Noise and non-linearities in high-throughput data

Abstract

High throughput data analysis are becoming common in biology, communications, economics and sociology. This vast amount of data are usually represented in form of matrices and can be considered as knowledge networks. Spectralbased approaches have been proved useful in extracting hidden information within such networks and to estimate missing data, but these methods are based essentially on linear assumptions. The physical models of matching, when available, often suggest nonlinear mechanisms, that may sometimes be identified as noise. The use of nonlinear models in data analysis, however, may require the introduction of many parameters, that lowers the statistical weight of the model. According with the quality of data, a simpler linear analysis may be more convenient than more complex approaches.In this paper, we show how a simple nonparametric Bayesian model may be used to explore the role of nonlinearities and noise in synthetic and experimental datasets.

Cognitive Network Structure: An Experimental Study

“…From each communication channel, private and public textual chats and private radar, we have extracted the relevant events from the log file, and computed several indicators to describe both the communication and the cognitive network. Since the semantic content of a message crucially depends on the cultural context, and having already scheduled future experiments with people coming from different countries (and cultures), we decided to base our analysis only on the timing and number of messages exchanged by agents [19]. Of course the semantic aspects represent a very relevant dimension to analyze and such analysis will require a dedicated work.…”