Low-power analogue neural networks use trainable nonlinear connections

Researchers have developed low-power analogue neural networks that utilize trainable nonlinear connections, inspired by Kolmogorov-Arnold networks. These networks compute directly with analogue device physics, placing trainable nonlinear functions on each connection as analogue band-pass filters. While effective for tasks involving smooth, continuous values like robotic kinematics and control, they do not offer parameter efficiency for classification tasks. The networks have been successfully transferred to hardware with quantified fidelity and are projected to operate at very low power consumption. AI

IMPACT This research could lead to more energy-efficient AI hardware for continuous control tasks.

RANK_REASON The cluster contains an academic paper detailing a new approach to neural network architecture and hardware implementation. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.AI →

• arXiv
• CMOS
• Eleni Vasilaki
• field-programmable analogue arrays
• Kolmogorov-Arnold Networks
• Memristive Probabilistic Computing
• multilayer perceptron

• paper
• infra

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