AI新算法模型CapsNet在方法论上的革新

The Innovation in Methodology of a New AI Algorithm Model CapsNet

深度学习之父杰弗里·辛顿构造了新算法模型—Caps Net,用胶囊(Capsule)来模拟大脑皮质中的皮质柱并用之来存储概念性知识,并革命性地重新把分析还原方法引进卷积神经网络(CNN)中。本文认为辛顿在方法论上的革新体现在重归模拟人脑之路和建构非完全性还原方法,反映了他关于"如何实现类人智能"的思考:"意识"产生于同时具备概念能力和非概念能力的神经网络。这或将在AI领域引发范式变革。

Recently,Geoffrey Hinton built a new algorithm model-CapsNet. In the model,Hinton used Capsules to simulate cortical minicolumns in the cortex and to store conceptual knowledge and also reintroduced the analytical reduction method to CNN in a revolutionary way. This paper holds that Hinton made two major methodological innovations-reconstructing the road to simulate human brain and constructing the incomplete reduction method-which reflect his thinking on how to achieve human-like intelligence: " consciousness" is generated from the neural network with both conceptual ability and non-conceptual ability. This could lead to paradigm shifts in AI.