arXiv subjects

Lingyu Yang

Publications and source records attributed to Lingyu Yang.

17 recordsLinked to original sources

arXiv2026

The MiniMax-M2 Series: Mini Activations Unleashing Max Real-World Intelligence

We introduce the MiniMax-M2 series, a family of Mixture-of-Experts language models built around the principle that mini activations can unleash maximum real-world intelligence. The flagship M2 contains 229.9B total parameters with only 9.8B activated per token. Designed end-to-end for agentic deployment, the M2 series rests on three components: (i) agent-driven data pipelines producing large-scale, verifiable trajectories across agentic coding and agentic cowork, each grounded in an executable workspace and an artifact-aligned reward; (ii) Forge, a scalable agent-native RL system that adapts to long-horizon agent trajectories, paired with windowed-FIFO scheduling, prefix-tree merging, inference optimization, and a clean training-inference-agent decoupling that supports both white-box and black-box agents; (iii) the latest M2.7 checkpoint takes an early step toward self-evolution -- autonomously debugging training runs and modifying its own scaffold. Across M2 through M2.7, this combination translates a mini-activation footprint into frontier-tier performance on agentic coding, deep search, office-task, and reasoning benchmarks.

cs.AI↗

arXiv2026

Contexting as Recommendation: Evolutionary Collaborative Filtering for Context Engineering

Large Language Models (LLMs) are highly sensitive to their input contexts, motivating the development of automated context engineering. However, existing methods predominantly treat this as a global search problem, seeking a single context strategy that maximizes average performance across a dataset. This restrictive assumption overlooks the fact that different inputs often require distinct guidance, leaving substantial instance-level performance gains untapped. In this paper, we propose a paradigm shift by formulating context engineering as a recommendation problem. We introduce \textbf{Neural Collaborative Context Engineering (NCCE)}, a framework that transitions optimization from a static global search to dynamic, instance-wise routing. NCCE first bootstraps a diverse catalog of anchor contexts and then employs a novel \textbf{Context-CF Co-Evolution} mechanism. This stage establishes a synergistic feedback loop: a lightweight Neural Collaborative Filtering (NCF) model learns instance-context preferences to guide the generation of specialized context variants, while the newly evaluated contexts continuously refine the NCF model's understanding of latent preferences. At inference time, the trained NCF model acts as a context router, dynamically assigning the most suitable context strategy to each unseen instance. Theoretical Proofs and comprehensive experiments demonstrate that by matching individual inputs with their optimal contexts, NCCE significantly improves task accuracy, highlighting the critical importance of personalization in LLM context engineering.

cs.CL↗

arXiv2026

Holder Policy Optimisation

Group Relative Policy Optimisation (GRPO) enhances large language models by estimating advantages across a group of sampled trajectories. However, mapping these trajectory-level advantages to policy updates requires aggregating token-level probabilities within each sequence. Relying on a fixed aggregation mechanism for this step fundamentally limits the algorithm's adaptability. Empirically, we observe a critical trade-off: certain fixed aggregations frequently suffer from training collapse, while others fail to yield satisfactory performance. To resolve this, we propose \textbf{H\"{o}lderPO}, a generalised policy optimisation framework unifying token-level probability aggregation via the H\"{o}lder mean. By explicitly modulating the parameter $p$, our framework provides continuous control over the trade-off between gradient concentration and variance bounds. Theoretically, we prove that a larger $p$ concentrates the gradient to amplify sparse learning signals, whereas a smaller $p$ strictly bounds gradient variance. Because no static configuration can universally resolve this concentration-stability trade-off, we instantiate the framework with a dynamic annealing algorithm that progressively schedules $p$ across the training lifecycle. Extensive evaluations demonstrate superior stability and convergence over existing baselines. Specifically, our approach achieves a state-of-the-art average accuracy of $54.9\%$ across multiple mathematical benchmarks, yielding a substantial $7.2\%$ relative gain over standard GRPO and secures an exceptional $93.8\%$ success rate on ALFWorld.

