New Fractional Stochastic Neural Networks Improve Long Memory Recovery and Robustness

Researchers have introduced Fractional Stochastic Neural Networks (FSNNs), a novel architecture that incorporates fractional Brownian motion to drive residual dynamics. This approach utilizes a discrete stochastic maximum principle to establish an adjoint recursion, enabling the proof of mean square convergence for projected samplewise stochastic gradient descent under deterministic network parameters. The FSNNs have demonstrated potential in various applications, including noisy regression with uncertainty quantification, long-memory time series generation, and image classification under structured perturbations, showing improvements in long-memory recovery and robustness compared to traditional Brownian and deterministic baselines. AI

IMPACT Introduces a new neural network architecture that may enhance performance in time-series generation and image classification tasks.

RANK_REASON The cluster contains a research paper detailing a new type of neural network architecture. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.LG →

• fractional Brownian motion
• Fractional Stochastic Neural Networks
• stochastic gradient descent

• paper
• model release

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