Searcharxiv⌕ Search

arXiv subjects

Huijuan Wang

Publications and source records attributed to Huijuan Wang.

At least 19 recordsLinked to original sources

K2-Horizon: A Frontier Fleet of Open Source Language Models

We present K2-Horizon, a fleet of six fully open language models, from a 0.9B dense model to a 375B-A23B MoE, built for general-purpose reasoning and agentic work across math and science, software engineering, deep research, office automation, and beyond. The models lead or are competitive with strong open models of their size on agentic and coding benchmarks. On the modeling side, to strengthen reasoning, we introduce Mixture-of-Value Attention (MoVA), applying MoE to attention value projections, and softplus attention gating, a more flexible, theoretically motivated gate. Our 36B-A4B model uses both and nearly matches a 32B dense model trained on the same data, with one-seventh the active parameters. We also introduce Uno, lossless diffusion adapters that speed up decoding by up to 2.2x without a separate draft model. Our release follows the LLM360 definition of open source for transparency and reproducibility: the full development history, including intermediate checkpoints from every training stage, pre-training data, training logs, and all training, evaluation, and infrastructure code, notably our production-scale xLLM framework for dense and MoE training and the RL360 reinforcement-learning stack. Beyond the data itself, we release and document in detail every pipeline that built it, the part of model development that open releases most rarely disclose. These pipelines produced roughly 10T synthetic pre-training tokens, showing that synthetic data can be a major ingredient of pre-training at scale. Four of the models share one 22T-token pre-training corpus and recipe, revealing consistent scaling behavior and enabling controlled studies of scale and sparsity. All artifacts are available at https://ifm.ai/k2/

cs.AI↗

Influence of Assortativity on the SICP Spreading Process in Hypergraphs

Many real-world systems involve group interactions that cannot be captured by pairwise graphs. They can be better represented by hypergraphs where each hyperedge could connect an arbitrary number of nodes. The Susceptible-Infected Contact Process (SICP) has been proposed to model the spreading of information and epidemics on hypergraphs. However, how higher-order topological properties of a hypergraph influence the SICP spreading process remains understudied. In this paper, we examine the impact of hypergraph assortativity based on the Neighborhood Coefficient (NC)-the average size of hyperedges incident to a node-on the SICP spreading process. By introducing a low-complexity measure of hypergraph assortativity grounded in analysis of variance, we apply hyperdegree-preserving rewirings to real-world networks and synthetic hypergraph models to tune their assortativity. We find and explain why lower NC assortativity yields higher average epidemic prevalence and lower prevalence variance across different seed selections. This phenomenon is evident when hyperdegree-preserving rewirings are applied to real-world networks with heterogeneous topological structures, but it is largely absent when comparing synthetic random hypergraphs to their rewired counterparts.

cs.SI↗

DDBench: A Benchmark for Agentic Debugging on Distributed Systems

LLM-based coding agents have advanced rapidly on single-process SWE tasks, with frontier models now clustering in the high-70s on SWE-bench Verified. Distributed-system debugging, however, remains an under-explored regime: bugs span processes, nodes, and protocol interactions, with root causes rarely recoverable from source alone and brute-force exploration intractable across non-deterministic interleavings. This leaves two gaps in LLM and agent evaluation: no code-repair benchmark targets distributed-system bugs, and no controlled study isolates how much externally provided debugging context changes agent success on them. We introduce DDBench, a code-repair benchmark of 60 historical bugs mined from 13 open-source distributed systems, partitioned into three reasoning-scope tiers. DDBench evaluates every case under two matched conditions: a symptom-only condition where the agent receives only the bug symptom and repository, and a context-augmented condition where it additionally receives bounded debugging context (logs, traces, runtime state, and targeted code-investigation notes), isolating the effect of debugging context from model capability. The evaluation of 10 LLMs on DDBench reveals several findings. First, distributed debugging exercises a reasoning dimension that single-process benchmarks do not surface: models' pass rates span 45 pp, and pairwise bootstrap separates 6 of 15 top-tier model pairs at p < 0.05 on DDBench's hardest case-set. Second, bounded debugging context lifts aggregate pass rate by +19.0 pp, and the lift is asymmetric: weaker models gain more in pass rate, while stronger models gain more in efficiency. Third, debugging context requires careful curation, as even faithful debugging context can anchor a strong model on the wrong layer of the system when its observation sits far from the fault.

