed as SEEMOO, which is capable of scheduling different surrogate models within a single optimization process, hence boosting the overall optimization performance in a cooperative paradigm. Specifically, we focus on expensive multi-objective optimization problems, where multiple objective functions shape a compositional landscape and hence challenge surrogate selection. SEEMOO comprises following core designs: 1) A pre-collected model pool that maintains different surrogate models; 2) An attention-based state-extractor supports universal optimization state representation of problems with varied objective numbers; 3) a deep Q-network serves as dynamic surrogate selector: Given the optimization state, it selects desired surrogate model for current-step evaluation. SEEMOO is trained to maximize the overall optimization performance under a training problem distribution. Extensive benchmark results demonstrate SEEMOO's surrogate ensemble paradigm boosts the optimization performance of single-surrogate baselines. Further ablation studies underscore the importance of SEEMOO's design components.

cs.NE↗

arXiv2026

Dowling's polynomial conjecture for independent sets of matroids

The celebrated Mason's conjecture states that the sequence of independent set numbers of any matroid is log-concave, and even ultra log-concave. The strong form of Mason's conjecture was independently solved by Anari, Liu, Oveis Gharan and Vinzant, and by Brändén and Huh. The weak form of Mason's conjecture was also generalized to a polynomial version by Dowling in 1980 by considering certain polynomial analogue of independent set numbers. In this paper we completely solve Dowling's polynomial conjecture by using the theory of Lorentzian polynomials.

math.CO↗

arXiv2025

Kimi Linear: An Expressive, Efficient Attention Architecture

We introduce Kimi Linear, a hybrid linear attention architecture that, for the first time, outperforms full attention under fair comparisons across various scenarios -- including short-context, long-context, and reinforcement learning (RL) scaling regimes. At its core lies Kimi Delta Attention (KDA), an expressive linear attention module that extends Gated DeltaNet with a finer-grained gating mechanism, enabling more effective use of limited finite-state RNN memory. Our bespoke chunkwise algorithm achieves high hardware efficiency through a specialized variant of the Diagonal-Plus-Low-Rank (DPLR) transition matrices, which substantially reduces computation compared to the general DPLR formulation while remaining more consistent with the classical delta rule. We pretrain a Kimi Linear model with 3B activated parameters and 48B total parameters, based on a layerwise hybrid of KDA and Multi-Head Latent Attention (MLA). Our experiments show that with an identical training recipe, Kimi Linear outperforms full MLA with a sizeable margin across all evaluated tasks, while reducing KV cache usage by up to 75% and achieving up to 6 times decoding throughput for a 1M context. These results demonstrate that Kimi Linear can be a drop-in replacement for full attention architectures with superior performance and efficiency, including tasks with longer input and output lengths. To support further research, we open-source the KDA kernel and vLLM implementations, and release the pre-trained and instruction-tuned model checkpoints.

cs.CL↗

arXiv2025

Physics-informed machine learning for combustion: A review

Physics-informed machine learning (PIML) represents an emerging paradigm that integrates various forms of physical knowledge into machine learning (ML) components, thereby enhancing the physical consistency of ML models compared to purely data-driven paradigms. The field of combustion, characterized by a rich foundation of physical laws and abundant data, is undergoing a transformation due to PIML. This paper aims to provide a comprehensive overview of PIML for combustion, systematically outlining fundamental principles, significant contributions, key advancements, and available resources. The application of PIML in combustion is categorized into three domains: combustion chemical kinetics, combustion reacting flows, and other combustion-related scenarios. Additionally, current challenges, potential solutions, and practical guidelines for researchers and engineers will be discussed. A primary focus of this review is to demonstrate how combustion laws can be integrated into ML, either through soft or hard constraints, via loss functions or representation models, and within coordinate-to-variable or field-to-field paradigms. This paper shows that PIML offers a unified framework linking physics, model, and data in combustion--integrating physical knowledge in model-to-data simulation and reconstruction tasks, as well as data-to-model modeling tasks--resulting in enhanced data, improved physical models, and more reliable ML models. PIML for combustion presents significant opportunities for both the combustion and ML communities, encouraging greater collaboration and cross-disciplinary engagement.

