Deep learning is going to be an important tool in business, as I’ve argued in previous issues. Neural nets are sometimes posited as brain-like structures that don’t need any prior information or parameterisation.
In reality, in production, tuning a neural net is an art and
a science itself. Even the best examples of deep learning we’ve seen require substantial prior knowledge (selecting the depth of the network, number of layers, learning rate, quantization/tokenization of input data and the selection of hyperparameters
Of course, what we can do with modern neural nets and todays CPUs and data sources is kinda amazing. Here are a few examples from this week: