arXiv subjects

Maximilian Schmidt

Publications and source records attributed to Maximilian Schmidt.

27 records · Page 2Linked to original sources

arXiv2019

Deep Relevance Regularization: Interpretable and Robust Tumor Typing of Imaging Mass Spectrometry Data

Neural networks have recently been established as a viable classification method for imaging mass spectrometry data for tumor typing. For multi-laboratory scenarios however, certain confounding factors may strongly impede their performance. In this work, we introduce Deep Relevance Regularization, a method of restricting what the neural network can focus on during classification, in order to improve the classification performance. We demonstrate how Deep Relevance Regularization robustifies neural networks against confounding factors on a challenging inter-lab dataset consisting of breast and ovarian carcinoma. We further show that this makes the relevance map -- a way of visualizing the discriminative parts of the mass spectrum -- sparser, thereby making the classifier easier to interpret

cs.LG↗

arXiv2019

Integrality of Seshadri constants and irreducibility of principal polarizations on products of two isogenous elliptic curves

In this paper we consider the question of when all Seshadri constants on a product of two isogenous elliptic curves $E_1\times E_2$ without complex multiplication are integers. By studying elliptic curves on $E_1\times E_2$ we translate this question into a purely numerical problem expressed by quadratic forms. By solving that problem, we show that all Seshadri constants on $E_1\times E_2$ are integers if and only if the minimal degree of an isogeny $E_1\to E_2$ equals 1 or 2. Furthermore, this method enables a characterization of irreducible principal polarizations on $E_1\times E_2$.

math.AG↗

arXiv2019

On the integrality of Seshadri constants of abelian surfaces

In this paper we consider the question of when Seshadri constants on abelian surfaces are integers. Our first result concerns self-products $E\times E$ of elliptic curves: If $E$ has complex multiplication in $\Z[i]$ or in $\Z[\frac12(1+i\sqrt3)]$ or if $E$ has no complex multiplication at all, then it is known that for every ample line bundle $L$ on $E\times E$, the Seshadri constant $\eps(L)$ is an integer. We show that, contrary to what one might expect, these are in fact the only elliptic curves for which this integrality statement holds. Our second result answers the question how -- on any abelian surface~-- integrality of Seshadri constants is related to elliptic curves.

math.AG↗

arXiv2019

Normalizing flows for novelty detection in industrial time series data

Flow-based deep generative models learn data distributions by transforming a simple base distribution into a complex distribution via a set of invertible transformations. Due to the invertibility, such models can score unseen data samples by computing their exact likelihood under the learned distribution. This makes flow-based models a perfect tool for novelty detection, an anomaly detection technique where unseen data samples are classified as normal or abnormal by scoring them against a learned model of normal data. We show that normalizing flows can be used as novelty detectors in time series. Two flow-based models, Masked Autoregressive Flows and Free-form Jacobian of Reversible Dynamics restricted by autoregressive MADE networks, are tested on synthetic data and motor current data from an industrial machine and achieve good results, outperforming a conventional novelty detection method, the Local Outlier Factor.

cs.LG↗

arXiv2018

Dating glacier ice of the last millennium by quantum technology

Radiometric dating with 39Ar covers a unique timespan and offers key advances in interpreting environmental archives of the last millennium. Although this tracer has been acknowledged for decades, studies so far have been limited by the low abundance and radioactivity, thus requiring huge sample sizes. Atom Trap Trace Analysis, an application of techniques from quantum physics such as laser cooling and trapping, allows to reduce the sample volume by several orders of magnitude, compared to conventional techniques. Here we show that the adaptation of this method to 39Ar is now available for glaciological applications, by demonstrating the first Argon Trap Trace Analysis (ArTTA) dating of alpine glacier ice samples. Ice blocks as small as a few kilograms are sufficient and have been obtained at two artificial glacier caves. Importantly, both sites offer direct access to the stratigraphy at the glacier base and validation against existing age constraints. The ice blocks obtained at Chli Titlis glacier in 3030 m all (Swiss Alps) have been dated by state-of-the-art micro-radiocarbon analysis in a previous study. The unique finding of a bark fragment and a larch needle within the ice of Schaufelferner glacier in 2870 m asl (Stubai Alps, Austria) allows for conventional radiocarbon dating. At both sites, results of 39Ar dating match the existing age information based on radiocarbon dating and visual stratigraphy. With our results, we establish Argon Trap Trace Analysis as the key to decipher so far untapped glacier archives of the last millennium.

