arXiv subjects

Haoran Zhou

Publications and source records attributed to Haoran Zhou.

At least 19 recordsLinked to original sources

arXiv2026

Earnings25: A Comprehensive 500-Hour Speech Benchmark for Finance

We introduce Earnings25, a finance-domain benchmark for evaluating automatic speech recognition (ASR) on English-language earnings calls under realistic conditions. Earnings25 comprises two complementary test sets: (i) testset-full, 498 hours of full English-language S&P 500 earnings calls from Q4 2025, and (ii) testset-segmented, a 46-hour industry-balanced set of 290 segments sampled from English-language U.S. earnings calls in 2025. The benchmark provides aligned transcripts and structured metadata, including speaker roles, industry labels, and call structure, enabling speaker- and industry-aware evaluation beyond aggregate word error rate (WER). We report reproducible baselines for Whisper and Parakeet-TDT using standardized scoring.

cs.CL↗

arXiv2026

SANEmerg: An Emergent Communication Framework for Semantic-aware Agentic AI Networking

Future networking systems are envisioned to become part of an agentic AI-native ecosystem in which a vast number of heterogeneous and specialized AI agents cooperate seamlessly to fulfill complex user requirements in real time. However, traditional networking paradigms are characterized by a rigid decoupling of communication and computation, which often leads to significant inefficiencies in large-scale agentic AI networking (AgentNet) systems. Emergent communication offers a novel solution by enabling autonomous agents that support task-specific signaling protocols for information exchange and collaborative coordination. In this paper, we consider a multi-agent emergent communication framework, tailored for semantic-aware AgentNet systems in which the user's semantic intent can be automatically detected, inferred, and linked to a set of sub-tasks to be assigned to a set of agents. We investigate how communication and signaling protocols can emerge among collaborative agents with computationally bounded intelligence under stringent bandwidth constraints. Our proposed framework, called SANEmerg, is designed to facilitate the emergence of communication for collaborative task fulfillment while adhering to the physical limits of AgentNet. SANEmerg incorporates a bandwidth-adaptable importance-filter that dynamically prioritizes the transmission of higher-contribution message dimensions, ensuring robust performance in bandwidth-limited environments. Furthermore, SANEmerg integrates a complexity-regularizer grounded in the Minimum Description Length (MDL) principle to facilitate the emergence of computationally bounded signaling. Evaluated via an AgentNet prototype and extensive experimentation, SANEmerg demonstrates significant performance improvements over state-of-the-art solutions, achieving superior task accuracy while significantly reducing bandwidth and computational overhead.

cs.AI↗

arXiv2026

A Discrete-Time Random Feature Method for Nonlinear Evolution Equations with Implicit-Explicit Runge--Kutta Time Stepping

We study a discrete-time random feature method for nonlinear, time-dependent partial differential equations. In contrast to continuous-time formulations that treat time as an additional input variable, the method advances the solution step by step, with each time level computed from previously available states. The spatial solution at each step is represented in the random feature trial space, and the time discretization is given by an implicit-explicit Runge--Kutta (IMEX-RK, 4 stages, third-order) scheme. After splitting the operator into linear and nonlinear parts, each stage admits a linear least-squares formulation, which avoids nonlinear least-squares solves. We also derive a global error estimate for the fully discrete method, separating the contributions of the stage-wise RFM approximation, perturbations in the least-squares coefficients, and the temporal discretization. Numerical experiments for the Allen--Cahn, Burgers, Korteweg--De Vries, and Cahn--Hilliard equations show relative $L^2$-errors of order $10^{-6}$ and convergence rates consistent with the third-order IMEX scheme. A comparison with an IMEX-PINN variant shows that the proposed method achieves higher accuracy at substantially lower computational cost.

math.NA↗

arXiv2026

MotionScale: Reconstructing Appearance, Geometry, and Motion of Dynamic Scenes with Scalable 4D Gaussian Splatting

