Showing 1–28 of 28 results for author: Klotz, D

Energy-based Hopfield Boosting for Out-of-Distribution Detection

Authors: Claus Hofmann, Simon Schmid, Bernhard Lehner, Daniel Klotz, Sepp Hochreiter

Abstract: Out-of-distribution (OOD) detection is critical when deploying machine learning models in the real world. Outlier exposure methods, which incorporate auxiliary outlier data in the training process, can drastically improve OOD detection performance compared to approaches without advanced training strategies. We introduce Hopfield Boosting, a boosting approach, which leverages modern Hopfield energy… ▽ More Out-of-distribution (OOD) detection is critical when deploying machine learning models in the real world. Outlier exposure methods, which incorporate auxiliary outlier data in the training process, can drastically improve OOD detection performance compared to approaches without advanced training strategies. We introduce Hopfield Boosting, a boosting approach, which leverages modern Hopfield energy (MHE) to sharpen the decision boundary between the in-distribution and OOD data. Hopfield Boosting encourages the model to concentrate on hard-to-distinguish auxiliary outlier examples that lie close to the decision boundary between in-distribution and auxiliary outlier data. Our method achieves a new state-of-the-art in OOD detection with outlier exposure, improving the FPR95 metric from 2.28 to 0.92 on CIFAR-10 and from 11.76 to 7.94 on CIFAR-100. △ Less

Submitted 14 May, 2024; originally announced May 2024.

AI Increases Global Access to Reliable Flood Forecasts

Authors: Grey Nearing, Deborah Cohen, Vusumuzi Dube, Martin Gauch, Oren Gilon, Shaun Harrigan, Avinatan Hassidim, Daniel Klotz, Frederik Kratzert, Asher Metzger, Sella Nevo, Florian Pappenberger, Christel Prudhomme, Guy Shalev, Shlomo Shenzis, Tadele Tekalign, Dana Weitzner, Yoss Matias

Abstract: Floods are one of the most common natural disasters, with a disproportionate impact in developing countries that often lack dense streamflow gauge networks. Accurate and timely warnings are critical for mitigating flood risks, but hydrological simulation models typically must be calibrated to long data records in each watershed. Using AI, we achieve reliability in predicting extreme riverine event… ▽ More Floods are one of the most common natural disasters, with a disproportionate impact in developing countries that often lack dense streamflow gauge networks. Accurate and timely warnings are critical for mitigating flood risks, but hydrological simulation models typically must be calibrated to long data records in each watershed. Using AI, we achieve reliability in predicting extreme riverine events in ungauged watersheds at up to a 5-day lead time that is similar to or better than the reliability of nowcasts (0-day lead time) from a current state of the art global modeling system (the Copernicus Emergency Management Service Global Flood Awareness System). Additionally, we achieve accuracies over 5-year return period events that are similar to or better than current accuracies over 1-year return period events. This means that AI can provide flood warnings earlier and over larger and more impactful events in ungauged basins. The model developed in this paper was incorporated into an operational early warning system that produces publicly available (free and open) forecasts in real time in over 80 countries. This work highlights a need for increasing the availability of hydrological data to continue to improve global access to reliable flood warnings. △ Less

Submitted 3 November, 2023; v1 submitted 29 July, 2023; originally announced July 2023.

Mastering Nordschleife -- A comprehensive race simulation for AI strategy decision-making in motorsports

Authors: Max Boettinger, David Klotz

Abstract: In the realm of circuit motorsports, race strategy plays a pivotal role in determining race outcomes. This strategy focuses on the timing of pit stops, which are necessary due to fuel consumption and tire performance degradation. The objective of race strategy is to balance the advantages of pit stops, such as tire replacement and refueling, with the time loss incurred in the pit lane. Current rac… ▽ More In the realm of circuit motorsports, race strategy plays a pivotal role in determining race outcomes. This strategy focuses on the timing of pit stops, which are necessary due to fuel consumption and tire performance degradation. The objective of race strategy is to balance the advantages of pit stops, such as tire replacement and refueling, with the time loss incurred in the pit lane. Current race simulations, used to estimate the best possible race strategy, vary in granularity, modeling of probabilistic events, and require manual input for in-laps. This paper addresses these limitations by developing a novel simulation model tailored to GT racing and leveraging artificial intelligence to automate strategic decisions. By integrating the simulation with OpenAI's Gym framework, a reinforcement learning environment is created and an agent is trained. The study evaluates various hyperparameter configurations, observation spaces, and reward functions, drawing upon historical timing data from the 2020 Nürburgring Langstrecken Serie for empirical parameter validation. The results demonstrate the potential of reinforcement learning for improving race strategy decision-making, as the trained agent makes sensible decisions regarding pit stop timing and refueling amounts. Key parameters, such as learning rate, decay rate and the number of episodes, are identified as crucial factors, while the combination of fuel mass and current race position proves most effective for policy development. The paper contributes to the broader application of reinforcement learning in race simulations and unlocks the potential for race strategy optimization beyond FIA Formula~1, specifically in the GT racing domain. △ Less

Submitted 28 June, 2023; originally announced June 2023.

