arXiv subjects

Kangyu Wang

Publications and source records attributed to Kangyu Wang.

11 recordsLinked to original sources

arXiv2026

Asymmetric physics enables efficient learning in quadrupedal robot swarms

Animal collectives navigate cluttered environments through local coordination, yet robot swarms still struggle to reproduce this capability in the physical world. End-to-end learning offers a route to such coordination, but scaling it to embodied swarms remains difficult: standard sampling-based reinforcement learning becomes inefficient when visual perception, dense robot-robot interaction, and contact-rich locomotion must be learned together. Here we show that asymmetric physics enables efficient end-to-end learning of vision-based, decentralized control in large swarms of quadrupedal robots. During training, quadrupeds interact in shared environments, where a high-fidelity, non-differentiable simulator generates realistic motion and contact dynamics, and differentiable surrogate models provide gradients for navigation and locomotion policies. This separation enables up to 512 quadrupeds to learn coordinated navigation policies in obstacle-rich environments. At deployment, each robot acts from a single forward-facing depth camera, without explicit communication, centralized planning, or global maps. The policies generalize across forests, bridges, enclosures, narrow passages, and mazes, and zero-shot transfer to six physical quadrupeds across five real-world scenarios. The resulting swarms exhibit predictive avoidance, right-side yielding, pausing before bottlenecks, and wall following, showing that asymmetric physics enables efficient training of scalable decentralized control policies for quadrupedal robot swarms.

cs.RO↗

arXiv2026

VidLaDA: Bidirectional Diffusion Large Language Models for Efficient Video Understanding

Current Video Large Language Models (Video LLMs) typically encode frames via a vision encoder and employ an autoregressive (AR) LLM for understanding and generation. However, this AR paradigm inevitably faces a dual efficiency bottleneck: strictly unidirectional attention compromises understanding efficiency by hindering global spatiotemporal aggregation, while serial decoding restricts generation efficiency. To address this, we propose VidLaDA, a Video LLM based on Diffusion Language Models (DLMs) that leverages bidirectional attention to unlock comprehensive spatiotemporal modeling and decode tokens in parallel. To further mitigate the computational overhead of diffusion decoding, we introduce MARS-Cache, an acceleration strategy that prunes redundancy by combining asynchronous visual cache refreshing with frame-wise chunk attention. Experiments show VidLaDA rivals state-of-the-art AR baselines (e.g., Qwen2.5-VL and LLaVA-Video) and outperforms DLM baselines, with MARS-Cache delivering over 12x speedup without compromising accuracy. Code and checkpoints are open-sourced at https://github.com/ziHoHe/VidLaDA.

cs.CV↗

arXiv2025

Let It Flow: Agentic Crafting on Rock and Roll, Building the ROME Model within an Open Agentic Learning Ecosystem

Agentic crafting requires LLMs to operate in real-world environments over multiple turns by taking actions, observing outcomes, and iteratively refining artifacts. Despite its importance, the open-source community lacks a principled, end-to-end ecosystem to streamline agent development. We introduce the Agentic Learning Ecosystem (ALE), a foundational infrastructure that optimizes the production pipeline for agentic model. ALE consists of three components: ROLL, a post-training framework for weight optimization; ROCK, a sandbox environment manager for trajectory generation; and iFlow CLI, an agent framework for efficient context engineering. We release ROME, an open-source agent grounded by ALE and trained on over one million trajectories. Our approach includes data composition protocols for synthesizing complex behaviors and a novel policy optimization algorithm, Interaction-Perceptive Agentic Policy Optimization (IPA), which assigns credit over semantic interaction chunks rather than individual tokens to improve long-horizon training stability. Empirically, we evaluate ROME within a structured setting and introduce Terminal Bench Pro, a benchmark with improved scale and contamination control. ROME demonstrates strong performance across benchmarks like SWE-bench Verified and Terminal Bench, proving the effectiveness of ALE.

