Haoyue Tan | Searcharxiv

arXiv2026

AdaCluster: Adaptive Query-Key Clustering for Sparse Attention in Video Generation

Video diffusion transformers (DiTs) suffer from prohibitive inference latency due to quadratic attention complexity. Existing sparse attention methods either overlook semantic similarity or fail to adapt to heterogeneous token distributions across layers, leading to model performance degradation. We propose AdaCluster, a training-free adaptive clustering framework that accelerates the generation of DiTs while preserving accuracy. AdaCluster applies an angle-similarity-preserving clustering method to query vectors for higher compression, and designs a euclidean-similarity-preserving clustering method for keys, covering cluster number assignment, threshold-wise adaptive clustering, and efficient critical cluster selection. Experiments on CogVideoX-2B, HunyuanVideo, and Wan-2.1 on one A40 GPU demonstrate up to 1.67-4.31x speedup with negligible quality degradation.

cs.CV↗

arXiv2025

ModFus-DM: Explore the Representation in Modulated Signal Diffusion Generated Models

Automatic modulation classification (AMC) is essential for wireless communication systems in both military and civilian applications. However, existing deep learning-based AMC methods often require large labeled signals and struggle with non-fixed signal lengths, distribution shifts, and limited labeled signals. To address these challenges, we propose a modulation-driven feature fusion via diffusion model (ModFus-DM), a novel unsupervised AMC framework that leverages the generative capacity of diffusion models for robust modulation representation learning. We design a modulated signal diffusion generation model (MSDGM) to implicitly capture structural and semantic information through a progressive denoising process. Additionally, we propose the diffusion-aware feature fusion (DAFFus) module, which adaptively aggregates multi-scale diffusion features to enhance discriminative representation. Extensive experiments on RML2016.10A, RML2016.10B, RML2018.01A and RML2022 datasets demonstrate that ModFus-DM significantly outperforms existing methods in various challenging scenarios, such as limited-label settings, distribution shifts, variable-length signal recognition and channel fading scenarios. Notably, ModFus-DM achieves over 88.27% accuracy in 24-type recognition tasks at SNR $\geq $ 12dB with only 10 labeled signals per type.

eess.SP↗