cs.LG↗

arXiv2026

AcademiClaw: When Students Set Challenges for AI Agents

Benchmarks within the OpenClaw ecosystem have thus far evaluated exclusively assistant-level tasks, leaving the academic-level capabilities of OpenClaw largely unexamined. We introduce AcademiClaw, a bilingual benchmark of 80 complex, long-horizon tasks sourced directly from university students' real academic workflows -- homework, research projects, competitions, and personal projects -- that they found current AI agents unable to solve effectively. Curated from 230 student-submitted candidates through rigorous expert review, the final task set spans 25+ professional domains, ranging from olympiad-level mathematics and linguistics problems to GPU-intensive reinforcement learning and full-stack system debugging, with 16 tasks requiring CUDA GPU execution. Each task executes in an isolated Docker sandbox and is scored on task completion by multi-dimensional rubrics combining six complementary techniques, with an independent five-category safety audit providing additional behavioral analysis. Experiments on six frontier models show that even the best achieves only a 55\% pass rate. Further analysis uncovers sharp capability boundaries across task domains, divergent behavioral strategies among models, and a disconnect between token consumption and output quality, providing fine-grained diagnostic signals beyond what aggregate metrics reveal. We hope that AcademiClaw and its open-sourced data and code can serve as a useful resource for the OpenClaw community, driving progress toward agents that are more capable and versatile across the full breadth of real-world academic demands. All data and code are available at https://github.com/GAIR-NLP/AcademiClaw.

cs.AI↗

arXiv2026

Turing Test on Screen: A Benchmark for Mobile GUI Agent Humanization

The rise of autonomous GUI agents has triggered adversarial countermeasures from digital platforms, yet existing research prioritizes utility and robustness over the critical dimension of anti-detection. We argue that for agents to survive in human-centric ecosystems, they must evolve Humanization capabilities. We introduce the ``Turing Test on Screen,'' formally modeling the interaction as a MinMax optimization problem between a detector and an agent aiming to minimize behavioral divergence. We then collect a new high-fidelity dataset of mobile touch dynamics, and conduct our analysis that vanilla LMM-based agents are easily detectable due to unnatural kinematics. Consequently, we establish the Agent Humanization Benchmark (AHB) and detection metrics to quantify the trade-off between imitability and utility. Finally, we propose methods ranging from heuristic noise to data-driven behavioral matching, demonstrating that agents can achieve high imitability theoretically and empirically without sacrificing performance. This work shifts the paradigm from whether an agent can perform a task to how it performs it within a human-centric ecosystem, laying the groundwork for seamless coexistence in adversarial digital environments.

cs.AI↗

arXiv2026

Pseudospin Formulation of Quench Dynamics in the Semiclassical Holstein Model

We present a pseudospin formulation for the post-quench dynamics of charge-density-wave (CDW) order in the half-filled spinless Holstein model on a square lattice, assuming spatially homogeneous evolution. This Anderson pseudospin description captures the coherent nonequilibrium dynamics of the coupled electron-lattice system. Numerical simulations reveal three distinct dynamical regimes of the CDW order parameter following a quench-locked oscillations, Landau-damped dynamics, and overdamped relaxation-closely paralleling quench dynamics in BCS superconductors and other electronically driven symmetry-breaking phases. Crucially, however, the presence of dynamical lattice degrees of freedom leads to qualitatively different long-time behavior. In particular, while the oscillation amplitude is reduced in the damped regimes, CDW oscillations do not fully decay but instead persist indefinitely due to feedback from the lattice field. We further show that these persistent oscillations are characterized by a nonequilibrium electronic distribution, which provides an intuitive understanding of both their amplitude and the renormalization of the oscillation frequency relative to the bare Holstein phonon frequency. Our results highlight the essential role of lattice dynamics in nonequilibrium ordered phases and establish a clear distinction between electron-lattice-driven CDW dynamics and their purely electronic counterparts.

cond-mat.str-el↗

arXiv2025

Do LLMs Know When to Flip a Coin? Strategic Randomization through Reasoning and Experience

