Special Session in INCONIP012 in Doha, Quatar: “Non-stationary Time Series Processing in Computational Neuroscience”

Event Date: 
12 Nov 2012 (All day) - 15 Nov 2012 (All day)

The 19th International Conference on Neural Information Processing

Special Session Organized by the INFER team: Non-stationary Time Series Processing in Computational Neuroscience

Doha, 12-15 November 2012

One of the most fundamental and yet unsolved problems in neuroscience is the understanding of the dynamics underlying the immediate and effortless perceptual processes. For instance, the reliable identification of speaker’s age, sex, identity or musical instruments in a noisy environment requires the on-line disentanglement of sophisticated temporal patterns of sounds, characterized by hundreds of variables (frequencies) which are separated by only 1/10 of a millisecond.

In other words, the enormous complexity of the human cognitive landscape requires the existence of an exceptionally fast and accurate dynamical code within the peripheral machinery and the cortex; which simultaneously adapts to continuous changes of external inputs and evaluates the flux of top-down information from higher cognitive areas. Performing such almost instantaneous robust pattern recognition is a very challenging task, not only for neurocompuational models but also for the state-of-the-art time series prediction algorithms. The nature of this incredibly precise machinery is still poorly understood. However, efforts in time series analyses of non-stationary neural recordings during the last decade could provide relevant insights for designing a new class of highly adaptive predictive models for rapidly varying underlying probability distributions.

Therefore, modeling the subtle trade-off between integration of complex information and exquisite temporal resolution in the human brain may enable us to build algorithms which perform with high precision and generalization capability in real-life, non-stationary environments. In this special session we propose the contribution of works which are focused in temporal pattern recognition on particularly challenging environments; which aim to go beyond out-of-sample performance measures on relatively stationary test datasets.