Signals Under Noise and Networks
Leonid Perlovsky
Air Force Research Laboratory and Harvard University
Abstract:
Detecting weak signals in strong noise is mathematically difficult. I briefly analyze fundamental reasons for these difficulties and relate them to logic and Gödelian incompleteness. It turned out that most algorithms and neural networks specifically designed to overcoming logical limitations, still use logic at some algorithmic steps. Dynamic logic (DL) is a breakthrough mathematical technique overcoming these limitations and resulting in fast and efficient algorithms. The talk describes few examples of detecting objects and situations in noise, where DL solves problems considered unsolvable, and detect signals below noise 100 times better than has been considered possible for decades. DL can be used to detect networks, and the talk addresses finding actual neural connectivity from EEG, EMG, fMRI. The last part of the talk discusses that DL models functionality of the mind: concepts, emotions, instincts. Recent neuro-imaging experiments demonstrated that DL models actual operation of human visual system. I briefly touch on a possible model of interacting language and cognition hierarchies; role of beautiful and music in cognition. Mathematically, DL is an extension of fuzzy logic toward a process “from vague-to-crisp.” No specific mathematical knowledge is assumed.
The seminar will take place
in 44 Cummington St. Room 705