Strategic randomization is a key principle in game theory, yet it remains underexplored in large language models (LLMs). Prior work often conflates the cognitive decision to randomize with the mechanical generation of randomness, leading to incomplete evaluations. To address this, we propose a novel zero-sum game inspired by the Tian Ji Horse Race, where the Nash equilibrium corresponds to a maximal entropy strategy. The game's complexity masks this property from untrained humans and underdeveloped LLMs. We evaluate five LLMs across prompt styles -- framed, neutral, and hinted -- using competitive multi-tournament gameplay with system-provided random choices, isolating the decision to randomize. Results show that weaker models remain deterministic regardless of prompts, while stronger models exhibit increased randomization under explicit hints. When facing weaker models, strong LLMs adopt deterministic strategies to exploit biases, but converge toward equilibrium play when facing peers. Through win/loss outcomes and Bayes factor analysis, we demonstrate meaningful variation in LLMs' strategic reasoning capabilities, highlighting opportunities for improvement in abstract reasoning and adaptive learning. We make our implementation publicly available at https://github.com/ocelopus/llm-when-to-throw-coin to ensure full reproducibility.

cs.AI↗

arXiv2025

MiniMax-M1: Scaling Test-Time Compute Efficiently with Lightning Attention

We introduce MiniMax-M1, the world's first open-weight, large-scale hybrid-attention reasoning model. MiniMax-M1 is powered by a hybrid Mixture-of-Experts (MoE) architecture combined with a lightning attention mechanism. The model is developed based on our previous MiniMax-Text-01 model, which contains a total of 456 billion parameters with 45.9 billion parameters activated per token. The M1 model natively supports a context length of 1 million tokens, 8x the context size of DeepSeek R1. Furthermore, the lightning attention mechanism in MiniMax-M1 enables efficient scaling of test-time compute. These properties make M1 particularly suitable for complex tasks that require processing long inputs and thinking extensively. MiniMax-M1 is trained using large-scale reinforcement learning (RL) on diverse problems including sandbox-based, real-world software engineering environments. In addition to M1's inherent efficiency advantage for RL training, we propose CISPO, a novel RL algorithm to further enhance RL efficiency. CISPO clips importance sampling weights rather than token updates, outperforming other competitive RL variants. Combining hybrid-attention and CISPO enables MiniMax-M1's full RL training on 512 H800 GPUs to complete in only three weeks, with a rental cost of just $534,700. We release two versions of MiniMax-M1 models with 40K and 80K thinking budgets respectively, where the 40K model represents an intermediate phase of the 80K training. Experiments on standard benchmarks show that our models are comparable or superior to strong open-weight models such as the original DeepSeek-R1 and Qwen3-235B, with particular strengths in complex software engineering, tool utilization, and long-context tasks. We publicly release MiniMax-M1 at https://github.com/MiniMax-AI/MiniMax-M1.

cs.CL↗

arXiv2024

Recurrent convolutional neural networks for modeling non-adiabatic dynamics of quantum-classical systems

Recurrent neural networks (RNNs) have recently been extensively applied to model the time-evolution in fluid dynamics, weather predictions, and even chaotic systems thanks to their ability to capture temporal dependencies and sequential patterns in data. Here we present a RNN model based on convolution neural networks for modeling the nonlinear non-adiabatic dynamics of hybrid quantum-classical systems. The dynamical evolution of the hybrid systems is governed by equations of motion for classical degrees of freedom and von Neumann equation for electrons. The physics-aware recurrent convolution (PARC) neural network structure incorporates a differentiator-integrator architecture that inductively models the spatiotemporal dynamics of generic physical systems. We apply our RNN approach to learn the space-time evolution of a one-dimensional semi-classical Holstein model after an interaction quench. For shallow quenches (small changes in electron-lattice coupling), the deterministic dynamics can be accurately captured using a single-CNN-based recurrent network. In contrast, deep quenches induce chaotic evolution, making long-term trajectory prediction significantly more challenging. Nonetheless, we demonstrate that the PARC-CNN architecture can effectively learn the statistical climate of the Holstein model under deep-quench conditions.

