This week, we’re bringing you links from Charles, Simon, and Carly featuring neural networks, a video advocating for morality in algorithms, and incredible music made by machines.
To use a simple analogy, you might think of a neural network as an information assembly line: some piece of information — a picture say — comes in, is processed, and the result of that processing is passed along to another stage until a final information product — a label for the picture — is produced. In most current architectures, the assembly line has to remain fixed. Learning amounts to adjusting the proportion a given stage takes as input from its predecessors. Using this analogy, DyNet allows the network to adjust the structure of those assembly line stages during training. Some impressive results have been reported for language processing. Seems to be worth a serious look.
I’m a huge Cathy O’Neil fan: not only is she a long-time and active blogger on WordPress.com, she’s also an outspoken advocate for explicitly including morality into our thinking around algorithms and the real and often unjust impact those algorithms can have on peoples’ lives. In the TED talk, entitled “The era of blind faith in big data must end,” she tells us “Algorithms don’t make things fair — they repeat our past practices, they automate the status quo,” and discusses the impact that this kind of automation can have on our world.
Once more Magenta, with feeling! Magenta, a Google Brain project that uses machine intelligence to create art and music, has trained a model to generate music, with feeling! Their latest blog post details the data set and representation used to train the LSTM-based model. It features a number of sound clips generated by the model that I found fascinating to listen to.
What’s on your data reading night stand? Please feel welcome to share links to great reads and your insight on why you love them in the comments.