physics.chem-ph↗

arXiv2025

Autonomous Elemental Characterization Enabled by a Low Cost Robotic Platform Built Upon a Generalized Software Architecture

Despite the rapidly growing applications of robots in industry, the use of robots to automate tasks in scientific laboratories is less prolific due to lack of generalized methodologies and high cost of hardware. This paper focuses on the automation of characterization tasks necessary for reducing cost while maintaining generalization, and proposes a software architecture for building robotic systems in scientific laboratory environment. A dual-layer (Socket.IO and ROS) action server design is the basic building block, which facilitates the implementation of a web-based front end for user-friendly operations and the use of ROS Behavior Tree for convenient task planning and execution. A robotic platform for automating mineral and material sample characterization is built upon the architecture, with an open source, low-cost three-axis computer numerical control gantry system serving as the main robot. A handheld laser induced breakdown spectroscopy (LIBS) analyzer is integrated with a 3D printed adapter, enabling automated 2D chemical mapping. We demonstrate the utility of automated chemical mapping by scanning of the surface of a spodumene-bearing pegmatite core sample with a 1071-point dense hyperspectral map acquired at a rate of 1520 bits per second. Automated LIBS scanning enables controlled chemical quantification in the laboratory that complements field-based measurements acquired with the same handheld device, linking resource exploration and processing steps in the supply chain for lithium-based battery materials.

cs.RO↗

arXiv2025

Kimi-VL Technical Report

We present Kimi-VL, an efficient open-source Mixture-of-Experts (MoE) vision-language model (VLM) that offers advanced multimodal reasoning, long-context understanding, and strong agent capabilities - all while activating only 2.8B parameters in its language decoder (Kimi-VL-A3B). Kimi-VL demonstrates strong performance across challenging domains: as a general-purpose VLM, Kimi-VL excels in multi-turn agent tasks (e.g., OSWorld), matching flagship models. Furthermore, it exhibits remarkable capabilities across diverse challenging vision language tasks, including college-level image and video comprehension, OCR, mathematical reasoning, and multi-image understanding. In comparative evaluations, it effectively competes with cutting-edge efficient VLMs such as GPT-4o-mini, Qwen2.5-VL-7B, and Gemma-3-12B-IT, while surpassing GPT-4o in several key domains. Kimi-VL also advances in processing long contexts and perceiving clearly. With a 128K extended context window, Kimi-VL can process diverse long inputs, achieving impressive scores of 64.5 on LongVideoBench and 35.1 on MMLongBench-Doc. Its native-resolution vision encoder, MoonViT, further allows it to see and understand ultra-high-resolution visual inputs, achieving 83.2 on InfoVQA and 34.5 on ScreenSpot-Pro, while maintaining lower computational cost for common tasks. Building upon Kimi-VL, we introduce an advanced long-thinking variant: Kimi-VL-Thinking-2506. Developed through long chain-of-thought (CoT) supervised fine-tuning (SFT) and reinforcement learning (RL), the latest model exhibits strong long-horizon reasoning capabilities (64.0 on MMMU, 46.3 on MMMU-Pro, 56.9 on MathVision, 80.1 on MathVista, 65.2 on VideoMMMU) while obtaining robust general abilities. Code and models are publicly accessible at https://github.com/MoonshotAI/Kimi-VL.