cs.SE↗

A Network Science Perspective on Evaluating Deep Graph Generative Models

Traditional network models from network science, such as the Erdos-Renyi and configuration models, generate random networks that reproduce few selected topological properties observed in real-world networks. Deep graph generative models emerge as a data-driven approach, leveraging deep neural network architectures to learn complex structural distributions directly from real-world networks to generate more realistic synthetic networks. Because real social contact networks cannot be shared due to privacy risks, synthetic networks serve as an alternative for developing and evaluating epidemic mitigation strategies. In this work, we evaluate deep graph generative models as well as the configuration from a network science perspective by assessing both the topological similarity between generated and real-world networks and their utility in identifying effective node immunization strategies to sup- press epidemic/misinformation spreading. It is found that two deep graph generative models produce synthetic networks that closely resemble the structural properties of real-world networks, enabling them to identify effective immunization strategies.

cs.SI↗

Recursive Self-Improvement in Unified Multimodal Models

Unified multimodal models (UMMs) understand and generate both text and images, which lets a model produce its own training data. Existing self-improvement in UMMs keeps supervision on the visual side, where image understanding judges image generation. We propose recursive cross-capability self-improvement (RSI), a training loop in which the text and visual abilities of a UMM supply training data for one another. In each round, the model generates images and reads them to find where it falls short. It then writes programs aimed at these shortcomings, and execution verifies every result against its specification. Verified renders train image generation, while labeled renders and the model's own correct programs train visual understanding and program writing. Program execution thus acts as a source of truth outside the model, so errors do not accumulate across rounds. We study RSI on charts and build BasicChartBench to evaluate open models early in training. On requests worded differently from training, four rounds of RSI raise the score from 45.7% to 60.2%, while continued training stays at 46.3%. Verified construction carries most of the gain, and targeting the model's failures adds 3.5%. Along the way, the share of verified programs rises from 48.9% to 95.2%, and the reader's accuracy on edited renders rises from 55.6% to 87.4%.

cs.CL↗

InfoEdit: Probing Global Layout Reasoning in Infographic Editing

Multimodal foundation models edit natural photographs at production quality, yet the same models struggle with structured visual content such as infographics. Unlike photographs, infographics encode information through logical relations; editing one element often requires surrounding elements to be adapted. We refer to this global layout reasoning capability as reflow. Existing image-editing benchmarks neither provide a dedicated setting for structured visual content nor evaluate the reflow capability. We introduce InfoEdit, a novel benchmark of 1,000 infographics across eight logical-relation families, paired with 4,000 editing instructions across four editing tasks, and a reflow-aware evaluation protocol. Across eight frontier editors, only GPT-Image-2 clears 60% average success rate; most models fall below 7%, and no editor exceeds 36% on the Swap-Block task even with perfect target localization. We further show that code-level editing can match the strongest pixel-level editor, revealing complementary strengths across tasks. InfoEdit identifies reflow as a central challenge in structured visual content editing and provides a diagnostic benchmark to facilitate future progress.

cs.CV↗

Network-based modeling of cocaine trafficking flows and displacement effects

The worldwide cocaine market is undergoing an extraordinary surge. Insights on how cocaine is trafficked from production countries to consumer markets are limited, and often based on observed trafficking routes (e.g., seizures) alone. In this paper, we introduce a network-based model of actual transnational cocaine flows, beyond flows seized. In this model, cocaine is routed through an international transportation network of land and sea connections, where each link is assigned an interception risk metric for containerized transport. This interception risk metric combines features related to law enforcement inspections, containerized trade, and transport connectivity into three key driving components of interception risk using Principal Component Analysis (PCA). The interception risk metric on each link is then constructed from these driving components, where the importance of each component is characterized by a strength parameter. Given a set of strength parameters, cocaine is assumed to be routed from production countries to consumption countries via routes that minimize the total risk of interception, subject to country-specific supply and demand constraints. In the end, the strength parameters are calibrated by comparing the corresponding flow outcomes with known trafficking activity. From these final trafficking flows associated with the calibrated parameters, we are able to identify `blind spots' (e.g., nonzero flows on links with small or none observed trafficking activity) and provide more evidence of the waterbed effect (e.g., changing routes as a result of increased interdiction efforts by law enforcement).