cs.AI↗

arXiv2025

dInfer: An Efficient Inference Framework for Diffusion Language Models

Diffusion-based large language models (dLLMs) have emerged as a promising alternative to autoregressive (AR) LLMs, leveraging denoising-based generation to enable inherent parallelism. Even more and more open-sourced dLLM models emerge, yet their widespread adoption remains constrained by the lack of a standardized and efficient inference framework. We present dInfer, an efficient and extensible framework for dLLM inference. dInfer decomposes the inference pipeline into four modular components--model, diffusion iteration manager, decoding strategy, and KV-cache manager--and integrates novel algorithms for each component alongside system-level optimizations. Through this combination of algorithmic innovations and system enhancements, dInfer achieves substantial efficiency gains without compromising output quality on LLaDA-MoE. At batch size 1, it surpasses 1,100 tokens per second on HumanEval and averages over 800 tokens per second across six benchmarks on $8\times$ H800 GPUs. Compared to prior systems, dInfer delivers a $10\times$ speedup over Fast-dLLM while maintaining similar model performance. Even compared to the AR model (with a comparable number of activation parameters and performance) QWen2.5-3B, which is highly optimized with the latest vLLM inference engine, dInfer still delivers a $2$-$3\times$ speedup. The implementation of dInfer is open-sourced at https://github.com/inclusionAI/dInfer.

cs.CL↗

arXiv2025

CreditDecoding: Accelerating Parallel Decoding in Diffusion Large Language Models with Trace Credit

Diffusion large language models (dLLMs) generate text through iterative denoising. In commonly adopted parallel decoding schemes, each step confirms only high-confidence positions while remasking the others. By analyzing dLLM denoising traces, we uncover a key inefficiency: models often predict the correct target token several steps before its confidence becomes high enough to be decoded. This gap between early prediction and late decoding forces repeated remasking of already-correct tokens, causing redundant iterations and limiting acceleration. To exploit this temporal redundancy, we introduce Trace Credit to quantify a token's decoding potential by accumulating historical evidence. Building on this, we propose CreditDecoding, a training-free parallel decoding method that fuses Trace Credit with current logits to boost the confidence of correct but underconfident tokens, thereby accelerating denoising and improving robustness. On eight benchmarks, CreditDecoding achieves up to 5.48 times speedup with +0.48 accuracy on LLaDA-8B and consistently improves performance across diverse dLLM architectures and parameter scales. It further scales to long contexts and remains orthogonal to mainstream inference optimizations, making it a practical and widely applicable solution.

cs.CL↗

arXiv2025

LLaDA-MoE: A Sparse MoE Diffusion Language Model

We introduce LLaDA-MoE, a large language diffusion model with the Mixture-of-Experts (MoE) architecture, trained from scratch on approximately 20T tokens. LLaDA-MoE achieves competitive performance with significantly reduced computational overhead by maintaining a 7B-parameter capacity while activating only 1.4B parameters during inference. Our empirical evaluation reveals that LLaDA-MoE achieves state-of-the-art performance among diffusion language models with larger parameters, surpassing previous diffusion language models LLaDA, LLaDA 1.5, and Dream across multiple benchmarks. The instruct-tuned model LLaDA-MoE-7B-A1B-Instruct demonstrates capabilities comparable to Qwen2.5-3B-Instruct in knowledge understanding, code generation, mathematical reasoning, agent and alignment tasks, despite using fewer active parameters. Our results show that integrating a sparse MoE architecture into the training objective of masked diffusion language models still brings out MoE's strengths under efficient inference with few active parameters, and opens ample room for further exploration of diffusion language models. LLaDA-MoE models are available at Huggingface.

cs.CL↗

arXiv2025

Congruence families modulo powers of $7$ for $4$-colored generalized Frobenius partitions

In 2012, Peter Paule and Cristian-Silviu Radu proved an infinite family of Ramanujan type congruences for $2$-colored Frobenius partitions $c\phi_2$ introduced by George Andrews. Recently, Frank Garvan, James Sellers and Nicolas Smoot showed that this family of congruences is equivalent to the family of congruences for $(2,0)$-colored Frobenius partitions $c\psi_{2,0}$ introduced by Brian Drake and by Yuze Jiang, Larry Rolen and Michael Woodbury for the general case. Motivated by Garvan, Sellers and Smoot's work, Rong Chen and Xiao-Jie Zhu found modular transformations relating the $c\psi_{k,\beta}$ for fixed $k$ and varying $\beta$. As an example, they proved a family of congruences for $c\psi_{3,1/2}$ following Paule and Radu's work and then proved the equivalence between $c\psi_{3,1/2}$ and $c\phi_3=c\psi_{3,3/2}$. In the present paper, we give a new example of Chen and Zhu's framework for $c\psi_{4,\beta}$. Our proof is considerably simpler.

math.CO↗

arXiv2025

Inclusion Arena: An Open Platform for Evaluating Large Foundation Models with Real-World Apps