Realistic reconstruction of dynamic 4D scenes from monocular videos is essential for understanding the physical world. Despite recent progress in neural rendering, existing methods often struggle to recover accurate 3D geometry and temporally consistent motion in complex environments. To address these challenges, we propose MotionScale, a 4D Gaussian Splatting framework that scales efficiently to large scenes and extended sequences while maintaining high-fidelity structural and motion coherence. At the core of our approach is a scalable motion field parameterized by cluster-centric basis transformations that adaptively expand to capture diverse and evolving motion patterns. To ensure robust reconstruction over long durations, we introduce a progressive optimization strategy comprising two decoupled propagation stages: 1) A background extension stage that adapts to newly visible regions, refines camera poses, and explicitly models transient shadows; 2) A foreground propagation stage that enforces motion consistency through a specialized three-stage refinement process. Extensive experiments on challenging real-world benchmarks demonstrate that MotionScale significantly outperforms state-of-the-art methods in both reconstruction quality and temporal stability. Project page: https://hrzhou2.github.io/motion-scale-web/.

cs.CV↗

arXiv2026

Efficient Counting and Simulation in Content-Oblivious Rings

In the content-oblivious (CO) model (proposed by Censor-Hillel et al.), processes inhabit an asynchronous network and communicate only by exchanging pulses. A series of works has clarified the computational power of this model. In particular, it was shown that, when a leader is present and the network is 2-edge-connected, content-oblivious communication can simulate classical asynchronous message passing. Subsequent results extended this equivalence to leaderless oriented and unoriented rings, and, under non-uniform assumptions, to general 2-edge-connected networks. The simulator of Censor-Hillel et al. requires $O(n^3b+n^3\log n)$ pulses to emulate the send of a single $b$-bit message, making it impractical even on modest-size networks. We focus on message-efficient computation in CO networks. We study the fundamental problem of counting in ring topologies, both because knowing the exact network size is a basic prerequisite for many distributed tasks and because counting immediately implies a broad class of aggregation primitives. We give an algorithm that counts using $O(n^{1.5})$ pulses in anonymous rings with a leader, an $O(n\log^2 n)$ algorithm for counting in rings with IDs. Moreover, we show that any counting algorithm in CO requires $\Omega(n\log n)$ pulses. Interestingly, in the course of this investigation, we design a simulator for classic message passing: in one simulated round, each process can send a $b$-bit message to each of its neighbors using only $O(b)$ pulses per process. The simulator extends to general 2-edge-connected networks, after a pre-processing step that requires $O(n^{8}\log n)$ pulses, where $n$ is the number of processes, allowing thus efficient simulation of asynchronous message passing in general 2-edge-connected networks.

cs.DC↗

arXiv2025

SANet: A Semantic-aware Agentic AI Networking Framework for Cross-layer Optimization in 6G

Agentic AI networking (AgentNet) is a novel AI-native networking paradigm in which a large number of specialized AI agents collaborate to perform autonomous decision-making, dynamic environmental adaptation, and complex missions. It has the potential to facilitate real-time network management and optimization functions, including self-configuration, self-optimization, and self-adaptation across diverse and complex environments. This paper proposes SANet, a novel semantic-aware AgentNet architecture for wireless networks that can infer the semantic goal of the user and automatically assign agents associated with different layers of the network to fulfill the inferred goal. Motivated by the fact that AgentNet is a decentralized framework in which collaborating agents may generally have different and even conflicting objectives, we formulate the decentralized optimization of SANet as a multi-agent multi-objective problem, and focus on finding the Pareto-optimal solution for agents with distinct and potentially conflicting objectives. We propose three novel metrics for evaluating SANet. Furthermore, we develop a model partition and sharing (MoPS) framework in which large models, e.g., deep learning models, of different agents can be partitioned into shared and agent-specific parts that are jointly constructed and deployed according to agents' local computational resources. Two decentralized optimization algorithms are proposed. We derive theoretical bounds and prove that there exists a three-way tradeoff among optimization, generalization, and conflicting errors. We develop an open-source RAN and core network-based hardware prototype that implements agents to interact with three different layers of the network. Experimental results show that the proposed framework achieved performance gains of up to 14.61% while requiring only 44.37% of FLOPs required by state-of-the-art algorithms.

cs.AI↗

arXiv2025

Enabling Disaggregated Multi-Stage MLLM Inference via GPU-Internal Scheduling and Resource Sharing