Conformal Prediction for Time Series with Modern Hopfield Networks

Authors: Andreas Auer, Martin Gauch, Daniel Klotz, Sepp Hochreiter

Abstract: To quantify uncertainty, conformal prediction methods are gaining continuously more interest and have already been successfully applied to various domains. However, they are difficult to apply to time series as the autocorrelative structure of time series violates basic assumptions required by conformal prediction. We propose HopCPT, a novel conformal prediction approach for time series that not o… ▽ More To quantify uncertainty, conformal prediction methods are gaining continuously more interest and have already been successfully applied to various domains. However, they are difficult to apply to time series as the autocorrelative structure of time series violates basic assumptions required by conformal prediction. We propose HopCPT, a novel conformal prediction approach for time series that not only copes with temporal structures but leverages them. We show that our approach is theoretically well justified for time series where temporal dependencies are present. In experiments, we demonstrate that our new approach outperforms state-of-the-art conformal prediction methods on multiple real-world time series datasets from four different domains. △ Less

Submitted 2 November, 2023; v1 submitted 22 March, 2023; originally announced March 2023.

Comments: presented at NeurIPS 2023

Few-Shot Learning by Dimensionality Reduction in Gradient Space

Authors: Martin Gauch, Maximilian Beck, Thomas Adler, Dmytro Kotsur, Stefan Fiel, Hamid Eghbal-zadeh, Johannes Brandstetter, Johannes Kofler, Markus Holzleitner, Werner Zellinger, Daniel Klotz, Sepp Hochreiter, Sebastian Lehner

Abstract: We introduce SubGD, a novel few-shot learning method which is based on the recent finding that stochastic gradient descent updates tend to live in a low-dimensional parameter subspace. In experimental and theoretical analyses, we show that models confined to a suitable predefined subspace generalize well for few-shot learning. A suitable subspace fulfills three criteria across the given tasks: it… ▽ More We introduce SubGD, a novel few-shot learning method which is based on the recent finding that stochastic gradient descent updates tend to live in a low-dimensional parameter subspace. In experimental and theoretical analyses, we show that models confined to a suitable predefined subspace generalize well for few-shot learning. A suitable subspace fulfills three criteria across the given tasks: it (a) allows to reduce the training error by gradient flow, (b) leads to models that generalize well, and (c) can be identified by stochastic gradient descent. SubGD identifies these subspaces from an eigendecomposition of the auto-correlation matrix of update directions across different tasks. Demonstrably, we can identify low-dimensional suitable subspaces for few-shot learning of dynamical systems, which have varying properties described by one or few parameters of the analytical system description. Such systems are ubiquitous among real-world applications in science and engineering. We experimentally corroborate the advantages of SubGD on three distinct dynamical systems problem settings, significantly outperforming popular few-shot learning methods both in terms of sample efficiency and performance. △ Less

Submitted 7 June, 2022; originally announced June 2022.

Comments: Accepted at Conference on Lifelong Learning Agents (CoLLAs) 2022. Code: https://github.com/ml-jku/subgd Blog post: https://ml-jku.github.io/subgd

Journal ref: Proceedings of The 1st Conference on Lifelong Learning Agents, PMLR 199:1043-1064 (2022)

MC-LSTM: Mass-Conserving LSTM

Authors: Pieter-Jan Hoedt, Frederik Kratzert, Daniel Klotz, Christina Halmich, Markus Holzleitner, Grey Nearing, Sepp Hochreiter, Günter Klambauer

