John D. Hunter

Clustering neural spike trains with transient responses

The detection of transient responses, i.e. nonstationarities, that arise in a varying and small fraction of the total number of neural spike trains recorded from chronically implanted multielectrode grids becomes increasingly difficult as the number of electrodes grows. This paper presents a novel application of an unsupervised neural network for clustering neural spike trains with transient responses. This network is constructed by incorporating projective clustering into an adaptive resonance type neural network (ART) architecture resulting in a PART neural network. Since comparisons are made between inputs and learned patterns using only a subset of the total number of available dimensions, PART neural networks are ideally suited to the detection of transients. We show that PART neural networks are an effective tool for clustering neural spike trains that is easily implemented, computationally inexpensive, and well suited for detecting neural responses to dynamic environmental stimuli.

Frequency domain analysis of human subdural recordings.

Summary: It is possible to localize many aspects of cortical function and dysfunction without the use of direct electrical stimulation of cortex. This study explores the degree to which information can be obtained about functional cortical organization relative to epileptogenic regions through analysis of electrocorticographic recordings in the frequency domain. Information about the extent of seizure regions and the location of the normal sensory and motor homunculus and some higher language and memory related areas can be obtained through the analysis of task-related power spectrum changes and changes in lateral interelectrode coherence patterns calculated from interictal and ictal recordings.

Projective clustering using neural networks with adaptive delay and signal transmission loss

We develop a new neural network architecture for projective clustering of data sets that incorporates adaptive transmission delays and signal transmission information loss. The resultant selective output signaling mechanism does not require the addition of multiple hidden layers but instead is based on the assumption that the signal transmission velocity between input processing neurons and clustering neurons is proportional to the similarity between the input pattern and the feature vector (the top-down weights) of the clustering neuron. The mathematical model governing the evolution of the signal transmission delay, the short-term memory traces, and the long-term memory traces represents a new class of large-scale delay differential equations where the evolution of the delay is described by a nonlinear differential equation involving the similarity measure already noted. We give a complete description of the computational performance of the network for a wide range of parameter values.