Joshua Chover

The early growth of cancer.

In this paper, we compare two types of stochastic models for the initial growth of cancerous tumors. In one type, the random element enters via the initial time of growth or via the initial size of the growth clone. In the other type, tumors differ from one another essentially via their growth rates. We present a simple test to distinguish between the two types when tumor size distributions are available from several time points. Size distributions are the key elements of such kinetic analysis given the limitation that an individual tumor can be measured only once, at the time of sacrifice of an experimental animal. We discuss these concepts in connection with data from particular experiments on carcinogenic growth in the livers of mice.

Convergence of a local lattice process

We analyze a discrete time Markov process on the 0-1 patterns of an infinite lattice, with transitions determined by nearest neighbors; and give a sufficient condition for ergodicity.

Neural correlation via random connections

A simple neural network is studied, which has sparse, random, plastic, excitatory connections and also feedback loops between sensory cells and correlator cells. Time is limited to several discrete instants, where firing is synchronous. For parameter values within biological ranges, the system exhibits a capacity for associative recall, with a controlled amount of extraneous firing, following Hebb-like synaptic changes.

Recall via transient neuronal firing

Abstract A neural network with random afferent and collateral connections is examined during the first few waves of firing in response to a stimulus. (Cells fire when linear combinations of EPSPs exceed thresholds.) Such response is compared with response to a noise-altered version of the given stimulus (mediated by altered synaptic weights), and concordances are averaged over possible stimuli. Formulae are calculated for system performance characteristics by use of normal approximations. With reasonable parameter sets, there is indication that the network is capable of associative recall for a significant fraction of stimuli.