Abstract: The success of Convolutional Neural Networks (CNNs) in computer vision is mainly driven by their strong inductive bias, which is strong enough to allow CNNs to solve vision-related tasks with random weights, meaning without learning. Similarly, Long Short-Term Memory (LSTM) has a strong inductive bias towards storing information over time. However, many real-world systems are governed by conservat… ▽ More The success of Convolutional Neural Networks (CNNs) in computer vision is mainly driven by their strong inductive bias, which is strong enough to allow CNNs to solve vision-related tasks with random weights, meaning without learning. Similarly, Long Short-Term Memory (LSTM) has a strong inductive bias towards storing information over time. However, many real-world systems are governed by conservation laws, which lead to the redistribution of particular quantities -- e.g. in physical and economical systems. Our novel Mass-Conserving LSTM (MC-LSTM) adheres to these conservation laws by extending the inductive bias of LSTM to model the redistribution of those stored quantities. MC-LSTMs set a new state-of-the-art for neural arithmetic units at learning arithmetic operations, such as addition tasks, which have a strong conservation law, as the sum is constant over time. Further, MC-LSTM is applied to traffic forecasting, modelling a pendulum, and a large benchmark dataset in hydrology, where it sets a new state-of-the-art for predicting peak flows. In the hydrology example, we show that MC-LSTM states correlate with real-world processes and are therefore interpretable. △ Less

Submitted 10 June, 2021; v1 submitted 13 January, 2021; originally announced January 2021.

Comments: 13 pages (8.5 without references) + 17 pages appendix

Uncertainty Estimation with Deep Learning for Rainfall-Runoff Modelling

Authors: Daniel Klotz, Frederik Kratzert, Martin Gauch, Alden Keefe Sampson, Günter Klambauer, Sepp Hochreiter, Grey Nearing

Abstract: Deep Learning is becoming an increasingly important way to produce accurate hydrological predictions across a wide range of spatial and temporal scales. Uncertainty estimations are critical for actionable hydrological forecasting, and while standardized community benchmarks are becoming an increasingly important part of hydrological model development and research, similar tools for benchmarking un… ▽ More Deep Learning is becoming an increasingly important way to produce accurate hydrological predictions across a wide range of spatial and temporal scales. Uncertainty estimations are critical for actionable hydrological forecasting, and while standardized community benchmarks are becoming an increasingly important part of hydrological model development and research, similar tools for benchmarking uncertainty estimation are lacking. We establish an uncertainty estimation benchmarking procedure and present four Deep Learning baselines, out of which three are based on Mixture Density Networks and one is based on Monte Carlo dropout. Additionally, we provide a post-hoc model analysis to put forward some qualitative understanding of the resulting models. Most importantly however, we show that accurate, precise, and reliable uncertainty estimation can be achieved with Deep Learning. △ Less

Submitted 15 December, 2020; originally announced December 2020.

Comments: 32 pages, 11 figures

MSC Class: 86A05 ACM Class: J.2.5

Different roles of Fe1-xNixOOH co-catalyst on hematite (α-Fe2O3) photoanodes with different dopants

Authors: Anton Tsyganok, Dino Klotz, Kirtiman Deo Malviya, Avner Rothschild, Daniel A Grave

Abstract: Transparent Fe1-xNixOOH overlayers (~2 nm thick) were deposited photoelectrochemically on (001) oriented heteroepitaxial Sn- and Zn-doped hematite (Fe2O3) thin film photoanodes. In both cases, the water photo-oxidation performance was improved by the co-catalyst overlayers. Intensity modulated photocurrent spectroscopy (IMPS) was applied to study the changes in the hole current and recombination c… ▽ More Transparent Fe1-xNixOOH overlayers (~2 nm thick) were deposited photoelectrochemically on (001) oriented heteroepitaxial Sn- and Zn-doped hematite (Fe2O3) thin film photoanodes. In both cases, the water photo-oxidation performance was improved by the co-catalyst overlayers. Intensity modulated photocurrent spectroscopy (IMPS) was applied to study the changes in the hole current and recombination current induced by the overlayers. For the Sn-doped hematite photoanode, the improvement in performance after deposition of the Fe1-xNixOOH overlayer was entirely due to reduction in the recombination current, leading to a cathodic shift in the onset potential. For the Zn-doped hematite photoanode, in addition to a reduction in recombination current, an increase in the hole current to the surface was also observed after the overlayer deposition, leading to a cathodic shift in the onset potential as well as an enhancement in the plateau photocurrent. These results demonstrate that Fe1-xNixOOH co-catalysts can play different roles depending on the underlying hematite photoanode. The effect of the co-catalyst is not always limited to changes in the surface properties, but also to an increase in hole current from the bulk to the surface that indicates a possible crosslink between surface and bulk processes. △ Less

Submitted 8 December, 2020; originally announced December 2020.

