AI Insight
PhasorFlow is a new open-source Python library that performs computations on the unit circle by encoding inputs as complex phasors and processing them through unitary wave-interference gates while preserving global norm. The library introduces three key components: a 22-gate Phasor Circuit model for unit-circle computation, Variational Phasor Circuits (VPC) that optimize phase parameters for classification tasks, and a Phasor Transformer that replaces standard attention mechanisms with parameter-free discrete Fourier transform token-mixing. Testing on diverse tasks including EEG motor-imagery classification showed the VPC matched baseline performance with significantly fewer parameters, though the authors transparently note fundamental limitations including a parity ceiling that additional depth cannot overcome.
Why it matters
This work presents a computationally lightweight alternative to standard neural networks that could be particularly valuable for resource-constrained applications where parameter efficiency is critical. The deterministic, unit-circle-based approach runs on classical hardware while achieving comparable performance to traditional methods in several domains, offering a new paradigm for machine learning model design.
Understand the Science
arXiv:2603.15886v3 Announce Type: replace
Abstract: We present PhasorFlow, an open-source Python library for computing on the $S^1$ unit circle. Inputs are encoded as complex phasors $z=e^{iphi}$ on the $N$-torus ($mathbb{T}^N$); as computation proceeds through unitary wave-interference gates, global norm is preserved while components drift into $mathbb{C}^N$, letting algorithms leverage continuous geometric gradients. PhasorFlow makes three contributions. First, we formalize the Phasor Circuit model ($N$ threads, $M$ gates) with a 22-gate library spanning standard-unitary, non-linear, neuromorphic, and encoding operations under full matrix-algebra simulation. Second, we present the Variational Phasor Circuit (VPC), analogous to variational quantum circuits, optimizing continuous phase parameters for classification. Third, we introduce the Phasor Transformer, replacing $QK^TV$ attention with a parameter-free DFT token-mixing layer inspired by FNet. We validate on spatial classification, time-series prediction, financial volatility, neuromorphic tasks, and — for the VPC — real motor-imagery EEG, where it matches standard baselines at a fraction of their parameters. We characterize the models honestly: the VPC is a parameter-efficient phase-linear classifier with a parity ceiling that depth cannot raise, and the Phasor Transformer benefits from depth before saturating, competitive but not superior. This positions unit-circle computing as a deterministic, lightweight paradigm on classical hardware. Available at https://github.com/mindverse-computing/phasorflow.
Source: PhasorFlow: A Python Library for Unit Circle Based Computing