Multimodal large language models (MLLMs) extend LLMs with visual understanding through a three-stage pipeline: multimodal preprocessing, vision encoding, and LLM inference. While these stages enhance capability, they introduce significant system bottlenecks. First, multimodal preprocessing-especially video decoding-often dominates Time-to-First-Token (TTFT). Most systems rely on CPU-based decoding, which severely limits throughput, while existing GPU-based approaches prioritize throughput-oriented parallelism and fail to meet the latency-sensitive requirements of MLLM inference. Second, the vision encoder is a standalone, compute-intensive stage that produces visual embeddings and cannot be co-batched with LLM prefill or decoding. This heterogeneity forces inter-stage blocking and increases token-generation latency. Even when deployed on separate GPUs, these stages underutilize available compute and memory resources, reducing overall utilization and constraining system throughput. To address these challenges, we present FlashCodec and UnifiedServe, two complementary designs that jointly optimize the end-to-end MLLM pipeline. FlashCodec accelerates the multimodal preprocessing stage through collaborative multi-GPU video decoding, reducing decoding latency while preserving high throughput. UnifiedServe optimizes the vision-to-text and inference stages using a logically decoupled their execution to eliminate inter-stage blocking, yet physically sharing GPU resources to maximize GPU system utilization. By carefully orchestrating execution across stages and minimizing interference, UnifiedServe Together, our proposed framework forms an end-to-end optimized stack that can serve up to 3.0$\times$ more requests or enforce 1.5$\times$ tighter SLOs, while achieving up to 4.4$\times$ higher throughput compared to state-of-the-art systems.

cs.DC↗

arXiv2025

Motion4D: Learning 3D-Consistent Motion and Semantics for 4D Scene Understanding

Recent advancements in foundation models for 2D vision have substantially improved the analysis of dynamic scenes from monocular videos. However, despite their strong generalization capabilities, these models often lack 3D consistency, a fundamental requirement for understanding scene geometry and motion, thereby causing severe spatial misalignment and temporal flickering in complex 3D environments. In this paper, we present Motion4D, a novel framework that addresses these challenges by integrating 2D priors from foundation models into a unified 4D Gaussian Splatting representation. Our method features a two-part iterative optimization framework: 1) Sequential optimization, which updates motion and semantic fields in consecutive stages to maintain local consistency, and 2) Global optimization, which jointly refines all attributes for long-term coherence. To enhance motion accuracy, we introduce a 3D confidence map that dynamically adjusts the motion priors, and an adaptive resampling process that inserts new Gaussians into under-represented regions based on per-pixel RGB and semantic errors. Furthermore, we enhance semantic coherence through an iterative refinement process that resolves semantic inconsistencies by alternately optimizing the semantic fields and updating prompts of SAM2. Extensive evaluations demonstrate that our Motion4D significantly outperforms both 2D foundation models and existing 3D-based approaches across diverse scene understanding tasks, including point-based tracking, video object segmentation, and novel view synthesis. Our code is available at https://hrzhou2.github.io/motion4d-web/.

cs.CV↗

arXiv2025

Beyond 2-Edge-Connectivity: Algorithms and Impossibility for Content-Oblivious Leader Election

The content-oblivious model, introduced by Censor-Hillel, Cohen, Gelles, and Sel (PODC 2022; Distributed Computing 2023), captures an extremely weak form of communication where nodes can only send asynchronous, content-less pulses. Censor-Hillel, Cohen, Gelles, and Sel showed that no non-constant function $f(x,y)$ can be computed correctly by two parties using content-oblivious communication over a single edge, where one party holds $x$ and the other holds $y$. This seemingly ruled out many natural graph problems on non-2-edge-connected graphs. In this work, we show that, with the knowledge of network topology $G$, leader election is possible in a wide range of graphs. Impossibility: Graphs symmetric about an edge admit no randomized terminating leader election algorithm, even when nodes have unique identifiers and full knowledge of $G$. Leader election algorithms: Trees that are not symmetric about any edge admit a quiescently terminating leader election algorithm with topology knowledge, even in anonymous networks, using $O(n^2)$ messages, where $n$ is the number of nodes. Moreover, even-diameter trees admit a terminating leader election given only the knowledge of the network diameter $D = 2r$, with message complexity $O(nr)$. Necessity of topology knowledge: In the family of graphs $\mathcal{G} = \{P_3, P_5\}$, both the 3-path $P_3$ and the 5-path $P_5$ admit a quiescently terminating leader election if nodes know the topology exactly. However, if nodes only know that the underlying topology belongs to $\mathcal{G}$, then terminating leader election is impossible.