Journal ref: ACS Catal. 2018

Empirical Analysis of the Photoelectrochemical Impedance Response of Hematite Photoanodes for Water Photo-Oxidation

Authors: Dino Klotz, Daniel A. Grave, Hen Dotan, Avner Rothschild

Abstract: Photoelectrochemical impedance spectroscopy (PEIS) is a useful tool for the characterization of photoelectrodes for solar water splitting. However, the analysis of PEIS spectra often involves a priori assumptions that might bias the results. This work puts forward an empirical method that analyzes the distribution of relaxation times (DRT), obtained directly from the measured PEIS spectra of a mod… ▽ More Photoelectrochemical impedance spectroscopy (PEIS) is a useful tool for the characterization of photoelectrodes for solar water splitting. However, the analysis of PEIS spectra often involves a priori assumptions that might bias the results. This work puts forward an empirical method that analyzes the distribution of relaxation times (DRT), obtained directly from the measured PEIS spectra of a model hematite photoanode. By following how the DRT evolves as a function of control parameters such as the applied potential and composition of the electrolyte solution, we obtain unbiased insights into the underlying mechanisms that shape the photocurrent. In a subsequent step, we fit the data to a process-oriented equivalent circuit model (ECM) whose makeup is derived from the DRT analysis in the first step. This yields consistent quantitative trends of the dominant polarization processes observed. Our observations reveal a common step for the photo-oxidation reactions of water and H2O2 in alkaline solution △ Less

Submitted 8 December, 2020; originally announced December 2020.

Journal ref: J. Phys. Chem. Lett. 2018

The spatial collection efficiency of photogenerated charge carriers in photovoltaic and photoelectrochemical devices

Authors: Gideon Segev, Hen Dotan, David S. Ellis, Yifat Piekner, Dino Klotz, Jeffrey W. Beeman, Jason K. Cooper, Daniel A. Grave, Ian D. Sharp, Avner Rothschild

Abstract: The spatial collection efficiency portrays the driving forces and loss mechanisms in photovoltaic and photoelectrochemical devices. It is defined as the fraction of photogenerated charge carriers created at a specific point within the device that contribute to the photocurrent. In stratified planar structures, the spatial collection efficiency can be extracted out of photocurrent action spectra me… ▽ More The spatial collection efficiency portrays the driving forces and loss mechanisms in photovoltaic and photoelectrochemical devices. It is defined as the fraction of photogenerated charge carriers created at a specific point within the device that contribute to the photocurrent. In stratified planar structures, the spatial collection efficiency can be extracted out of photocurrent action spectra measurements empirically, with few a priori assumptions. Although this method was applied to photovoltaic cells made of well-understood materials, it has never been used to study unconventional materials such as metal-oxide semiconductors that are often employed in photoelectrochemical cells. This perspective shows the opportunities that this method has to offer for investigating new materials and devices with unknown properties. The relative simplicity of the method, and its applicability to operando performance characterization, makes it an important tool for analysis and design of new photovoltaic and photoelectrochemical materials and devices. △ Less

Submitted 8 December, 2020; originally announced December 2020.

Journal ref: Joule 2017

Accurate Determination of the Charge Transfer Efficiency of Photoanodes for Solar Water Splitting

Authors: Dino Klotz, Daniel A. Grave, Avner Rothschild

Abstract: The oxygen evolution reaction (OER) at the surface of semiconductor photoanodes involves photo-generated holes that oxidize water. A certain fraction of the holes that reach the surface recombine with electrons from the conduction band, giving rise to the surface recombination loss. The charge transfer efficiency, xt, defined as the ratio between the flux of holes that contribute to the water oxid… ▽ More The oxygen evolution reaction (OER) at the surface of semiconductor photoanodes involves photo-generated holes that oxidize water. A certain fraction of the holes that reach the surface recombine with electrons from the conduction band, giving rise to the surface recombination loss. The charge transfer efficiency, xt, defined as the ratio between the flux of holes that contribute to the water oxidation reaction and the total flux of holes that reach the surface, is an important parameter that helps to distinguish between bulk and surface recombination losses. However, accurate determination of xt by conventional voltammetry measurements is complicated because only the total current is measured and it is difficult to discern between different contributions to the current. Chopped light measurement and hole scavenger measurement techniques are widely employed to determine xt, but they often lead to errors. Intensity modulated photocurrent spectroscopy (IMPS) is better suited for accurate determination of xt because it provides direct information on both the total photocurrent and the surface recombination current. Careful analysis of IMPS measurements at different light intensities is required to account for nonlinear effects. We compare the xt values obtained by these methods using heteroepitaxial hematite photoanodes. A wide spread of xt values is obtained by different analysis methods and different light sources and light intensities. Statistical analysis of the results show good correlation between different methods for measurements carried out with the same light source, light intensity and potential. However, there is a considerable spread in the results obtained by different methods. For accurate determination of xt, we recommend IMPS measurements with a bias light intensity such that the irradiance is as close as possible to the standard solar spectrum. △ Less

Submitted 8 December, 2020; originally announced December 2020.

