Adaptive integration in the visual cortex by depressing recurrent cortical circuits

Neurons in the visual cortex receive a large amount of input from recurrent connections, yet the functional role of these connections remains unclear. Here we explore networks with strong recurrence in a computational model and show that short-term depression of the synapses in the recurrent loops implements an adaptive filter. This allows the visual system to respond reliably to deteriorated stimuli yet quickly to high-quality stimuli. For low-contrast stimuli, the model predicts long response latencies, whereas latencies are short for high-contrast stimuli. This is consistent with physiological data showing that in higher visual areas, latencies can increase more than 100 ms at low contrast compared to high contrast. Moreover, when presented with briefly flashed stimuli, the model predicts stereotypical responses that outlast the stimulus, again consistent with physiological findings. The adaptive properties of the model suggest that the abundant recurrent connections found in visual cortex serve to adapt the network's time constant in accordance with the stimulus and normalizes neuronal signals such that processing is as fast as possible while maintaining reliability.     

Modelling the fate of glucosinolates during thermal processing of Brassica vegetables

Glucosinolates are secondary metabolites of Brassica vegetables that have been associated with health benefits. The concentrations of these compounds are strongly affected by processing of the vegetables. Various mechanisms are responsible for these changes: Lysis of plant cells and compartments, diffusion and leaching of glucosinolates and myrosinase (a plant enzyme able to hydrolyze glucosinolates) into the cooking water, enzymatic hydrolysis of glucosinolates, inactivation of myrosinase and thermal degradation of glucosinolates. This publication presents a dynamic mathematical model that includes these mechanisms with their estimated parameters and the effect of temperature on them.Simulations made by the model for several process conditions show losses of glucosinolates as a consequence of domestic boiling of 69%, microwaving of 8%, industrial blanching of 37% and by industrial sterilization of 82%.The model can assist in adapting the processing conditions like the time–temperature profile and the vegetable–water ratio to optimize industrial and domestic processing of Brassica vegetables in terms of health benefits. In addition this model can help to add the effect of vegetable preparation practices on the quantitative glucosinolate intake in epidemiological studies.     

Learning potential functions and their representations for multi-task reinforcement learning

In multi-task learning, there are roughly two approaches to discovering representations. The first is to discover task relevant representations, i.e., those that compactly represent solutions to particular tasks. The second is to discover domain relevant representations, i.e., those that compactly represent knowledge that remains invariant across many tasks. In this article, we propose a new approach to multi-task learning that captures domain-relevant knowledge by learning potential-based shaping functions, which augment a task’s reward function with artificial rewards. We address two key issues that arise when deriving potential functions. The first is what kind of target function the potential function should approximate; we propose three such targets and show empirically that which one is best depends critically on the domain and learning parameters. The second issue is the representation for the potential function. This article introduces the notion of $k$ -relevance, the expected relevance of a representation on a sample sequence of $k$ tasks, and argues that this is a unifying definition of relevance of which both task and domain relevance are special cases. We prove formally that, under certain assumptions, $k$ -relevance converges monotonically to a fixed point as $k$ increases, and use this property to derive Feature Selection Through Extrapolation of k-relevance (FS-TEK), a novel feature-selection algorithm. We demonstrate empirically the benefit of FS-TEK on artificial domains.     

Adaptive efficient global optimization of systems with independent components

We present a novel approach for efficient optimization of systems consisting of expensive to simulate components and relatively inexpensive system-level simulations. We consider the types of problem in which the components of the system problem are independent in the sense that they do not exchange coupling variables, however, design variables can be shared across components. Component metamodels are constructed using Kriging. The metamodels are adaptively sampled based on a system level infill sampling criterion using Efficient Global Optimization. The effectiveness of the technique is demonstrated by applying it on numerical examples and an engineering case study. Results show steady and fast converge to the global deterministic optimum of the problems.