cs.DC↗

arXiv2025

TacRefineNet: Goal-Conditioned Tactile Grasp Refinement for Edge-Prominent Objects

Accurate final grasp alignment remains challenging for edge-prominent objects such as thin plates, discs, and rods, whose sparse contacts are easily occluded and poorly resolved by depth sensing. We present TacRefineNet, a tactile-only, goal-conditioned framework for local refinement along tactilely observable pose dimensions. Given current and target multi-finger tactile images and their corresponding hand-joint configurations, a Siamese policy network directly predicts corrective wrist pose increments. The hand iteratively opens, moves, and regrasps, forming an external-dexterity tactile servoing loop. Cross-combination training pairs current and target samples, allowing targets within the sampled pose range to be specified without retraining. We collect 156,007 simulated samples from 15 plates, discs, and rods and train the policy entirely in MuJoCo before zero-shot deployment to an 11-DoF five-fingered hand with piezoresistive sensors. On seen objects, the real system achieves 80.7\% and 59.3\% success under the $10^\circ$/10\,mm criterion for fixed and random targets, respectively; after five steps, the mean errors are approximately 5.2\,mm and $3.5^\circ$. Experiments further show continuous correction under long-horizon perturbations and limited within-category transfer to unseen objects, with reduced performance for symmetric or weakly discriminative contacts. Project website is available at https://sites.google.com/view/tacrefinenet

cs.RO↗

arXiv2025

Non-Uniform Content-Oblivious Leader Election on Oriented Asynchronous Rings

We study the leader election problem in oriented ring networks under content-oblivious asynchronous message-passing systems, where an adversary may arbitrarily corrupt message contents. Frei et al. (DISC 2024) presented a uniform terminating leader election algorithm for oriented rings in this setting, with message complexity $O(n \cdot \mathsf{ID}_{\max})$ on a ring of size $n$, where $\mathsf{ID}_{\max}$ is the largest identifier in the system, this result has been recently extended by Chalopin et al. (DISC 2025) to unoriented rings. In this paper, we investigate the message complexity of leader election on ring networks in the content-oblivious model, showing that no uniform algorithm can solve the problem if each process is limited to sending a constant number of messages in one direction. Interestingly, this limitation hinges on the uniformity assumption. In the non-uniform setting, where processes know an upper bound $U \geq n$ on the ring size, we present an algorithm with message complexity $O(n \cdot U \cdot \mathsf{ID}_{\min})$, in which each process sends $O(U \cdot \mathsf{ID}_{\min})$ messages clockwise and only three messages counter-clockwise. Here, $\mathsf{ID}_{\min}$ is the smallest identifier in the system. This dependence on the identifiers compares favorably with the dependence on $\mathsf{ID}_{\max}$ of Frei et al. We also show a non-uniform algorithm where each process sends $O(U \cdot \log\mathsf{ID}_{\min})$ messages in one direction and $O(\log\mathsf{ID}_{\min})$ in the other. The factor $\log \mathsf{ID}_{\min}$ is optimal, matching the lower bound of Frei et al. Finally, in the anonymous setting, where processes do not have identifiers, we propose a randomized algorithm where each process sends only $O(\log^2 U)$ messages, with a success probability of $1 - U^{-c}$.

cs.DC↗

arXiv2025

DyDexHandover: Human-like Bimanual Dynamic Dexterous Handover using RGB-only Perception

Dynamic in air handover is a fundamental challenge for dual-arm robots, requiring accurate perception, precise coordination, and natural motion. Prior methods often rely on dynamics models, strong priors, or depth sensing, limiting generalization and naturalness. We present DyDexHandover, a novel framework that employs multi-agent reinforcement learning to train an end to end RGB based policy for bimanual object throwing and catching. To achieve more human-like behavior, the throwing policy is guided by a human policy regularization scheme, encouraging fluid and natural motion, and enhancing the generalization capability of the policy. A dual arm simulation environment was built in Isaac Sim for experimental evaluation. DyDexHandover achieves nearly 99 percent success on training objects and 75 percent on unseen objects, while generating human-like throwing and catching behaviors. To our knowledge, it is the first method to realize dual-arm in-air handover using only raw RGB perception.