Journal ref: Phys. Chem. Chem. Phys.2017, 19, 20383-20392

Rainfall-Runoff Prediction at Multiple Timescales with a Single Long Short-Term Memory Network

Authors: Martin Gauch, Frederik Kratzert, Daniel Klotz, Grey Nearing, Jimmy Lin, Sepp Hochreiter

Abstract: Long Short-Term Memory Networks (LSTMs) have been applied to daily discharge prediction with remarkable success. Many practical scenarios, however, require predictions at more granular timescales. For instance, accurate prediction of short but extreme flood peaks can make a life-saving difference, yet such peaks may escape the coarse temporal resolution of daily predictions. Naively training an LS… ▽ More Long Short-Term Memory Networks (LSTMs) have been applied to daily discharge prediction with remarkable success. Many practical scenarios, however, require predictions at more granular timescales. For instance, accurate prediction of short but extreme flood peaks can make a life-saving difference, yet such peaks may escape the coarse temporal resolution of daily predictions. Naively training an LSTM on hourly data, however, entails very long input sequences that make learning hard and computationally expensive. In this study, we propose two Multi-Timescale LSTM (MTS-LSTM) architectures that jointly predict multiple timescales within one model, as they process long-past inputs at a single temporal resolution and branch out into each individual timescale for more recent input steps. We test these models on 516 basins across the continental United States and benchmark against the US National Water Model. Compared to naive prediction with a distinct LSTM per timescale, the multi-timescale architectures are computationally more efficient with no loss in accuracy. Beyond prediction quality, the multi-timescale LSTM can process different input variables at different timescales, which is especially relevant to operational applications where the lead time of meteorological forcings depends on their temporal resolution. △ Less

Submitted 15 October, 2020; originally announced October 2020.

Journal ref: Hydrol. Earth Syst. Sci., 25, 2045-2062, 2021

Estimating action plans for smart poultry houses

Authors: Darlan Felipe Klotz, Richardson Ribeiro, Fabrício Enembreck, Gustavo Denardin, Marco Barbosa, Dalcimar Casanova, Marcelo Teixeira

Abstract: In poultry farming, the systematic choice, update, and implementation of periodic (t) action plans define the feed conversion rate (FCR[t]), which is an acceptable measure for successful production. Appropriate action plans provide tailored resources for broilers, allowing them to grow within the so-called thermal comfort zone, without wast or lack of resources. Although the implementation of an a… ▽ More In poultry farming, the systematic choice, update, and implementation of periodic (t) action plans define the feed conversion rate (FCR[t]), which is an acceptable measure for successful production. Appropriate action plans provide tailored resources for broilers, allowing them to grow within the so-called thermal comfort zone, without wast or lack of resources. Although the implementation of an action plan is automatic, its configuration depends on the knowledge of the specialist, tending to be inefficient and error-prone, besides to result in different FCR[t] for each poultry house. In this article, we claim that the specialist's perception can be reproduced, to some extent, by computational intelligence. By combining deep learning and genetic algorithm techniques, we show how action plans can adapt their performance over the time, based on previous well succeeded plans. We also implement a distributed network infrastructure that allows to replicate our method over distributed poultry houses, for their smart, interconnected, and adaptive control. A supervision system is provided as interface to users. Experiments conducted over real data show that our method improves 5% on the performance of the most productive specialist, staying very close to the optimal FCR[t]. △ Less

Submitted 17 August, 2020; originally announced August 2020.

Comments: To be submitted to the journal Expert Systems with Applications

MSC Class: 68T35 (Primary) 93C40 (Secondary) ACM Class: C.3; C.5; I.2; J.6

Accurate Hydrologic Modeling Using Less Information

Authors: Guy Shalev, Ran El-Yaniv, Daniel Klotz, Frederik Kratzert, Asher Metzger, Sella Nevo

Abstract: Joint models are a common and important tool in the intersection of machine learning and the physical sciences, particularly in contexts where real-world measurements are scarce. Recent developments in rainfall-runoff modeling, one of the prime challenges in hydrology, show the value of a joint model with shared representation in this important context. However, current state-of-the-art models dep… ▽ More Joint models are a common and important tool in the intersection of machine learning and the physical sciences, particularly in contexts where real-world measurements are scarce. Recent developments in rainfall-runoff modeling, one of the prime challenges in hydrology, show the value of a joint model with shared representation in this important context. However, current state-of-the-art models depend on detailed and reliable attributes characterizing each site to help the model differentiate correctly between the behavior of different sites. This dependency can present a challenge in data-poor regions. In this paper, we show that we can replace the need for such location-specific attributes with a completely data-driven learned embedding, and match previous state-of-the-art results with less information. △ Less

Submitted 21 November, 2019; originally announced November 2019.

