The rationale for ensemble and meta-algorithmic architectures in signal and information processing

Abstract

We are living through an historic era in computing. As I . I N T R O D U C T I O NMachine intelligence is all around us. Our financial institutions sift through our purchasing data to define patterns of usual -and by their complement unusual -activity in order to minimize fraud. Most likely words and queries show up in typed, spoken, and gestured applications and services. Selfdriving cars are part of the everyday conversation. It is clear that a growing percentage of the intelligent decisions made in our world are performed by the machines, and not the minds, of humans.This "additional level of indirection" for decision-making comes at a price. There is the risk that reliance on machine algorithms may make humans less capable in some intellectual capacities; however, this issue will not be addressed here. What is more germane -and potentially more frightening -is that machines will make critical mistakes which go unnoticed. This argument is not based on an apocalyptic or Luddite fear of computers. Computers have arguably improved human lives more than any other invention (water purification and agriculture notwithstanding). Instead, it is because machine intelligence is so very different from human intelligence that the reliability of machine intelligence output must be questioned. Computers and their associated algorithms, systems and intelligence engines outperform humans on many mathematical and analytical tasks. To date, however, humans significantly outperform computers on more distributed, Gestalt, intuitive and unstructured tasks. As computers are tasked more and more in this purported "Big Data" era to discern information from ever-growing amounts of unstructured data, how can we be sure that the machine intelligence is consistent with human intelligence? One approach is to design machine intelligence systems that share the precepts of human intelligence. Paramount among these are the following precepts:• The ability to incorporate new rules of discernment and models for analysis into an existing model or set of models; • The ability to selectively apply different learned content based on the context; • The ability to correctly solve problems in the absence of sufficient input; • The ability to perform "real time ground truthing"; that is, the ability to assess a situation after a particular output is assumed and efficiently perform a minimal set of validation experiments to prove or disprove the correctness of the output.These precepts will be re-considered (Section VII) after the stage for meta-algorithm-based architectures is given. The ensemble and meta-algorithmic approaches and their