physics.app-ph↗

arXiv2017

Fundamental activity constraints lead to specific interpretations of the connectome

The continuous integration of experimental data into coherent models of the brain is an increasing challenge of modern neuroscience. Such models provide a bridge between structure and activity, and identify the mechanisms giving rise to experimental observations. Nevertheless, structurally realistic network models of spiking neurons are necessarily underconstrained even if experimental data on brain connectivity are incorporated to the best of our knowledge. Guided by physiological observations, any model must therefore explore the parameter ranges within the uncertainty of the data. Based on simulation results alone, however, the mechanisms underlying stable and physiologically realistic activity often remain obscure. We here employ a mean-field reduction of the dynamics, which allows us to include activity constraints into the process of model construction. We shape the phase space of a multi-scale network model of the vision-related areas of macaque cortex by systematically refining its connectivity. Fundamental constraints on the activity, i.e., prohibiting quiescence and requiring global stability, prove sufficient to obtain realistic layer- and area-specific activity. Only small adaptations of the structure are required, showing that the network operates close to an instability. The procedure identifies components of the network critical to its collective dynamics and creates hypotheses for structural data and future experiments. The method can be applied to networks involving any neuron model with a known gain function.

q-bio.NC↗

arXiv2016

Full-density multi-scale account of structure and dynamics of macaque visual cortex

We present a multi-scale spiking network model of all vision-related areas of macaque cortex that represents each area by a full-scale microcircuit with area-specific architecture. The layer- and population-resolved network connectivity integrates axonal tracing data from the CoCoMac database with recent quantitative tracing data, and is systematically refined using dynamical constraints. Simulations reveal a stable asynchronous irregular ground state with heterogeneous activity across areas, layers, and populations. Elicited by large-scale interactions, the model reproduces longer intrinsic time scales in higher compared to early visual areas. Activity propagates down the visual hierarchy, similar to experimental results associated with visual imagery. Cortico-cortical interaction patterns agree well with fMRI resting-state functional connectivity. The model bridges the gap between local and large-scale accounts of cortex, and clarifies how the detailed connectivity of cortex shapes its dynamics on multiple scales.

q-bio.NC↗

arXiv2014

Extracting partial decay rates of helium from complex rotation: autoionizing resonances of the one-dimensional configurations

Partial autoionization rates of doubly excited one-dimensional helium in the collinear Zee and eZe configuration are obtained by means of the complex rotation method. The approach presented here relies on a projection of back-rotated resonance wave functions onto singly ionized $\textrm{He}^{+}$ channel wave functions and the computation of the corresponding particle fluxes. In spite of the long-range nature of the Coulomb potential between the electrons and the nucleus, an asymptotic region where the fluxes are stationary is clearly observed. Low-lying doubly excited states are found to decay predomintantly into the nearest single-ionization continuum. This approach paves the way for a systematic analysis of the decay rates observed in higher-dimensional models, and of the role of electronic correlations and atomic structure in recent photoionization experiments.

physics.atom-ph↗

arXiv2012

Nonthermal fixed points and solitons in a one-dimensional Bose gas

Single-particle momentum spectra for a dynamically evolving one-dimensional Bose gas are analysed in the semi-classical wave limit. Representing one of the simplest correlation functions these give information about possible universal scaling behaviour. Motivated by the previously discovered connection between (quasi-)topological field configurations, strong wave turbulence, and nonthermal fixed points of quantum field dynamics, soliton formation is studied with respect to the appearance of transient power-law spectra. A random-soliton model is developed to describe the spectra analytically, and the analogies and difference between the appearing power laws and those found in a field theory approach to strong wave turbulence are discussed. The results open a view on solitary wave dynamics from the point of view of critical phenomena far from thermal equilibrium and on a possibility to study this dynamics in experiment without the necessity of detecting solitons in situ.

cond-mat.quant-gas↗