Using LSTMs for climate change assessment studies on droughts and floods

Authors: Frederik Kratzert, Daniel Klotz, Johannes Brandstetter, Pieter-Jan Hoedt, Grey Nearing, Sepp Hochreiter

Abstract: Climate change affects occurrences of floods and droughts worldwide. However, predicting climate impacts over individual watersheds is difficult, primarily because accurate hydrological forecasts require models that are calibrated to past data. In this work we present a large-scale LSTM-based modeling approach that -- by training on large data sets -- learns a diversity of hydrological behaviors.… ▽ More Climate change affects occurrences of floods and droughts worldwide. However, predicting climate impacts over individual watersheds is difficult, primarily because accurate hydrological forecasts require models that are calibrated to past data. In this work we present a large-scale LSTM-based modeling approach that -- by training on large data sets -- learns a diversity of hydrological behaviors. Previous work shows that this model is more accurate than current state-of-the-art models, even when the LSTM-based approach operates out-of-sample and the latter in-sample. In this work, we show how this model can assess the sensitivity of the underlying systems with regard to extreme (high and low) flows in individual watersheds over the continental US. △ Less

Submitted 28 November, 2019; v1 submitted 10 November, 2019; originally announced November 2019.

Towards Learning Universal, Regional, and Local Hydrological Behaviors via Machine-Learning Applied to Large-Sample Datasets

Authors: Frederik Kratzert, Daniel Klotz, Guy Shalev, Günter Klambauer, Sepp Hochreiter, Grey Nearing

Abstract: Regional rainfall-runoff modeling is an old but still mostly out-standing problem in Hydrological Sciences. The problem currently is that traditional hydrological models degrade significantly in performance when calibrated for multiple basins together instead of for a single basin alone. In this paper, we propose a novel, data-driven approach using Long Short-Term Memory networks (LSTMs), and demo… ▽ More Regional rainfall-runoff modeling is an old but still mostly out-standing problem in Hydrological Sciences. The problem currently is that traditional hydrological models degrade significantly in performance when calibrated for multiple basins together instead of for a single basin alone. In this paper, we propose a novel, data-driven approach using Long Short-Term Memory networks (LSTMs), and demonstrate that under a 'big data' paradigm, this is not necessarily the case. By training a single LSTM model on 531 basins from the CAMELS data set using meteorological time series data and static catchment attributes, we were able to significantly improve performance compared to a set of several different hydrological benchmark models. Our proposed approach not only significantly outperforms hydrological models that were calibrated regionally but also achieves better performance than hydrological models that were calibrated for each basin individually. Furthermore, we propose an adaption to the standard LSTM architecture, which we call an Entity-Aware-LSTM (EA-LSTM), that allows for learning, and embedding as a feature layer in a deep learning model, catchment similarities. We show that this learned catchment similarity corresponds well with what we would expect from prior hydrological understanding. △ Less

Submitted 10 November, 2019; v1 submitted 19 July, 2019; originally announced July 2019.

NeuralHydrology -- Interpreting LSTMs in Hydrology

Authors: Frederik Kratzert, Mathew Herrnegger, Daniel Klotz, Sepp Hochreiter, Günter Klambauer

Abstract: Despite the huge success of Long Short-Term Memory networks, their applications in environmental sciences are scarce. We argue that one reason is the difficulty to interpret the internals of trained networks. In this study, we look at the application of LSTMs for rainfall-runoff forecasting, one of the central tasks in the field of hydrology, in which the river discharge has to be predicted from m… ▽ More Despite the huge success of Long Short-Term Memory networks, their applications in environmental sciences are scarce. We argue that one reason is the difficulty to interpret the internals of trained networks. In this study, we look at the application of LSTMs for rainfall-runoff forecasting, one of the central tasks in the field of hydrology, in which the river discharge has to be predicted from meteorological observations. LSTMs are particularly well-suited for this problem since memory cells can represent dynamic reservoirs and storages, which are essential components in state-space modelling approaches of the hydrological system. On basis of two different catchments, one with snow influence and one without, we demonstrate how the trained model can be analyzed and interpreted. In the process, we show that the network internally learns to represent patterns that are consistent with our qualitative understanding of the hydrological system. △ Less

Submitted 12 November, 2019; v1 submitted 19 March, 2019; originally announced March 2019.

