Some thoughts on machine learning

Misunderstanding of deep learning ：
Many people think that deep learning is equivalent to deep neural network , But neural networks have many defects , Today's neural networks rely on BP Algorithm ,BP The algorithm has many limitations , For example, only differentiable activation functions can be used , But in the real world, many construction models are non differentiable , Therefore, it is proposed that it does not depend on BP Depth model of algorithm ： Deep forest . Deep forests use fewer superparameters , And it doesn't depend on bp Algorithm , what's more , Deep forest provides a new research direction for artificial intelligence ： We can not only use neural networks when building deep learning models , Instead, you can use other algorithms to achieve deep learning
———— come from 2018 World AI summit zhouzhihua