Author Archives: admin

Tuesday 18 February 2020 16:00 – 17:00 A. Payatakes Seminar Room

“Adaptive Wavelet Galerkin Methods”

Dr Nikos Rekatsinas Institute of Applied and Computational Mathematics

Abstract
The mathematical modelling of numerous phenomena in science, engineering and technology often leads to (systems of) Partial Differential Equations (PDEs) which typically can only be solved by numerical methods. In several applications, the solutions exhibit strong singularities caused, for instance, by non-smooth boundary parts of the domain or non-smooth forcing. In those cases, one seeks methods which are designed to be adapted to the features of interest of the solution. Our focus is on the Adaptive Wavelet Galerkin Method (awgm) for the optimal adaptive solution of stationary, and evolutionary PDEs. Adaptive approximation allows the local resolution of the approximation space to be adjusted to the local smoothness of the solution. Optimality of the solution means it can be approximated at the best possible rate allowed by the order of the basis -being in our case a wavelet Riesz basis- in linear computational complexity. The optimality and the overall qualitative properties of the awgm depend crucially on the efficiency of the involved approximate residual evaluation scheme. An improvement of the latter is based on the reformulation of a 2nd order PDE as a well-posed first order system least squares (FOSLS) problem. As an alternative to the usual time-marching schemes, the FOSLS approach is extended to the optimal adaptive solution of simultaneous space-time variational formulations of parabolic evolutionary PDEs, better suited to efficiently approximate singularities that are local in both space and time. The use of tensor products of temporal and spatial wavelets allows for the whole time evolution problem to be solved at a complexity of solving one instance of the corresponding stationary problem. The theoretical findings are illustrated with numerical results.

Tuesday 11 February 2020 16:00 – 17:00 A. Payatakes Seminar Room

“Localization of Sperm Whales with a 3-Hydrophone Array”

Dr Despoina Pavlidi Institute of Applied and Computational Mathematics

Abstract
Passive acoustic monitoring of marine mammals involves the use of hydrophones to localize vocalizing animals by merely listening to them and not disturbing them in any way. Sperm whales produce trains of pulsed sounds, so called clicks, during their long dives to depths of 1000 m or more. These sounds serve for communication and foraging (biosonar) and if they are picked up by a hydrophone array they can be used for animal localization and study. The focus here is on localization, which, apart from monitoring, can help the design of mitigation protocols to protect endangered populations from human activities. An underwater localization method based on +me-differences of arrival at 3 hydrophones will be presented. This is an extension to a previously proposed algorithm using 2 hydrophones and aims at the removal of pitfalls observed in the 2-hydrophone localization. Refraction of acous+c paths in the underwater environment is taken into account using ray theory. Further, a Bayesian framework is adopted allowing for the estimation of localization uncertainties. Three different array topologies are studied and their performance is compared against the 2hydrophone topology. The improvements obtained by the 3-hydrophone localization approach, in terms of removal of left-right ambiguity in azimuth estimation and reduction of uncertainty in broadside range estimation, will be highlighted.

Tuesday 04 February 2020 16:00 – 17:00 A. Payatakes Seminar Room

“Electrochemical Raman spectroscopy of 2D materials”

Dr. Kyriakos Filintoglou Institute of Chemical Engineering Sciences (ICE-CT)

Abstract
Two- dimensional (2D) transition metal dichalcogenides (TMDCs) have attracted attention recently due their unique physical properties (e.g. carrier mobility, appropriate bandgaps, rich excitonic effects, large spin-orbit coupling, spin-valley coupling). Besides, TMDCs raised a great interest for many applications such as optoelectronics, nanophotonics and valleytronics. Doping in 2D materials is an important strategy to precisely control their electronic and optical properties, without causing any induced structural disorder. In this study, we have successfully developed and tested a three-electrode electrochemical cell suitable for the electrical and in-situ Raman measurements. The single and few layer MoS2 and WS2 crystals have grown by a recently developed CVD method and have been investigated by a combination of Raman spectroscopy, Cyclic Voltammetry, and Impedance Spectroscopy. Impedance spectra suggest that the samples show pseudo-capacitive behavior. The Raman bands shift to higher frequencies with voltage application without any significant change in their widths and relative intensities. Finally, the MoS2 and WS2 were also studied by the application of biaxial mechanical strain. Our purpose is to construct a strain vs doping correlation plot based on the Raman measurements, providing an easy and fast way to determine the level of mechanical strain and carrier’s concentration of an unknown TMDC sample.

Tuesday 28 January 2020 16:00 – 17:00 A. Payatakes Seminar Room

” Nanomaterials for Energy Conversion and Energy Saving Devices”

Dr. George Sirrokostas Institute of Chemical Engineering Sciences (ICE-CT)

Abstract
The increasing demand for energy worldwide and the climate change makes the exploitation of renewable energy sources, covering today around 10% of the total energy consumption, a critical issue. It is predicted that energy demand will grow by more than 25% to 2040, due to rising incomes and a global population growing by 1.7 billion people, mostly in urban areas of developing economies. Low-carbon technologies, led by renewables and natural gas, will meet more than 80% of this increase in global energy demand. From the different renewable energy sources, solar energy is the most abundant one. Different types of photovoltaic devices have been used for solar energy harvesting. Among them, during the last few years, the pioneering breakthroughs on perovskite solar cells (PSCs) resulted in lab cells with a record efficiency above 20%, surpassing other more mature technologies such as dye sensitized solar cells (DSSCs), organic photovoltaics (OPV) and a-Si, after only few years of development. From the different device architectures that have been proposed, the carbon-based triple layer mesoscopic C-PSCs are a promising route towards potential commercialization. On the other hand, their efficiency is lower than that of planar devices, hence certain challenges must be addressed to improve further their efficiency and stability. A few parameters have been examined for improving the properties of the perovskite layer, while a novel integrated PV device, that combines energy production and the electrochromism functionality into one device, has been introduced.

Tuesday 21 January 2020 16:00 – 17:00 A. Payatakes Seminar Room

“Neuro-inspired deep learning architectures”

Dr Chavlis Spyridon Institute of Molecular Biology and Biotechnology (IMBB)

Abstract

A typical biological neuron, such as a pyramidal neuron of the hippocampus or the neocortex, receives thousands of afferent synaptic inputs to its dendritic tree, propagating its output downstream via efferent axonal transmission. In conventional Artificial Neural Networks (ANNs), dendritic trees are modeled as linear structures that sum weighted synaptic inputs to the cell bodies. However, numerous experimental and theoretical studies have shown that dendritic arbors are far more than simple linear integrators. That is, synaptic inputs can actively modulate neighboring synaptic activity; therefore, dendritic structures are highly nonlinear. In addition, the most widely-used ANN learning algorithm is backpropagation, which retrogradely broadcasts the total cost in order to fine-tune model parameters. Nevertheless, in biological systems, neurons communicate with each other via different rules which are based on features other than a global error (e.g., variance, correlation). In this study, inspired by the rules governing animals’ brains, we model the dendritic structures accompanied by their non-linearities and also we add biologically plausible learning rules. We apply this novel architecture to a typical machine learning task, namely the classification of images. We also show that our proposed architecture surpasses naive deep neural networks given the same complexity, i.e., number of parameters.

Tuesday 14 January 2020 16:00 – 17:00 A. Payatakes Seminar Room

“Functional architecture of spontaneous cortical networks in layer 2/3 of the primary visual cortex of the mouse”

Dr Vassilis Kehayas Institute of Computer Science (ICS)

Abtract
The brain’s neocortex is a six-layered structure that consists of billions of densely interconnected neurons arranged in topographic columns. Over time much has been learned about the computational properties of single neurons of the neocortex. However, there is uncertainty in the responses of single neurons to the same stimulus. Downstream neurons must integrate activity from large neuronal populations that exhibit coordinated activity. Nevertheless, on average, neurons display close to zero correlation with each other. We remain far from understanding how networks of cortical cells coordinate and interact with each other to process information. Here we study the functional topology of cortical networks during spontaneous activity, using layer 2/3 neurons of the primary visual cortex of the mouse as our experimental model system. We used multiple datasets acquired with different calcium reporters (OGB-1, GCaMP6s) containing a large population of neurons, ranging from ~100 to ~5000 cells. Such large samples were made possible with mesoscopic two-photon imaging, which allows the near-simultaneous recording of fields of views on the order of millimetres. Our hypothesis is that cortical networks are organized into functionally linked sub-networks that we can identify by studying spontaneous activity. We used a modified version of the spike time tiling coefficient, a metric that estimates directional temporal correlation and is robust to activity fluctuations, to construct network graphs of functional correlations. These graphs exhibit considerable temporal structure across multiple scales of correlation beyond that expected from networks constructed by circularly-shifting the observed activity patterns, in which the correlations between pairs of neurons are destroyed but the inter-event interval distributions are left intact. The observed networks had more functional connections, shorter average shortest paths, and higher average clustering coefficients compared to equivalent Erdös-Rényi networks, a model of irregular structure constructed by shuffling the edges between nodes. Consistent with this evidence, the observed graphs approach a “small-world” architecture across multiple scales of correlation. Our results show that spontaneous cortical activity exhibits substantial temporal structure despite that there is little correlation on average.

Abstract
The brain’s neocortex is a six-layered structure that consists of billions of densely interconnected neurons arranged in topographic columns. Over time much has been learned about the computational properties of single neurons of the neocortex. However, there is uncertainty in the responses of single neurons to the same stimulus. Downstream neurons must integrate activity from large neuronal populations that exhibit coordinated activity. Nevertheless, on average, neurons display close to zero correlation with each other. We remain far from understanding how networks of cortical cells coordinate and interact with each other to process information. Here we study the functional topology of cortical networks during spontaneous activity, using layer 2/3 neurons of the primary visual cortex of the mouse as our experimental model system. We used multiple datasets acquired with different calcium reporters (OGB-1, GCaMP6s) containing a large population of neurons, ranging from ~100 to ~5000 cells. Such large samples were made possible with mesoscopic two-photon imaging, which allows the near-simultaneous recording of fields of views on the order of millimetres. Our hypothesis is that cortical networks are organized into functionally linked sub-networks that we can identify by studying spontaneous activity. We used a modified version of the spike time tiling coefficient, a metric that estimates directional temporal correlation and is robust to activity fluctuations, to construct network graphs of functional correlations. These graphs exhibit considerable temporal structure across multiple scales of correlation beyond that expected from networks constructed by circularly-shifting the observed activity patterns, in which the correlations between pairs of neurons are destroyed but the inter-event interval distributions are left intact. The observed networks had more functional connections, shorter average shortest paths, and higher average clustering coefficients compared to equivalent Erdös-Rényi networks, a model of irregular structure constructed by shuffling the edges between nodes. Consistent with this evidence, the observed graphs approach a “small-world” architecture across multiple scales of correlation. Our results show that spontaneous cortical activity exhibits substantial temporal structure despite that there is little correlation on average.

Tuesday 25 December 2019 16:00 – 17:00 A. Payatakes Seminar Room

“Exploring the microcosmos with 3D Printing”

Dr. Areti Mourka Institute of Electronic Structure and Laser (IESL)

Abstract
Two-photon polymerization is a technique which allows the 3D printing of freeform structures with sub100nm resolution. It is a laser-based, 3D additive manufacturing technique which enables the ‘direct writing’ of computer-designed structures with resolution of a few tens of nanometers. It is based on two-photon absorption (2PA) by photopolymers; the focused beam of an ultra-fast laser ‘directly writes’ the 3D structure inside the volume of a transparent material, causing it to absorb two or more photons within the volume pixel (voxel) and polymerize locally. By removing the ‘unwritten’ area, one can obtain 3D copies of the computer design. The two-photon process has the advantages of reducing the volume within which photopolymerization occurs; increasing the resolution and also allowing the structures to be written within the volume of the photo-polymer. Two-photon polymerization enables 3D printing without the need for recoating, as required in classic 3D printing techniques. It is an appealing 3D printing technique for producing finely detailed microscale 3D structures due to its flexibility and it does not require extreme temperatures, harsh chemicals, or cleanroom facilities. In our studies, we demonstrate the tuning of the surface wetting performance via 3D microstructures. With two-photon polymerization, the role of the design of the 3D microstructures can be better understood, facilitating the applications, for which robust wetting control is required. Thus, we determine the intrinsic hydrophilicity of our homemade hybrid sol-gel polymer SZ2080 and subsequently micro-structured surfaces. Furthermore, the most commonly used photoresists in two-photon polymerization absorb in the ultraviolet (UV). These materials can therefore usually be excited with 2PA of visible or near-infrared (NIR) light. In our studies, we consider the characterization of general multi-photon absorption (MPA) correlating directly the properties of the features produced by MPA with the process parameters. Linewidth, power threshold, polymerization and damage thresholds, dynamic range and fabrication resolution have been object of investigation in our experiments. Moreover, this study complements the understanding of multi-photon polymerization (MPP) making use of machine learning for linking directly the various MPP printing parameters to the produced features.