physics.soc-ph↗

K2-V2: A 360-Open, Reasoning-Enhanced LLM

We introduce K2-V2, a 360-open LLM built from scratch as a superior base for reasoning adaptation, in addition to functions such as conversation and knowledge retrieval from general LLMs. It stands as the strongest fully open model, rivals open-weight leaders in its size class, outperforms Qwen2.5-72B and approaches the performance of Qwen3-235B. We actively infuse domain knowledge, reasoning, long-context, and tool use throughout the training process. This explicitly prepares the model for complex reasoning tasks. We demonstrate this potential using simple supervised fine-tuning, establishing a strong baseline that indicates significant headroom for advanced alignment. By releasing the full training history and data composition, we maximize the effectiveness of continuous training, a key open source production scenario. We release the model weights and signature LLM360 artifacts, such as complete training data, to empower the community with a capable, reasoning-centric foundation.

cs.LG↗

UReason: Benchmarking Reasoning-to-Generation Alignment in Unified Multimodal Models

Unified multimodal models (UMMs) aim to integrate multimodal understanding and generation within a unified architecture, yet it remains unclear to what extent textual and visual modalities are aligned. To investigate this question, we use reasoning-guided image generation as a diagnostic task, where models produce textual reasoning first and then generate images. We introduce UReason, a benchmark for evaluating reasoning-to-generation alignment in this paradigm, consisting of 2,000 human-curated and human-verified instances spanning five reasoning-intensive tasks: Code, Arithmetic, Spatial, Attribute, and Text. To enable controlled analysis, we develop an evaluation framework that compares direct generation, reasoning-guided generation, and decontextualized generation, which conditions only on the refined prompt extracted from reasoning. Across eight widely used open-source UMMs, while we find that reasoning-guided generation yields improvements over direct generation, somewhat surprisingly, decontextualized generation consistently outperforms reasoning-guided generation by a large margin. Our further analyses suggest that the intended visual semantics in textual reasoning are not reliably reflected in the generated images, despite their unified design and training. Overall, UReason serves as a practical litmus test for reasoning-to-generation alignment and provides a challenging benchmark for developing next-generation, more tightly aligned UMMs.

cs.CL↗

Fair Influence Maximization in Hypergraphs

The influence maximization problem aims to select a set of seed nodes that maximize the influence, i.e., the average number of influenced nodes, at the end of a spreading process. It has been widely studied with applications in viral marketing, public health campaigns, and social influence. In networks with pronounced community structure, existing approaches often yield an uneven distribution of influenced nodes across communities, which is unfair. Although the fair influence maximization (FIM) problem has been studied for pairwise networks, it remains largely unexplored for hyper graphs, which more accurately represent real-world systems involving group interactions. We introduce FIMH, a heuristic seed-selection algorithm for FIM on hyper graphs, under the Susceptible-Infected Contact Process (SICP) spreading model. FIMH iteratively estimates the contribution of each candidate node to influence and fairness and selects the node that best trades off these two objectives as an additional seed using a parameter-free utopia-distance criterion. Experiments on seven real-world hypergraphs demonstrate that FIMH achieves an influence comparable to that of state-of-the-art IM methods while significantly reducing influence disparity. Analysis of the topological properties of the selected seed nodes and their contributions to influence and fairness further supports the effectiveness of FIMH.

cs.SI↗

EMO: Frustratingly Easy Progressive Training of Extendable MoE

Sparse Mixture-of-Experts (MoE) models offer a powerful way to scale model size without increasing compute, as per-token FLOPs depend only on k active experts rather than the total pool of E experts. Yet, this asymmetry creates an MoE efficiency paradox in practice: adding more experts balloons memory and communication costs, making actual training inefficient. We argue that this bottleneck arises in part because current MoE training allocates too many experts from the beginning, even though early-stage data may not fully utilize such capacity. Motivated by this, we propose EMO, a simple progressive training framework that treats MoE capacity as expandable memory and grows the expert pool over the course of training. EMO explicitly models sparsity in scaling law to derive stage-wise compute-optimal token budgets for progressive expansion. Empirical results show that EMO matches the performance of a fixed-expert setup in large-scale experiments while improving wall-clock efficiency. It offers a surprisingly simple yet effective path to scalable MoE training, preserving the benefits of large expert pools while reducing both training time and GPU cost.