cs.RO↗

arXiv2025

Extracting Visual Facts from Intermediate Layers for Mitigating Hallucinations in Multimodal Large Language Models

Multimodal Large Language Models (MLLMs) have made significant strides by combining visual recognition and language understanding to generate content that is both coherent and contextually accurate. However, MLLMs continue to struggle with object hallucinations, where models produce seemingly plausible but factually incorrect outputs, including objects that do not exist in the image. Recent work has revealed that the prior knowledge in MLLMs significantly suppresses visual information in deep layers, causing hallucinatory outputs. However, how these priors suppress visual information at the intermediate layer stage in MLLMs remains unclear. We observe that visual factual knowledge and the differences between intermediate-layer prior/original probability distributions show similar evolutionary trends in intermediate layers. Motivated by this, we introduce Decoding by Extracting Visual Facts (EVA), a simple, training-free method that dynamically selects intermediate layers with the most significant visual factual information. By contrasting the output distributions of the selected layer derived from the original input and pure-text input, EVA extracts visual factual knowledge and proportionally incorporates it into the final layer to correct the output logits. Importantly, EVA is model-agnostic, seamlessly integrates with various classic decoding strategies, and is applicable across different MLLMs. We validate EVA on widely-used benchmarks, and the results show that it significantly reduces hallucination rates compared to baseline methods, underscoring its effectiveness in mitigating hallucinations.

cs.CV↗

arXiv2025

Content-Oblivious Leader Election in 2-Edge-Connected Networks

Censor-Hillel, Cohen, Gelles, and Sela (PODC 2022 & Distributed Computing 2023) studied fully-defective asynchronous networks, where communication channels may suffer an extreme form of alteration errors, rendering messages completely corrupted. The model is equivalent to content-oblivious computation, where nodes communicate solely via pulses. They showed that if the network is 2-edge-connected, then any algorithm for a noiseless setting can be simulated in the fully-defective setting; otherwise, no non-trivial computation is possible in the fully-defective setting. However, their simulation requires a predesignated leader, which they conjectured to be necessary for any non-trivial content-oblivious task. In this work, we present two results: General 2-edge-connected topologies: First, we show an asynchronous content-oblivious leader election algorithm that quiescently terminates in any 2-edge-connected network with message complexity $O(m \cdot N \cdot \mathsf{ID}_{\min})$, where $m$ is the number of edges, $N$ is a known upper bound on the number of nodes, and $\mathsf{ID}_{\min}$ is the smallest $\mathsf{ID}$. Combined with the above simulation, this result shows that whenever a size bound $N$ is known, any noiseless algorithm can be simulated in the fully-defective model without a preselected leader, fully refuting the conjecture. Unoriented rings: We then show that the knowledge of $N$ can be dropped in unoriented ring topologies by presenting a quiescently terminating election algorithm with message complexity $O(n \cdot \mathsf{ID}_{\max})$ that matches the previous bound. Consequently, this result constitutes a strict improvement over the previous leader election in oriented rings by Frei, Gelles, Ghazy, and Nolin (DISC 2024) and shows that, on rings, fully-defective and noiseless communication are computationally equivalent, with no additional assumptions.

cs.DC↗

arXiv2025

Adapting Whisper for Streaming Speech Recognition via Two-Pass Decoding

OpenAI Whisper is a family of robust Automatic Speech Recognition (ASR) models trained on 680,000 hours of audio. However, its encoder-decoder architecture, trained with a sequence-to-sequence objective, lacks native support for streaming ASR. In this paper, we fine-tune Whisper for streaming ASR using the WeNet toolkit by adopting a Unified Two-pass (U2) structure. We introduce an additional Connectionist Temporal Classification (CTC) decoder trained with causal attention masks to generate streaming partial transcripts, while the original Whisper decoder reranks these partial outputs. Our experiments on LibriSpeech and an earnings call dataset demonstrate that, with adequate fine-tuning data, Whisper can be adapted into a capable streaming ASR model. We also introduce a hybrid tokenizer approach, which uses a smaller token space for the CTC decoder while retaining Whisper's original token space for the attention decoder, resulting in improved data efficiency and generalization.

