3 papers
cs.CV2026
Competition Domains Matter: Selective Head Globalization for Training-Free Sparse Attention in Video Diffusion Transformers
Tiantian Liu, Meng Wan, Zhiqiang Liang +3
Training-free sparse attention is a practical route to efficient video diffusion transformer inference, yet existing methods typically fix the competition domain at each query row…
cs.LG2025
Lossless Compression: A New Benchmark for Time Series Model Evaluation
Meng Wan, Benxi Tian, Jue Wang +7
The evaluation of time series models has traditionally focused on four canonical tasks: forecasting, imputation, anomaly detection, and classification. While these tasks have drive…
cs.LG2025
Enlightenment Period Improving DNN Performance
Tiantian Liu, Meng Wan, Jue Wang +1
The start of deep neural network training is characterized by a brief yet critical phase that lasts from the beginning of the training until the accuracy reaches approximately 50\%…