cs.LG↗

GQA-μP: The maximal parameterization update for grouped query attention

Hyperparameter transfer across model architectures dramatically reduces the amount of compute necessary for tuning large language models (LLMs). The maximal update parameterization (μP) ensures transfer through principled mathematical analysis but can be challenging to derive for new model architectures. Building on the spectral feature-learning view of Yang et al. (2023a), we make two advances. First, we promote spectral norm conditions on the weights from a heuristic to the definition of feature learning, and as a consequence arrive at the Complete-P depth and weight-decay scalings without recourse to lazy-learning. Second, we consider a modified spectral norm that preserves the valid scaling law of network weights when weight matrices are not full rank. This enables (to our knowledge, the first) derivation of μP scalings for grouped-query attention (GQA). We demonstrate the efficacy of our theoretical derivations by showing learning rate transfer across the GQA repetition hyperparameter as well as experiments regarding transfer over weight decay.

cs.LG↗

Asymmetric Idiosyncrasies in Multimodal Models

In this work, we study idiosyncrasies in the caption models and their downstream impact on text-to-image models. We design a systematic analysis: given either a generated caption or the corresponding image, we train neural networks to predict the originating caption model. Our results show that text classification yields very high accuracy (99.70\%), indicating that captioning models embed distinctive stylistic signatures. In contrast, these signatures largely disappear in the generated images, with classification accuracy dropping to at most 50\% even for the state-of-the-art Flux model. To better understand this cross-modal discrepancy, we further analyze the data and find that the generated images fail to preserve key variations present in captions, such as differences in the level of detail, emphasis on color and texture, and the distribution of objects within a scene. Overall, our classification-based framework provides a novel methodology for quantifying both the stylistic idiosyncrasies of caption models and the prompt-following ability of text-to-image systems.

cs.CV↗

TimeWak: Temporal Chained-Hashing Watermark for Time Series Data

Synthetic time series generated by diffusion models enable sharing privacy-sensitive datasets, such as patients' functional MRI records. Key criteria for synthetic data include high data utility and traceability to verify the data source. Recent watermarking methods embed in homogeneous latent spaces, but state-of-the-art time series generators operate in data space, making latent-based watermarking incompatible. This creates the challenge of watermarking directly in data space while handling feature heterogeneity and temporal dependencies. We propose TimeWak, the first watermarking algorithm for multivariate time series diffusion models. To handle temporal dependence and spatial heterogeneity, TimeWak embeds a temporal chained-hashing watermark directly within the temporal-feature data space. The other unique feature is the $ε$-exact inversion, which addresses the non-uniform reconstruction error distribution across features from inverting the diffusion process to detect watermarks. We derive the error bound of inverting multivariate time series while preserving robust watermark detectability. We extensively evaluate TimeWak on its impact on synthetic data quality, watermark detectability, and robustness under various post-editing attacks, against five datasets and baselines of different temporal lengths. Our results show that TimeWak achieves improvements of 61.96% in context-FID score, and 8.44% in correlational scores against the strongest state-of-the-art baseline, while remaining consistently detectable.

cs.CR↗

Estimating Nodal Spreading Influence Using Partial Temporal Network

Temporal networks, whose links are activated or deactivated over time, are used to represent complex systems such as social interactions or collaborations occurring at specific times. Such networks facilitate the spread of information and epidemics. The average number of nodes infected via a spreading process on a network starting from a single seed node over a given period is called the influence of that node. In this paper, we address the question of how to utilize the partially observed temporal network (local and of short duration) around each node, to estimate the ranking of nodes in spreading influence on the full network over a long period. This is essential for target marketing and epidemic/misinformation mitigation where only partial network information is possibly accessible. This would also enable us to understand which network properties of a node, observed locally and shortly after the start of the spreading process, determine its influence. We systematically propose a set of nodal centrality metrics based on partial temporal network information, encoding diverse properties of (time-respecting) walks. It is found that distinct centrality metrics perform the best in estimating nodal influence depending on the infection probability of the spreading process. For a broad range of the infection probability, a node tends to be influential if it can reach many distinct nodes via time-respecting walks and if these nodes can be reached early in time. We find and explain why the proposed metrics generally outperform classic centrality metrics derived from both full and partial temporal networks.