Large Language Models (LLMs) and Multimodal Large Language Models (MLLMs) have ushered in a new era of AI capabilities, demonstrating near-human-level performance across diverse scenarios. While numerous benchmarks (e.g., MMLU) and leaderboards (e.g., Chatbot Arena) have been proposed to help evolve the development of LLMs and MLLMs, most rely on static datasets or crowdsourced general-domain prompts, often falling short of reflecting performance in real-world applications. To bridge this critical gap, we present Inclusion Arena, a live leaderboard that ranks models based on human feedback collected directly from AI-powered applications. Our platform integrates pairwise model comparisons into natural user interactions, ensuring evaluations reflect practical usage scenarios. For robust model ranking, we employ the Bradley-Terry model augmented with two key innovations: (1) Placement Matches, a cold-start mechanism to quickly estimate initial ratings for newly integrated models, and (2) Proximity Sampling, an intelligent comparison strategy that prioritizes battles between models of similar capabilities to maximize information gain and enhance rating stability. Extensive empirical analyses and simulations demonstrate that Inclusion Arena yields reliable and stable rankings, exhibits higher data transitivity compared to general crowdsourced datasets, and significantly mitigates the risk of malicious manipulation. By fostering an open alliance between foundation models and real-world applications, Inclusion Arena aims to accelerate the development of LLMs and MLLMs truly optimized for practical, user-centric deployments. The platform is publicly accessible at https://www.tbox.cn/about/model-ranking.

cs.AI↗

arXiv2025

CogStream: Context-guided Streaming Video Question Answering

Despite advancements in Video Large Language Models (Vid-LLMs) improving multimodal understanding, challenges persist in streaming video reasoning due to its reliance on contextual information. Existing paradigms feed all available historical contextual information into Vid-LLMs, resulting in a significant computational burden for visual data processing. Furthermore, the inclusion of irrelevant context distracts models from key details. This paper introduces a challenging task called Context-guided Streaming Video Reasoning (CogStream), which simulates real-world streaming video scenarios, requiring models to identify the most relevant historical contextual information to deduce answers for questions about the current stream. To support CogStream, we present a densely annotated dataset featuring extensive and hierarchical question-answer pairs, generated by a semi-automatic pipeline. Additionally, we present CogReasoner as a baseline model. It effectively tackles this task by leveraging visual stream compression and historical dialogue retrieval. Extensive experiments prove the effectiveness of this method.

cs.CV↗

arXiv2025

AI Agents and Hard Choices

Can AI agents deal with hard choices -- cases where options are incommensurable because multiple objectives are pursued simultaneously? Adopting a technologically engaged approach distinct from existing philosophical literature, I submit that the fundamental design of current AI agents as optimisers creates two limitations: the Identification Problem and the Resolution Problem. First, I demonstrate that agents relying on Multi-Objective Optimisation (MOO) are structurally unable to identify incommensurability. This inability generates three specific alignment problems: the blockage problem, the untrustworthiness problem, and the unreliability problem. I argue that standard mitigations, such as Human-in-the-Loop, are insufficient for many decision environments. As a constructive alternative, I conceptually explore an ensemble solution. Second, I argue that even if the Identification Problem is solved, AI agents face the Resolution Problem: they lack the autonomy to resolve hard choices rather than arbitrarily picking through self-modification of objectives. I conclude by examining the opaque normative trade-offs involved in granting AI this level of autonomy.

cs.AI↗

arXiv2024

Engineering band structures of two-dimensional materials with remote moire ferroelectricity

The stacking order and twist angle provide abundant opportunities for engineering band structures of two-dimensional materials, including the formation of moire bands, flat bands, and topologically nontrivial bands. The inversion symmetry breaking in rhombohedral-stacked transitional metal dichalcogenides (TMDCs) endows them with an interfacial ferroelectricity associated with an out-of-plane electric polarization. By utilizing twist angle as a knob to construct rhombohedral-stacked TMDCs, antiferroelectric domain networks with alternating out-of-plane polarization can be generated. Here, we demonstrate that such spatially periodic ferroelectric polarizations in parallel-stacked twisted WSe2 can imprint their moire potential onto a remote bilayer graphene. This remote moire potential gives rise to pronounced satellite resistance peaks besides the charge-neutrality point in graphene, which are tunable by the twist angle of WSe2. Our observations of ferroelectric hysteresis at finite displacement fields suggest the moire is delivered by a long-range electrostatic potential. The constructed superlattices by moire ferroelectricity represent a highly flexible approach, as they involve the separation of the moire construction layer from the electronic transport layer. This remote moire is identified as a weak potential and can coexist with conventional moire. Our results offer a comprehensive strategy for engineering band structures and properties of two-dimensional materials by utilizing moire ferroelectricity.

cond-mat.mtrl-sci↗