IEEE Signal Processing Magazine published a disappointing article on AI and the financial market.

This is a disappointing article.  Being published by the IEEE Signal Processing Magazine, I had hoped that it would shed new light on, or at least provide a survey of, the combined use of machine learning and signal processing in the hunt for new trading algorithms.  Instead, the author spent the bulk of the article on basic supply-and-demand principles and CAPM (Capital Asset Pricing Model).  I have two issues with this approach.  One, the author assumes that the reader has no prior knowledge about the financial market and he made a valiant but unsuccessful attempt to cover that in the confines of a few pages.  Two, he highlighted the use of CAPM, which is based on the erroneous premise that standard deviation equals risks, which in turn is based on the erroneous notion that the (statistical) distribution of returns is symmetrical.  This article is not unique in making the second mistake, since much of the financial industry is founded upon those fallacious ideas.

Deep Neural Networks are Easily Fooled: High Confidence Predictions for Unrecognizable Images

In the fast-moving world of deep learning, even though this research is not the latest, it is still relevant today:

…it is easy to produce images that are completely unrecognizable to humans, but that state-of-the-art DNNs believe to be recognizable objects with 99.99% confidence….

The lack of formal analysis in deep learning means it is mostly a “calibration” exercise, and it explains why universities are not yet teaching it as a semester-long course.

Because formal analysis hasn’t been developed yet, talking about deep learning methods very quickly devolves to this:

Read Hosea Siu‘s answer to Is there a reason why top universities like MIT, Berkeley, and Stanford, etc. don’t do that much deep learning? on Quora