cs.SI↗

Diffusion backbone of temporal higher-order networks

Temporal higher-order networks, where each hyperlink involving a group of nodes are activated or deactivated over time, are recently used to represent complex systems such as social contacts, interactions or collaborations that occur at specific times. Such networks are substrates for social contagion processes like the diffusion of information and opinions. In this work, we consider eight temporal higher-order networks derived from human face-to-face interactions in various contexts and the Susceptible-Infected threshold process on each of these networks: whenever a hyperlink is active and the number of infected nodes in the hyperlink exceeds a threshold $Θ$, each susceptible node in the hyperlink is infected independently with probability $β$. The objective is to understand (1) the contribution of each hyperlink to the diffusion process, namely, the average number of nodes that are infected directly via the activation of the hyperlink when the diffusion starts from an arbitrary seed node, and (2) hyperlinks with what network properties tend to contribute more. We first propose to construct the diffusion backbone. The backbone is a weighted higher-order network, where the weight of each hyperlink denotes the contribution of the hyperlink to a given diffusion process. Secondly, we find that the backbone, or the contribution of hyperlinks, is dependent on the parameters $β$ and $Θ$ of the diffusion process, which is also supported by our theoretical analysis of the backbone when $β\rightarrow 0$. Thirdly, we systematically design centrality metrics for hyperlinks in a temporal higher-order network, and each centrality metric is used to estimate the ranking of hyperlinks by the weight in the backbone. Finally, we find and explain why different centrality metrics can better estimate the contributions of hyperlinks for different parameters of the diffusion process.

physics.soc-ph↗

A constant potential reactor framework for electrochemical reaction simulations

Understanding the evolution of electrified solid-liquid interfaces during electrochemical reactions is crucial. However, capturing the dynamic behavior of the interfaces with high temporal resolution and accuracy over long timescales remains a major challenge for both experimental and computational techniques. Here, we present a constant potential reactor framework that enables the simulation of electrochemical reactions with ab initio accuracy over extended timescales, allowing for real-time atomic scale observations for the electrified solid-liquid interface evolution. By implementing an enhanced sampling active learning protocol, we develop fast, accurate, and scalable neural network potentials that generalize across systems with varying electron counts, based on high-throughput density functional theory computations within an explicit-implicit hybrid solvent model. The simulation of reactions in realistic electrochemical environments uncovers the intrinsic mechanisms through which alkali metal cations promote CO2 adsorption and suppress the hydrogen evolution reaction. These findings align with previous experimental results and clarify previously elusive observations, offering valuable computational insights. Our framework lay the groundwork for future studies exploring the dynamic interplay between interfacial structure and reactivity in electrochemical environments.

physics.chem-ph↗

BCIM: Budget and capacity constrained influence maximization in multilayer networks

Influence maximization (IM) seeks to identify a seed set that maximizes influence within a network, with applications in areas such as viral marketing, disease control, and political campaigns. The budgeted influence maximization (BIM) problem extends IM by incorporating cost constraints for different nodes. However, the current BIM problem, limited by budget alone, often results in the selection of numerous low-cost nodes, which may not be applicable to real-world scenarios. Moreover, considering that users can transmit information across multiple social platforms, solving the BIM problem across these platforms could lead to more optimized resource utilization. To address these challenges, we propose the Budget and Capacity Constrained Influence Maximization (BCIM) problem within multilayer networks and introduce a Multilayer Multi-population Genetic Algorithm (MMGA) to solve it. The MMGA employs modules, such as initialization, repair, and parallel evolution, designed not only to meet budget and capacity constraints but also to significantly enhance algorithmic efficiency. Extensive experiments on both synthetic and empirical multilayer networks demonstrate that MMGA improves spreading performance by at least 10% under the two constraints compared to baselines extended from classical IM problems. The BCIM framework introduces a novel direction in influence maximization, providing an effective and efficient solution to the problem.

cs.SI↗