A heterosynaptic learning rule for neural networks

In this article we intoduce a novel stochastic Hebb-like learning rule for neural networks that is neurobiologically motivated. This learning rule combines features of unsupervised (Hebbian) and supervised (reinforcement) learning and is stochastic with respect to the selection of the time points when a synapse is modified. Moreover, the learning rule does not only affect the synapse between pre- and postsynaptic neuron, which is called homosynaptic plasticity, but effects also further remote synapses of the pre-and postsynaptic neuron. This more complex form of synaptic plasticity has recently come under investigations in neurobiology and is called heterosynaptic plasticity. We demonstrate that this learning rule is useful in training neural networks by learning parity functions including the exclusive-or (XOR) mapping in a multilayer feed-forward network. We find, that our stochastic learning rule works well, even in the presence of noise. Importantly, the mean leaxning time increases with the number of patterns to be learned polynomially, indicating efficient learning.

Correlates of constipation in people with Parkinson's

Purpose: To investigate clinical, demographic and dietary factors associated with constipation in a sample of community dwelling people with Parkinson's disease, recruited through a specialist outpatient clinic. Partners/carers provided a convenience control group. Scope: Participants completed a baseline questionnaire (background information, diet and exercise, activities of daily living: mobility and manual dexterity, health-related quality of life (SF-12), stool frequency and characteristics, extent of concern due to constipation, laxative taking), and a four-week stool diary. The Rome criterion was used to determine constipation status. Multiple regression methods were used to explore the correlates of constipation. Baseline data were provided by 121 people with Parkinson's, (54 controls), of whom 73% (25%) met the Rome criterion. Prospective diary data from 106 people with Parkinson's (43 controls) showed lower proportions: 35% (7%) meeting the Rome criterion. Among all study subjects, i.e. Parkinson's patients and controls taken together, the presence of constipation is predicted by having Parkinson's disease (p=.003; odds ratio 4.80, 95% CI 1.64-14.04) and mobility score (p=.04; odds ratio 1.15, 95% CI 1.01-1.31), but not by dietary factors. Amongst people with Parkinson's constipation is predicted by number of medications (p=.027). Laxative taking masks constipation, and is significantly associated with wearing protection against bowel incontinence (p=.009; odds ratio 4.80, 95% CI: 1.48-15.52). Conclusions: Constipation is disease-related, not a lifestyle factor. More research is needed on optimal management and laxative use. © 2010 Elsevier Ltd.

How different are the correlates of onset and continuation of civil wars?

It is very common to analyse the factors associated with the onset and continuation of civil wars entirely separately, as if there were likely to be no similarity between them. This is an overstatement of the theoretical position, which has established only that they may be different (i.e. less than perfectly correlated). The hypothesis that the explanatory variables are the same is not theoretically excludable and is empirically testable, both for individual variables and for combinations of them. Starting from this approach yields a rather different picture of the factors associated with the continuation of civil wars, because the relatively small sample size means that confidence intervals on individual coefficients are wide in this case. It is shown here that country size, mountainous terrain and (in most datasets) ethnic diversity seem significant for the continuation of civil wars, starting from the null hypothesis that variables affect onset and continuation probabilities identically, rather than entirely independently. One variable that affects onset and continuation significantly differently is anocracy, which we find to matter only for onset. Civil war is more likely if it occurred two years previously, as well as one year previously, which indicates that wars are more likely to restart after only one year of peace, and also more likely to stop in their first year. The combined model strengthens the result that ethnic diversity matters (it is consistently significant across datasets, whereas it is not when onset is analysed separately), although in the UCD/PRIO dataset it is significant only for onset. By contrast, if continuation is analysed independently, virtually nothing is significant except a pre-1991 dummy and a dummy for civil war two years previously.