AI Insight
This paper introduces the Neural Langevin Machine, a generative neural network model that uses fixed points in recurrent networks to store and generate information. The model employs a biologically plausible learning rule that is asymmetric and local, meaning neurons only need information from their immediate connections rather than global network signals. The researchers demonstrate that the system exhibits a transition from memorization to generalization as training data increases and can perform tasks like image generation and denoising.
Why it matters
The locally-computed learning rule makes this approach more biologically realistic than traditional backpropagation-based methods, potentially informing our understanding of how biological neural networks learn. The model's ability to generate novel outputs and denoise images suggests practical applications in artificial intelligence and computational neuroscience.
Understand the Science
⚠️ Preprint – Noch nicht peer-reviewed
Dieser Artikel wurde noch nicht von unabhängigen Experten begutachtet. Die Ergebnisse sind vorläufig und sollten mit Vorsicht interpretiert werden.
Abstract: Fixed points of recurrent neural networks can be leveraged to store and generate information. These fixed points are captured by the Boltzmann-Gibbs measure, which leads to neural Langevin dynamics that relax to those fixed points for generative learning of a real dataset. We call this type of generative model a neural Langevin machine, which derives an asymmetric and firing-rate-speed-adjusted learning rule requiring only local neural signals, thereby bearing biological relevance in terms of local predictive learning. An out-of-equilibrium regime of the generative process is revealed, together with a memorization-to-generalization transition with increasing training data size. The neuro-inspired machine can also realize a continuous exploration of the phase space for different kinds of generative images and can denoise a corrupted image as well.
Source: Neural Langevin Machine: a local asymmetric learning rule can be creative