cs.SD↗

arXiv2025

SANNet: A Semantic-Aware Agentic AI Networking Framework for Multi-Agent Cross-Layer Coordination

Agentic AI networking (AgentNet) is a novel AI-native networking paradigm that relies on a large number of specialized AI agents to collaborate and coordinate for autonomous decision-making, dynamic environmental adaptation, and complex goal achievement. It has the potential to facilitate real-time network management alongside capabilities for self-configuration, self-optimization, and self-adaptation across diverse and complex networking environments, laying the foundation for fully autonomous networking systems in the future. Despite its promise, AgentNet is still in the early stage of development, and there still lacks an effective networking framework to support automatic goal discovery and multi-agent self-orchestration and task assignment. This paper proposes SANNet, a novel semantic-aware agentic AI networking architecture that can infer the semantic goal of the user and automatically assign agents associated with different layers of a mobile system to fulfill the inferred goal. Motivated by the fact that one of the major challenges in AgentNet is that different agents may have different and even conflicting objectives when collaborating for certain goals, we introduce a dynamic weighting-based conflict-resolving mechanism to address this issue. We prove that SANNet can provide theoretical guarantee in both conflict-resolving and model generalization performance for multi-agent collaboration in dynamic environment. We develop a hardware prototype of SANNet based on the open RAN and 5GS core platform. Our experimental results show that SANNet can significantly improve the performance of multi-agent networking systems, even when agents with conflicting objectives are selected to collaborate for the same goal.

cs.AI↗

arXiv2024

CRA-PCN: Point Cloud Completion with Intra- and Inter-level Cross-Resolution Transformers

Point cloud completion is an indispensable task for recovering complete point clouds due to incompleteness caused by occlusion, limited sensor resolution, etc. The family of coarse-to-fine generation architectures has recently exhibited great success in point cloud completion and gradually became mainstream. In this work, we unveil one of the key ingredients behind these methods: meticulously devised feature extraction operations with explicit cross-resolution aggregation. We present Cross-Resolution Transformer that efficiently performs cross-resolution aggregation with local attention mechanisms. With the help of our recursive designs, the proposed operation can capture more scales of features than common aggregation operations, which is beneficial for capturing fine geometric characteristics. While prior methodologies have ventured into various manifestations of inter-level cross-resolution aggregation, the effectiveness of intra-level one and their combination has not been analyzed. With unified designs, Cross-Resolution Transformer can perform intra- or inter-level cross-resolution aggregation by switching inputs. We integrate two forms of Cross-Resolution Transformers into one up-sampling block for point generation, and following the coarse-to-fine manner, we construct CRA-PCN to incrementally predict complete shapes with stacked up-sampling blocks. Extensive experiments demonstrate that our method outperforms state-of-the-art methods by a large margin on several widely used benchmarks. Codes are available at https://github.com/EasyRy/CRA-PCN.

cs.CV↗

arXiv2023

Dissecting RGB-D Learning for Improved Multi-modal Fusion

In the RGB-D vision community, extensive research has been focused on designing multi-modal learning strategies and fusion structures. However, the complementary and fusion mechanisms in RGB-D models remain a black box. In this paper, we present an analytical framework and a novel score to dissect the RGB-D vision community. Our approach involves measuring proposed semantic variance and feature similarity across modalities and levels, conducting visual and quantitative analyzes on multi-modal learning through comprehensive experiments. Specifically, we investigate the consistency and specialty of features across modalities, evolution rules within each modality, and the collaboration logic used when optimizing a RGB-D model. Our studies reveal/verify several important findings, such as the discrepancy in cross-modal features and the hybrid multi-modal cooperation rule, which highlights consistency and specialty simultaneously for complementary inference. We also showcase the versatility of the proposed RGB-D dissection method and introduce a straightforward fusion strategy based on our findings, which delivers significant enhancements across various tasks and even other multi-modal data.

cs.CV↗