השוואת שיטות

	הרפורמר: הטרנספורמר היעיל לרצפים ארוכים	מְיַדֵּעַ	Pyraformer: טרנספורמר קשב פירמידלי לחיזוי סדרות עתיות ארוכות טווח
תחום	למידה עמוקה	למידה עמוקה	למידה עמוקה
משפחה	Machine learning	Machine learning	Machine learning
שנת המקור	2020	2021	2022
הוגה השיטה	Nikita Kitaev, Łukasz Kaiser & Anselm Levskaya	Zhou, H. et al.	Shizhan Liu et al.
סוג	Memory-efficient attention-based sequence model	Transformer (ProbSparse self-attention)	Pyramidal self-attention transformer for time-series forecasting
מקור מכונן	Kitaev, N., Kaiser, Ł., & Levskaya, A. (2020). Reformer: The efficient transformer. ICLR.	Zhou, H. et al. (2021). Informer: Beyond Efficient Transformer for Long Sequence Time-Series Forecasting. AAAI.	Liu, S., Yu, H., Liao, C., Li, J., Lin, W., Liu, A. X., & Dustdar, S. (2022). Pyraformer: Low-complexity pyramidal attention for long-range time series modeling and forecasting. ICLR.
כינויים	Efficient Transformer, LSH Transformer, Locality-Sensitive Hashing Transformer, Verimli Dönüştürücü	Informer — Uzun Dizi Transformer Tahmini, Informer transformer, ProbSparse attention forecaster	Pyramidal Attention Transformer, Pyraformer Transformer, Piramit Dikkat Dönüştürücüsü, Low-Complexity Transformer
קשורות	2	5	3
תקציר	The Reformer is an efficient variant of the Transformer architecture introduced by Kitaev, Kaiser, and Levskaya at ICLR 2020. It addresses the prohibitive O(L²) memory and computational cost of standard self-attention for long sequences. The key innovations are locality-sensitive hashing (LSH) attention, which approximates full attention in O(L log L) time, and reversible residual layers that dramatically reduce activation memory during training.	Informer is a Transformer-based model introduced by Zhou et al. in 2021 for long-sequence time-series forecasting, using a ProbSparse self-attention mechanism that lowers the computational complexity of the standard Transformer to O(L log L). It is built for problems that demand predictions across thousands of future steps.	Pyraformer is a Transformer-based model for long-range time-series forecasting introduced by Liu et al. at ICLR 2022. Its central innovation is a Pyramidal Attention Module (PAM) that organizes tokens into a multi-resolution hierarchy, enabling the model to capture temporal dependencies across multiple scales while keeping time and memory complexity at O(L log L) rather than the quadratic cost of vanilla self-attention.