Which Transformer to Favor: A Comparative Analysis of Efficiency in Vision Transformers
Presentation at WACV 2025 on a large-scale benchmark of 45+ transformer models for image classification, evaluating accuracy, speed, and memory efficiency.
slides
Presentations.
2 talks
TaylorShift: Shifting the Complexity of Self-Attention from Squared to Linear (and Back) using Taylor-Softmax
Oral presentation at ICPR 2024 introducing TaylorShift, a novel reformulation of the attention mechanism using Taylor-Softmax that enables full token-to-token interactions in linear time.