quant-ph↗

arXiv2024

Electronic glasses from a broken gauge symmetry in disorder-free systems

Glass phases can be stabilized by quenched disorders, as in most spin-glass materials, or self-generated through kinetic freezing in disorder-free systems. A canonical example of the latter is structural glasses, which have been extensively studied for many decades. Yet, how the rugged energy landscape of a glass phase is spontaneously generated in disorder-free systems remains one of the key questions in glass physics. Here we present a general electronic mechanism for the emergence of glassy phase using the example of itinerant electrons coupled to XY spins on a lattice. This model can also be be viewed as the mean-field theory of a superconducting system with attractive density-density interactions. Intriguingly, the electron gauge symmetry in the strong pairing limit gives rise to a macroscopic degeneracy of XY spins. In the presence of electron hopping that breaks the gauge symmetry, the lifting of the extensive degeneracy leads to a glass phase with disordered pairings. Our findings highlight a novel scenario in which a glassy state originates from the breaking of quantum gauge symmetry without quenched disorders.

cond-mat.dis-nn↗

arXiv2024

Photo-induced pattern formations and melting of charge-density-wave order

We investigate the out-of-equilibrium dynamics of a photo-excited charge-density-wave (CDW) state in the square-lattice Holstein model, in a setup similar to a pump-probe experiment. At half-filling, the ground state of this system is characterized by a checkerboard modulation of particle densities, accompanied by a concomitant lattice distortion. An efficient real-space dynamics method integrating the von~Neumann equation for electron density matrix and Newton equation for classical lattice dynamics is developed to simulate the dynamical evolution of a photo-excited Holstein system. We find that the energy injected by a short pump pulse results in the reduction of the CDW order and the generation of coherent phonons. At strong photoexcitations, while the CDW order is melted in the sense that the time-averaged order parameter vanishes, a dynamical CDW state is self-sustained by the strong coherent phonon oscillations. Our large-scale simulations further uncover a dynamical regime of intermediate fluence where complex pattern formation is induced by the pump pulse. The emergent spatial textures are characterized by super density modulations on top of the short-range checkerboard CDW order. Our demonstration of pattern formation highlights the significance of dynamical inhomogeneity in quantum many-body systems in pump-probe experiments.

cond-mat.str-el↗

arXiv2024

Combinatorial Printing of Functionally Graded Solid-State Electrolyte for High-Voltage Lithium Metal Batteries

Heterogeneous multilayered solid-state electrolyte (HMSSE) has been widely explored for their broadened working voltage range and compatibility with electrodes. However, due to the limitations of traditional manufacturing methods such as casting, the interface between electrolyte layers in HMSSE can decrease the ionic conductivity severely. Here, a novel combinatory aerosol jet printing (CAJP) is introduced to fabricate functionally graded solid-state electrolyte (FGSSE) without sharp interface. Owing to the unique ability of CAJP (in-situ mixing and instantaneous tuning of the mixing ratio), FGSSE with smooth microscale compositional gradation is achieved. Electrochemical tests show that FGSSE has excellent oxidative stability exceeding 5.5 V and improved conductivity (>7 times of an analogous HMSSE). By decoupling the total resistance, we show that the resistance from the electrolyte/electrolyte interface of HMSSE is 5.7 times of the total resistance of FGSSE. The Li/FGSSE/NCM622 cell can be stably run for more than 200 cycles along with improved rate performance.

cond-mat.mtrl-sci↗

arXiv2023

Pattern formation in charge density wave states after a quantum quench

We study post-quench dynamics of charge-density-wave (CDW) order in the square-lattice $t$-$V$ model. The ground state of this system at half-filling is characterized by a checkerboard modulation of particle density. A generalized self-consistent mean-field method, based on the time-dependent variational principle, is employed to describe the dynamical evolution of the CDW states. Assuming a homogeneous CDW order throughout the quench process, the time-dependent mean-field approach is reduced to the Anderson pseudospin method. Quench simulations based on the Bloch equation for pseudospins produce three canonical behaviors of order-parameter dynamics: phase-locked persistent oscillation, Landau-damped oscillation, and dynamical vanishing of the CDW order. We further develop an efficient real-space von Neumann equation method to incorporate dynamical inhomogeneity into simulations of quantum quenches. Our large-scale simulations uncover complex pattern formations in the post-quench CDW states, especially in the strong quench regime. The emergent spatial textures are characterized by super density modulations on top of the short-period checkerboard CDW order. Our demonstration of pattern formation in quenched CDW states, described by a simple broken $Z_2$ symmetry, underscores the importance of dynamical inhomogeneity in quantum quenches of many-body systems with more complex orders.

