Approximating the Semantics of Logic Programs by Recurrent Neural Networks

In [1] we have shown how to construct a 3-layered recurrent neural network that computes the fixed point of the meaning function T_P of a given propositional logic program P, which corresponds to the computation of the semantics of P. In this article we consider the first order case. We define a notion of approximation for interpretations and prove that there exists a 3-layered feed forward neural network that approximates the calculation of T_P for a given first order acyclic logic program P with an injective level mapping arbitrarily well. Extending the feed forward network by recurrent connections we obtain a recurrent neural network whose iteration approximates the fixed point of T_P. This result is proven by taking advantage of the fact that for acyclic logic programs the function T_P is a contraction mapping on a complete metric space defined by the interpretations of the program. Mapping this space to the metric space R with Euclidean distance, a real valued function f_P can be defined which corresponds to T_P and is continuous as well as a contraction. Consequently it can be approximated by an appropriately chosen class of feed forward neural networks.     

A critical view on conservative mutations

By analysing the surface composition of a set of protein 3Dstructures, complemented with predicted surface compositionalinformation for homologous proteins, we have found significantevidence for a layer composition of protein structures. In theinnermost and outermost parts of proteins there is a net negativecharge, while the middle has a net positive charge. In addition,our findings indicate that the concept of conservative mutationneeds substantial revision, e.g. very different spatial preferenceswere found for glutamic acid and aspartic acid. The alaninescreening often used in protein engineering projects involvesthe substitution of residues to alanine, based on the assumptionthat alanine is a `neutral' residue. However, alanine has ahigh negative correlation with all but the non-polar residues.We therefore propose the use of, for example, serine as a substitutefor the residues that are negatively correlated with alanine.     

The effect of risk on trust attitude and trust behavior in interaction with information and decision automation

Situational risk has been postulated to be one of the most important contextual factors affecting operator’s trust in automation. However, experimentally, it has received only little attention and was directly manipulated even less. To close this gap, this study used a virtual reality multi-task environment where the main task entailed making a diagnosis by assessing different parameters. Risk was manipulated via the altitude, the task was set in including the possibility of virtually falling in case of a mistake. Participants were aided either by information or decision automation. Results revealed that trust attitude toward the automation was not affected by risk. While trust attitude was initially lower for the decision automation, it was equally high in both groups at the end of the experiment after experiencing reliable support. Trust behavior was significantly higher and increased during the experiment for the decision automation supported group in the form of less automation verification behavior. However, this detrimental effect was distinctly attenuated under high risk. This implies that negative consequences of decision automation in the real world might have been overestimated by studies not incorporating risk.