cs.CV↗

arXiv2025

Kimi k1.5: Scaling Reinforcement Learning with LLMs

Language model pretraining with next token prediction has proved effective for scaling compute but is limited to the amount of available training data. Scaling reinforcement learning (RL) unlocks a new axis for the continued improvement of artificial intelligence, with the promise that large language models (LLMs) can scale their training data by learning to explore with rewards. However, prior published work has not produced competitive results. In light of this, we report on the training practice of Kimi k1.5, our latest multi-modal LLM trained with RL, including its RL training techniques, multi-modal data recipes, and infrastructure optimization. Long context scaling and improved policy optimization methods are key ingredients of our approach, which establishes a simplistic, effective RL framework without relying on more complex techniques such as Monte Carlo tree search, value functions, and process reward models. Notably, our system achieves state-of-the-art reasoning performance across multiple benchmarks and modalities -- e.g., 77.5 on AIME, 96.2 on MATH 500, 94-th percentile on Codeforces, 74.9 on MathVista -- matching OpenAI's o1. Moreover, we present effective long2short methods that use long-CoT techniques to improve short-CoT models, yielding state-of-the-art short-CoT reasoning results -- e.g., 60.8 on AIME, 94.6 on MATH500, 47.3 on LiveCodeBench -- outperforming existing short-CoT models such as GPT-4o and Claude Sonnet 3.5 by a large margin (up to +550%).

cs.AI↗

arXiv2025

G1: Bootstrapping Perception and Reasoning Abilities of Vision-Language Model via Reinforcement Learning

Vision-Language Models (VLMs) excel in many direct multimodal tasks but struggle to translate this prowess into effective decision-making within interactive, visually rich environments like games. This ``knowing-doing'' gap significantly limits their potential as autonomous agents, as leading VLMs often performing badly in simple games. To address this, we introduce VLM-Gym, a curated reinforcement learning (RL) environment featuring diverse visual games with unified interfaces and adjustable, compositional difficulty, specifically designed for scalable multi-game parallel training. Leveraging VLM-Gym, we train G0 models using pure RL-driven self-evolution, which demonstrate emergent perception and reasoning patterns. To further mitigate challenges arising from game diversity, we develop G1 models. G1 incorporates a perception-enhanced cold start prior to RL fine-tuning. Our resulting G1 models consistently surpass their teacher across all games and outperform leading proprietary models like Claude-3.7-Sonnet-Thinking. Systematic analysis reveals an intriguing finding: perception and reasoning abilities mutually bootstrap each other throughout the RL training process. Source code including VLM-Gym and RL training are released at https://github.com/chenllliang/G1 to foster future research in advancing VLMs as capable interactive agents.

cs.CV↗

arXiv2025

Kimi-Audio Technical Report

We present Kimi-Audio, an open-source audio foundation model that excels in audio understanding, generation, and conversation. We detail the practices in building Kimi-Audio, including model architecture, data curation, training recipe, inference deployment, and evaluation. Specifically, we leverage a 12.5Hz audio tokenizer, design a novel LLM-based architecture with continuous features as input and discrete tokens as output, and develop a chunk-wise streaming detokenizer based on flow matching. We curate a pre-training dataset that consists of more than 13 million hours of audio data covering a wide range of modalities including speech, sound, and music, and build a pipeline to construct high-quality and diverse post-training data. Initialized from a pre-trained LLM, Kimi-Audio is continual pre-trained on both audio and text data with several carefully designed tasks, and then fine-tuned to support a diverse of audio-related tasks. Extensive evaluation shows that Kimi-Audio achieves state-of-the-art performance on a range of audio benchmarks including speech recognition, audio understanding, audio question answering, and speech conversation. We release the codes, model checkpoints, as well as the evaluation toolkits in https://github.com/MoonshotAI/Kimi-Audio.

eess.AS↗

arXiv2025

Muon is Scalable for LLM Training

Recently, the Muon optimizer based on matrix orthogonalization has demonstrated strong results in training small-scale language models, but the scalability to larger models has not been proven. We identify two crucial techniques for scaling up Muon: (1) adding weight decay and (2) carefully adjusting the per-parameter update scale. These techniques allow Muon to work out-of-the-box on large-scale training without the need of hyper-parameter tuning. Scaling law experiments indicate that Muon achieves $\sim\!2\times$ computational efficiency compared to AdamW with compute optimal training. Based on these improvements, we introduce Moonlight, a 3B/16B-parameter Mixture-of-Expert (MoE) model trained with 5.7T tokens using Muon. Our model improves the current Pareto frontier, achieving better performance with much fewer training FLOPs compared to prior models. We open-source our distributed Muon implementation that is memory optimal and communication efficient. We also release the pretrained, instruction-tuned, and intermediate checkpoints to support future research.

cs.LG↗