ΦasmaNet
A lightweight neural network architecture for fast prediction of gravitational-wave signal-to-noise ratios, employing a staged spectral token refinement design with Φ-Stages, Φ-Refractor, and Φ-Feedback modules.
Timeline (last 90 days)
ΦasmaNet is a compact neural network architecture with about 732k parameters, used for fast prediction of gravitational-wave signal-to-noise ratio (SNR).
ΦasmaNet uses staged spectral refinement as an architectural prior, comprising ordered attention mechanism stages (Φ-Stages), a Φ-Refractor subspace mixing module, and a gated top-down Φ-Feedback mechanism.
ΦasmaNet is trained on the IMRPhenomXPHM waveform model (precessing, higher-mode binary black hole signals), taking 11-dimensional parameters as input and outputting the phase-averaged optimal SNR for a single detector under fixed Advanced LIGO design sensitivity.
With 10^6 training samples, ΦasmaNet achieves 100-MAPE=98.73% on the held-out test set.
In the detection-relevant regime (ρ>8), ΦasmaNet reaches an accuracy of 99.39%, with R^2=0.9998.
Under matched training conditions with training sample sizes from 10^5 to 10^6, ΦasmaNet shows advantages in comparisons against baseline models such as MLP, 1D-ResNet, and Transformer-encoder.
On a single GPU, ΦasmaNet can perform about 1.2×10^5 SNR evaluations per second, and end-to-end processing of a catalog containing 10^6 injections takes only about 8.3 seconds.
Running traditional waveform and inner-product computations on a 16-core CPU takes about two orders of magnitude longer than ΦasmaNet's evaluation on a single GPU.
ΦasmaNet supports exact distance rescaling via ρ∝1/d_L, thereby enabling efficient repeated detectability assessments for the same catalog.
The current analysis is mainly based on a paper abstract from a single source and still lacks independent third-party reproduction verification, but the research provides a fully documented pipeline and explicitly positions it as a reproducible benchmark.
2 more timeline events
All Facts (12)
ΦasmaNet is a compact neural network architecture with about 732k parameters, used for fast prediction of gravitational-wave signal-to-noise ratio (SNR).
50%UnverifiedΦasmaNet uses staged spectral refinement as an architectural prior, comprising ordered attention mechanism stages (Φ-Stages), a Φ-Refractor subspace mixing module, and a gated top-down Φ-Feedback mechanism.
50%UnverifiedΦasmaNet is trained on the IMRPhenomXPHM waveform model (precessing, higher-mode binary black hole signals), taking 11-dimensional parameters as input and outputting the phase-averaged optimal SNR for a single detector under fixed Advanced LIGO design sensitivity.
50%UnverifiedWith 10^6 training samples, ΦasmaNet achieves 100-MAPE=98.73% on the held-out test set.
50%UnverifiedIn the detection-relevant regime (ρ>8), ΦasmaNet reaches an accuracy of 99.39%, with R^2=0.9998.
50%UnverifiedUnder matched training conditions with training sample sizes from 10^5 to 10^6, ΦasmaNet shows advantages in comparisons against baseline models such as MLP, 1D-ResNet, and Transformer-encoder.
50%UnverifiedOn a single GPU, ΦasmaNet can perform about 1.2×10^5 SNR evaluations per second, and end-to-end processing of a catalog containing 10^6 injections takes only about 8.3 seconds.
50%UnverifiedRunning traditional waveform and inner-product computations on a 16-core CPU takes about two orders of magnitude longer than ΦasmaNet's evaluation on a single GPU.
50%UnverifiedΦasmaNet supports exact distance rescaling via ρ∝1/d_L, thereby enabling efficient repeated detectability assessments for the same catalog.
50%UnverifiedThe current analysis is mainly based on a paper abstract from a single source and still lacks independent third-party reproduction verification, but the research provides a fully documented pipeline and explicitly positions it as a reproducible benchmark.
50%UnverifiedΦasmaNet's demonstration is limited to a fixed single-detector configuration, and its extension to detector networks and epoch-specific sensitivities has not yet been validated.
50%UnverifiedThe authors note that preliminary results on public tabular datasets suggest that the ΦasmaNet architecture's competitiveness may extend beyond the gravitational-wave domain, with potential for transfer to broader tabular regression tasks.
50%