Comments: Pre-print of published book chapter. See journal reference and DOI for more info

Journal ref: In: Samek W., Montavon G., Vedaldi A., Hansen L., Muller KR. (eds) Explainable AI: Interpreting, Explaining and Visualizing Deep Learning. Lecture Notes in Computer Science, vol 11700. Springer, Cham, 2019

The VLTI/MIDI view on the inner mass loss of evolved stars from the Herschel MESS sample

Authors: C. Paladini, D. Klotz, S. Sacuto, E. Lagadec, M. Wittkowski, A. Richichi, J. Hron, A. Jorissen, M. A. T. Groenewegen, F. Kerschbaum, T. Verhoelst, G. Rau, H. Olofsson, R. Zhao-Geisler, A. Matter

Abstract: The mass-loss process from evolved stars is a key ingredient for our understanding of many fields of astrophysics, including stellar evolution and the chemical enrichment of the interstellar medium via stellar yields. One the main unsolved questions is the geometry of the mass-loss process. Taking advantage of the results from the Herschel Mass loss of Evolved StarS (MESS) programme, we initiated… ▽ More The mass-loss process from evolved stars is a key ingredient for our understanding of many fields of astrophysics, including stellar evolution and the chemical enrichment of the interstellar medium via stellar yields. One the main unsolved questions is the geometry of the mass-loss process. Taking advantage of the results from the Herschel Mass loss of Evolved StarS (MESS) programme, we initiated a coordinated effort to characterise the geometry of mass loss from evolved red giants at various spatial scales. For this purpose we used the MID-infrared interferometric Instrument (MIDI) to resolve the inner envelope of 14 asymptotic giant branch stars (AGBs) in the MESS sample. In this contribution we present an overview of the interferometric data collected within the frame of our Large Programme, and we also add archive data for completeness. We studied the geometry of the inner atmosphere by comparing the observations with predictions from different geometric models. Asymmetries are detected for five O-rich and S-type, suggesting that asymmetries in the N band are more common among stars with such chemistry. We speculate that this fact is related to the characteristics of the dust grains. Except for one star, no interferometric variability is detected, i.e. the changes in size of the shells of non-mira stars correspond to changes of the visibility of less than 10%. The observed spectral variability confirms previous findings from the literature. The detection of dust in our sample follows the location of the AGBs in the IRAS colour-colour diagram: more dust is detected around oxygen-rich stars in region II and in the carbon stars in region VII. The SiC dust feature does not appear in the visibility spectrum of UAnt and SSct, which are two carbon stars with detached shells. This finding has implications for the theory of SiC dust formation. △ Less

Submitted 26 January, 2017; v1 submitted 19 January, 2017; originally announced January 2017.

Comments: 43 pages, 31 figures; accepted for publication in Astronomy & Astrophysics. Abstract shortened for compilation reasons. Metadata corrected

Journal ref: A&A 600, A136 (2017)

The complex environment of the bright carbon star TX Psc as probed by spectro-astrometry

Authors: J. Hron, S. Uttenthaler, B. Aringer, D. Klotz, T. Lebzelter, C. Paladini, G. Wiedemann

Abstract: Context: Stars on the asymptotic giant branch (AGB) show broad evidence of inhomogeneous atmospheres and circumstellar envelopes. These have been studied by a variety of methods on various angular scales. In this paper we explore the envelope of the well-studied carbon star TX Psc by the technique of spectro-astrometry. Aims: We explore the potential of this method for detecting asymmetries around… ▽ More Context: Stars on the asymptotic giant branch (AGB) show broad evidence of inhomogeneous atmospheres and circumstellar envelopes. These have been studied by a variety of methods on various angular scales. In this paper we explore the envelope of the well-studied carbon star TX Psc by the technique of spectro-astrometry. Aims: We explore the potential of this method for detecting asymmetries around AGB stars. Methods:We obtained CRIRES observations of several CO $Δ$v=1 lines near 4.6 $μ$m and HCN lines near 3 $μ$m in 2010 and 2013. These were then searched for spectro-astrometric signatures. For the interpretation of the results, we used simple simulated observations. Results: Several lines show significant photocentre shifts with a clear dependence on position angle. In all cases, tilde-shaped signatures are found where the positive and negative shifts (at PA 0deg) are associated with blue and weaker red components of the lines. The shifts can be modelled with a bright blob 70 mas to 210 mas south of the star with a flux of several percent of the photospheric flux. We estimate a lower limit of the blob temperature of 1000 K. The blob may be related to a mass ejection as found for AGB stars or red supergiants. We also consider the scenario of a companion object. Conclusions: Although there is clear spectro-astrometric evidence of a rather prominent structure near TX Psc, it does not seem to relate to the other evidence of asymmetries, so no definite explanation can be given. Our data thus underline the very complex structure of the environment of this star, but further observations that sample the angular scales out to a few hundred milli-arcseconds are needed to get a clearer picture. △ Less

Submitted 9 October, 2015; originally announced October 2015.

