Data science. Time complexity, inferential uncertainty, and spacekime analytics (Q2183427)

The main idea of this book is to allow time to be complex-valued -- so that any observed temporal trajectory corresponds to one of the lines \(c\cdot t\), \(t\in (-\infty, \infty)\) in the complex plane. The book starts with a convincing economic example explaining this idea. If there are more positive factors affecting the economy, then the economy grows exponentially, as \(\exp(a\cdot t)\) for some \(a>0\). If there are more negative factors, the economy shrinks -- same formula but with \(a<0\). But what happens in an intermediate situation, when there is approximately the same amount of positive and negative factors? In the usual exponential model, the value \(a\) between positive and negative is \(a=0\), which would mean that the Gross Domestic Product remains constant, but in reality, in such cases, the economy oscillates -- i.e., in effect, that we have an imaginary value \(a\). This is a good example of how complex-valued time models can lead to a better description of reality. The authors' idea is in line with many examples where extension to complex values helps process data: e.g., complex values are effectively used in the analysis of electric circuits and electromagnetic fields. The main difference is that, usually, quantities become complex while time remains real; in contrast, the book proposes to make time complex-valued too. The authors follow their idea in two directions. First, they extend physics equations to the case of complex-valued time -- which means, in effect, replacing the usual 4-D space-time with the 5-D one -- and study the properties of such extended equations. In particular, this leads to the 1920s Kaluza-Klein observation that a 5-D generalization of Einstein's General Relativity Theory automatically leads to Maxwell's equations of electrodynamics, but the book goes way beyond that. For example, the authors have interesting ideas about the arrow of time -- namely, one can use complex values to come up with a continuous transition between forward-in-time and backward-in-time processes. Another direction in which the authors explore their idea is applying it to data processing. In addition to analysis of economic data, the authors also apply this idea to the analysis of fMRI data and to image processing in general. This book can be recommended to physicists, philosophers, and data scientists interested in learning and developing new promising ideas. My only warning is about the title: the book definitely describes many data processing techniques, but it is not -- as some readers may mistakenly expect -- a textbook describing traditional data science techniques.