arXiv subjects

Wangzi Yao

Publications and source records attributed to Wangzi Yao.

3 recordsLinked to original sources

arXiv2026

Motif-Mamba: network motif improved mamba for long-range sequence modeling

Efficient long-sequence modeling remains a central challenge for large language models, as self-attention scales quadratically with sequence length. Mamba offers a linear-time alternative through selective state space recurrence, but its predominantly diagonal state transitions restrict explicit interactions among state dimensions. We propose Motif-Mamba, a structured state space model that augments Mamba with a motif-constrained low-rank recurrent pathway. Inspired by the dynamics of three-node network motifs, the proposed pathway projects hidden states into a compact dynamical subspace, imposes motif-guided interactions, and maps the resulting dynamics back to the original state space. This design enhances cross-dimensional communication while preserving the linear-time recurrent structure of Mamba. Experiments on long-sequence extrapolation, language modeling benchmarks, and brain--computer interface decoding show consistent improvements over Mamba backbones, suggesting that motif-guided low-rank dynamics provide an effective structural prior for long-range sequence modeling.

cs.AI↗

arXiv2026

VISTA: Triplet-Supervised Video Style Transfer with Diffusion Transformers

Video style transfer aims to render videos in a target artistic style while preserving content, structure, and motion. While image stylization has advanced rapidly, video stylization remains challenging due to temporal inconsistency. Most existing methods stylize frames or keyframes and enforce consistency via heuristic temporal propagation, which is brittle under occlusions, disocclusions, and long-term motion, leading to drift and flickering artifacts. We argue that a fundamental bottleneck lies in the lack of large-scale triplet data and a principled training paradigm that jointly models and disentangles style, content, and motion.To address this, we introduce VISTA-1000, a synthetic dataset with 1,000 styles and motion-aligned triplets of style reference, clean video, and stylized video, and propose a diffusion-transformer-based in-context video style transfer framework with a lightweight style adapter for robust style extraction. Extensive experiments demonstrate SOTA performance in style fidelity, temporal consistency, and content preservation.

cs.CV↗

arXiv2023

Neuromorphic Incremental on-chip Learning with Hebbian Weight Consolidation

As next-generation implantable brain-machine interfaces become pervasive on edge device, incrementally learning new tasks in bio-plasticity ways is urgently demanded for Neuromorphic chips. Due to the inherent characteristics of its structure, spiking neural networks are naturally well-suited for BMI-chips. Here we propose Hebbian Weight Consolidation, as well as an on-chip learning framework. HWC selectively masks synapse modifications for previous tasks, retaining them to store new knowledge from subsequent tasks while preserving the old knowledge. Leveraging the bio-plasticity of dendritic spines, the intrinsic self-organizing nature of Hebbian Weight Consolidation aligns naturally with the incremental learning paradigm, facilitating robust learning outcomes. By reading out spikes layer by layer and performing back-propagation on the external micro-controller unit, MLoC can efficiently accomplish on-chip learning. Experiments show that our HWC algorithm up to 23.19% outperforms lower bound that without incremental learning algorithm, particularly in more challenging monkey behavior decoding scenarios. Taking into account on-chip computing on Synsense Speck 2e chip, our proposed algorithm exhibits an improvement of 11.06%. This study demonstrates the feasibility of employing incremental learning for high-performance neural signal decoding in next-generation brain-machine interfaces.

cs.ET↗