Phasor Transformer offers efficient alternative to attention for time-series modeling

Researchers have introduced the Phasor Transformer, a novel block designed to address the quadratic bottleneck in attention mechanisms for time-series data. This new approach represents sequence states on the unit-circle manifold and utilizes a combination of trainable phase-shifts and a parameter-free Discrete Fourier Transform for global token coupling. The resulting Large Phasor Model (LPM) demonstrates competitive performance against self-attention models at a fraction of the parameter count, establishing a new efficiency-accuracy frontier for temporal modeling in oscillatory domains. AI

IMPACT Offers a more efficient approach to time-series modeling, potentially impacting applications requiring long-context understanding.

RANK_REASON This is a research paper detailing a new model architecture. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.AI →

• alphaXiv
• arXiv
• CatalyzeX
• DagsHub
• Dibakar Sigdel
• discrete Fourier transform
• Hugging Face
• Large Phasor Model
• Luyten Proper-Motion catalogue
• Phasor Transformer
• S1 (Berlin)
• Transformer++

• paper
• infra

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →