arXiv subjects

Alexander Schulz

Publications and source records attributed to Alexander Schulz.

26 records · Page 2Linked to original sources

arXiv2020

Reservoir memory machines

In recent years, Neural Turing Machines have gathered attention by joining the flexibility of neural networks with the computational capabilities of Turing machines. However, Neural Turing Machines are notoriously hard to train, which limits their applicability. We propose reservoir memory machines, which are still able to solve some of the benchmark tests for Neural Turing Machines, but are much faster to train, requiring only an alignment algorithm and linear regression. Our model can also be seen as an extension of echo state networks with an external memory, enabling arbitrarily long storage without interference.

cs.LG↗

arXiv2019

Quantum-dot-like states in molybdenum disulfide nanostructures due to the interplay of local surface wrinkling, strain, and dielectric confinement

The observation of quantum light emission from atomically thin transition metal dichalcogenides has opened a new field of applications for these material systems. The corresponding excited charge-carrier localization has been linked to defects and strain, while open questions remain regarding the microscopic origin. We demonstrate that the bending rigidity of these materials leads to wrinkling of the two-dimensional layer. The resulting strain field facilitates strong carrier localization due to its pronounced influence on the band gap. Additionally, we consider charge carrier confinement due to local changes of the dielectric environment and show that both effects contribute to modified electronic states and optical properties. The interplay of surface wrinkling, strain-induced confinement, and local changes of the dielectric environment is demonstrated for the example of nanobubbles that form when monolayers are deposited on substrates or other two-dimensional materials.

cond-mat.mes-hall↗

arXiv2017

Expectation maximization transfer learning and its application for bionic hand prostheses

Machine learning models in practical settings are typically confronted with changes to the distribution of the incoming data. Such changes can severely affect the model performance, leading for example to misclassifications of data. This is particularly apparent in the domain of bionic hand prostheses, where machine learning models promise faster and more intuitive user interfaces, but are hindered by their lack of robustness to everyday disturbances, such as electrode shifts. One way to address changes in the data distribution is transfer learning, that is, to transfer the disturbed data to a space where the original model is applicable again. In this contribution, we propose a novel expectation maximization algorithm to learn linear transformations that maximize the likelihood of disturbed data after the transformation. We also show that this approach generalizes to discriminative models, in particular learning vector quantization models. In our evaluation on data from the bionic prostheses domain we demonstrate that our approach can learn a transformation which improves classification accuracy significantly and outperforms all tested baselines, if few data or few classes are available in the target domain.

cs.LG↗

arXiv2017

Heterogeneous diffusion in comb and fractal grid structures

We give an exact analytical results for diffusion with a power-law position dependent diffusion coefficient along the main channel (backbone) on a comb and grid comb structures. For the mean square displacement along the backbone of the comb we obtain behavior $\langle x^2(t)\rangle\sim t^{1/(2-α)}$, where $α$ is the power-law exponent of the position dependent diffusion coefficient $D(x)\sim |x|^α$. Depending on the value of $α$ we observe different regimes, from anomalous subdiffusion, superdiffusion, and hyperdiffusion. For the case of the fractal grid we observe the mean square displacement, which depends on the fractal dimension of the structure of the backbones, i.e., $\langle x^2(t)\rangle\sim t^{(1+ν)/(2-α)}$, where $0<ν<1$ is the fractal dimension of the backbones structure. The reduced probability distribution functions for both cases are obtained by help of the Fox $H$-functions.

cond-mat.stat-mech↗

arXiv2015

Electronic Structures and Optical Properties of Partially and Fully Fluorinated Graphene

In this letter we study the electronic structures and optical properties of partially and fully fluorinated graphene by a combination of abinitio G0W0 calculations and large-scale multi-orbital tight-binding simulations. We find that for partially fluorinated graphene, the appearance of paired fluorine atoms is more favorable than unpaired atoms. We also show that different types of structural disorder, such as carbon vacancies, fluorine vacancies, fluorine vacancy-clusters and fluorine armchair- and zigzag-clusters, will introduce different types of midgap states and extra excitations within the optical gap. Furthermore we argue that the local formation of $sp^3$ bonds upon fluorination can be distinguished from other disorder inducing mechanisms which do not destroy the $sp^2$ hybrid orbitals by measuring the polarization rotation of passing polarized light.

cond-mat.mes-hall↗

arXiv2014

Scaling of phloem structure and optimality of photoassimilate transport in conifer needles

The phloem vascular system facilitates transport of energy-rich sugar and signaling molecules in plants, thus permitting long range communication within the organism and growth of non-photosynthesizing organs such as roots and fruits. The flow is driven by osmotic pressure, generated by differences in sugar concentration between distal parts of the plant. The phloem is an intricate distribution system, and many questions about its regulation and structural diversity remain unanswered. Here, we investigate the phloem structure in the simplest possible geometry: a linear leaf, found, for example, in the needles of conifer trees. We measure the phloem structure in four tree species representing a diverse set of habitats and needle sizes, from 1 cm (\textit{Picea omorika}) to 35 cm (\textit{Pinus palustris}). We show that the phloem shares common traits across these four species and find that the size of its conductive elements obeys a power law. We present a minimal model that accounts for these common traits and takes into account the transport strategy and natural constraints. This minimal model predicts a power law phloem distribution consistent with transport energy minimization, suggesting that energetics are more important than translocation speed at the leaf level.

physics.bio-ph↗

arXiv2014

Converting Reconfigurable Petri Nets to Maude

Model checking is an important aim of the theoretical computer science. It enables the verification of a model with a set of properties such as liveness, deadlock or safety. One of the typical modelling techniques are Petri nets they are well understood and can be used for a model checking. Reconfigurable Petri nets are based on a Petri nets with a set of rules. These rules can be used dynamically to change the net. Missing is the possibility to verify a reconfigurable net and properties such as deadlocks or liveness. This paper introduces a conversion from reconfigurable Petri net to Maude, that allows us to fill the gap. It presents a net transformation approach which is based on Maude's equation- and rewrite logic as well as the LTLR model checker.

cs.FL↗

arXiv2014

Diffusion and bulk flow in phloem loading - a theoretical analysis of the polymer trap mechanism in plants

Plants create sugar in the mesophyll cells of their leaves by photosynthesis. This sugar, mostly sucrose, has to be loaded via the bundle sheath into the phloem vascular system (the sieve elements), where it is distributed to growing parts of the plant. We analyze the feasibility of a particular loading mechanism, active symplasmic loading, also called the polymer trap mechanism, where sucrose is transformed into heavier sugars, such as raffinose and stachyose, in the intermediary-type companion cells bordering the sieve elements in the minor veins of the phloem. Keeping the heavier sugars from diffusing back requires that the plasmodesmata connecting the bundle sheath with the intermediary cell act as extremely precise filters, which are able to distinguish between molecules that differ by less than 20% in size. In our modeling, we take into account the coupled water and sugar movement across the relevant interfaces, without explicitly considering the chemical reactions transforming the sucrose into the heavier sugars. Based on the available data for plasmodesmata geometry, sugar concentrations and flux rates, we conclude that this mechanism can in principle function. We find that the water flow through the plasmodesmata, which has not been quantified before, contributes only 10-20% to the sucrose flux into the intermediary cells, while the main part is transported by diffusion. On the other hand, the subsequent sugar translocation into the sieve elements would very likely be carried predominantly by bulk water flow through the plasmodesmata. Thus, in contrast to apoplasmic loaders, all the necessary water for phloem translocation would be supplied in this way with no need for additional water uptake across the plasma membranes of the phloem.

physics.bio-ph↗