cond-mat.stat-mech↗

arXiv2023

Driven Majorana Modes: A Route to Synthetic $p_x+ip_y$ Superconductivity

We propose a protocol to realize synthetic $p_x+ip_y$ superconductors in one-dimensional topological systems that host Majorana fermions. By periodically driving a localized Majorana mode across the system, our protocol realizes a topological pumping of Majorana fermions, analogous to the adiabatic Thouless pumping of electrical charges. Importantly, similar to the realization of a Chern insulator through Thouless pumping, we show that pumping of Majorana zero modes could lead to a $p_x + ip_y$ superconductor in the two dimensions of space and synthetic time. The Floquet theory is employed to map the driven one-dimensional system to a two-dimensional synthetic system by considering frequency as a new dimension. We demonstrate such Floquet $p_x + i p_y$ superconductors using the Kitaev $p$-wave superconductor chain, a prototypical 1D topological system, as well as its more realistic realization in the 1D Kondo lattice model as examples. We further show the appearance of a new $π$ Majorana mode at the Floquet zone boundary in an intermediate drive frequency region. Our work suggests a driven magnetic spiral coupled to a superconductor as a promising platform for the realization of novel topological superconductors.

cond-mat.supr-con↗

arXiv2023

Neural Machine Translation with Dynamic Graph Convolutional Decoder

Existing wisdom demonstrates the significance of syntactic knowledge for the improvement of neural machine translation models. However, most previous works merely focus on leveraging the source syntax in the well-known encoder-decoder framework. In sharp contrast, this paper proposes an end-to-end translation architecture from the (graph \& sequence) structural inputs to the (graph \& sequence) outputs, where the target translation and its corresponding syntactic graph are jointly modeled and generated. We propose a customized Dynamic Spatial-Temporal Graph Convolutional Decoder (Dyn-STGCD), which is designed for consuming source feature representations and their syntactic graph, and auto-regressively generating the target syntactic graph and tokens simultaneously. We conduct extensive experiments on five widely acknowledged translation benchmarks, verifying that our proposal achieves consistent improvements over baselines and other syntax-aware variants.

cs.CL↗

arXiv2022

HHF: Hashing-guided Hinge Function for Deep Hashing Retrieval

Deep hashing has shown promising performance in large-scale image retrieval. However, latent codes extracted by Deep Neural Networks (DNNs) will inevitably lose semantic information during the binarization process, which damages the retrieval accuracy and makes it challenging. Although many existing approaches perform regularization to alleviate quantization errors, we figure out an incompatible conflict between metric learning and quantization learning. The metric loss penalizes the inter-class distances to push different classes unconstrained far away. Worse still, it tends to map the latent code deviate from ideal binarization point and generate severe ambiguity in the binarization process. Based on the minimum distance of the binary linear code, we creatively propose Hashing-guided Hinge Function (HHF) to avoid such conflict. In detail, the carefully-designed inflection point, which relies on the hash bit length and category numbers, is explicitly adopted to balance the metric term and quantization term. Such a modification prevents the network from falling into local metric optimal minima in deep hashing. Extensive experiments in CIFAR-10, CIFAR-100, ImageNet, and MS-COCO show that HHF consistently outperforms existing techniques, and is robust and flexible to transplant into other methods. Code is available at https://github.com/JerryXu0129/HHF.

cs.CV↗

arXiv2020

Acoustic anomaly detection via latent regularized gaussian mixture generative adversarial networks

Acoustic anomaly detection aims at distinguishing abnormal acoustic signals from the normal ones. It suffers from the class imbalance issue and the lacking in the abnormal instances. In addition, collecting all kinds of abnormal or unknown samples for training purpose is impractical and timeconsuming. In this paper, a novel Gaussian Mixture Generative Adversarial Network (GMGAN) is proposed under semi-supervised learning framework, in which the underlying structure of training data is not only captured in spectrogram reconstruction space, but also can be further restricted in the space of latent representation in a discriminant manner. Experiments show that our model has clear superiority over previous methods, and achieves the state-of-the-art results on DCASE dataset.

eess.AS↗