news

한어Русский языкEnglishFrançaisIndonesianSanskrit日本語DeutschPortuguêsΕλληνικάespañolItalianoSuomalainenLatina

it home reported on september 1 that the institute of microelectronics of the chinese academy of sciences and others combined artificial neural networks with the dynamic reconfigurability of the brain to developdynamic neural network based on semantic memory。

▲ brain-inspired dynamic neural network hardware and software co-design based on semantic memory

the brain's neural network has complex semantic memory and dynamic connectivity, which can link ever-changing inputs with experiences in the vast memory and efficiently perform complex and changing tasks.

currently, artificial intelligence systems are widely usedneural network models are mostly staticas the amount of data continues to grow, it generates a lot of energy consumption and time overhead in traditional digital computing systems, making it difficult to adapt to changes in the external environment.

compared with static networks, semantic memory dynamic neural networks cantrade-off between recognition accuracy and computational efficiency based on computational resources, which can show excellent performance on resource-constrained devices or distributed computing environments.

it home learned that in the classification tasks of the 2d image dataset mnist and the 3d point cloud dataset modelnet, the design achieved an accuracy comparable to that of the software.compared with static neural networks, the computational effort is reduced by 48.1% and 15.9%, which reduces computing energy consumption compared to traditional digital hardware systems.