Journal ref: A&A 584, A27 (2015)

Dissecting the AGB star L2 Puppis: a torus in the making

Authors: F. Lykou, D. Klotz, C. Paladini, J. Hron, A. A. Zijlstra, J. Kluska, B. R. M. Norris, P. G. Tuthill, S. Ramstedt, E. Lagadec, M. Wittkowski, M. Maercker, J. Meisner, A. Mayer

Abstract: The circumstellar environment of L2 Pup, an oxygen-rich semiregular variable, was observed to understand the evolution of mass loss and the shaping of ejecta in the late stages of stellar evolution. High-angular resolution observations from a single 8 m telescope were obtained using aperture masking in the near-infrared (1.64, 2.30 and 3.74 $\rmμm$) on the NACO/VLT, both in imaging and polarimetri… ▽ More The circumstellar environment of L2 Pup, an oxygen-rich semiregular variable, was observed to understand the evolution of mass loss and the shaping of ejecta in the late stages of stellar evolution. High-angular resolution observations from a single 8 m telescope were obtained using aperture masking in the near-infrared (1.64, 2.30 and 3.74 $\rmμm$) on the NACO/VLT, both in imaging and polarimetric modes. The aperture-masking images of L2 Pup at 2.30 $\rmμm$ show a resolved structure that resembles a toroidal structure with a major axis of ~140 milliarcseconds (mas) and an east-west orientation. Two clumps can be seen on either side of the star, ~65 mas from the star, beyond the edge of the circumstellar envelope (estimated diameter is ~27 mas), while a faint, hook-like structure appear toward the northeast. The patterns are visible both in the imaging and polarimetric mode, although the latter was only used to measure the total intensity (Stokes I). The overall shape of the structure is similar at the 3.74 $\rmμm$ pseudo-continuum (dust emission), where the clumps appear to be embedded within a dark, dusty lane. The faint, hook-like patterns are also seen at this wavelength, extending northeast and southwest with the central, dark lane being an apparent axis of symmetry. We interpret the structure as a circumstellar torus with inner radius of 4.2 au. With a rotation velocity of 10 km s$^{-1}$ as suggested by the SiO maser profile, we estimate a stellar mass of 0.7 M$_\odot$.} △ Less

Submitted 17 March, 2015; originally announced March 2015.

Comments: 9 pages, 7 figures, to be published in A&A

Journal ref: A&A 576, A46 (2015)

Large-scale environments of binary AGB stars probed by Herschel. II: Two companions interacting with the wind of pi1 Gruis

Authors: A. Mayer, A. Jorissen, C. Paladini, F. Kerschbaum, D. Pourbaix, C. Siopis, R. Ottensamer, M. Mečina, N. L. J. Cox, M. A. T. Groenewegen, D. Klotz, G. Sadowski, A. Spang, P. Cruzalèbes, C. Waelkens

Abstract: Context. The Mass loss of Evolved StarS (MESS) sample observed with PACS on board the Herschel Space Observatory revealed that several asymptotic giant branch (AGB) stars are surrounded by an asymmetric circumstellar envelope (CSE) whose morphology is most likely caused by the interaction with a stellar companion. The evolution of AGB stars in binary systems plays a crucial role in understanding t… ▽ More Context. The Mass loss of Evolved StarS (MESS) sample observed with PACS on board the Herschel Space Observatory revealed that several asymptotic giant branch (AGB) stars are surrounded by an asymmetric circumstellar envelope (CSE) whose morphology is most likely caused by the interaction with a stellar companion. The evolution of AGB stars in binary systems plays a crucial role in understanding the formation of asymmetries in planetary nebulæ (PNe), but at present, only a handful of cases are known where the interaction of a companion with the stellar AGB wind is observed. Aims. We probe the environment of the very evolved AGB star $π^1$ Gruis on large and small scales to identify the triggers of the observed asymmetries. Methods. Observations made with Herschel/PACS at 70 $μ$m and 160 $μ$m picture the large-scale environment of $π^1$ Gru. The close surroundings of the star are probed by interferometric observations from the VLTI/AMBER archive. An analysis of the proper motion data of Hipparcos and Tycho-2 together with the Hipparcos Intermediate Astrometric Data help identify the possible cause for the observed asymmetry. Results. The Herschel/PACS images of $π^1$ Gru show an elliptical CSE whose properties agree with those derived from a CO map published in the literature. In addition, an arc east of the star is visible at a distance of $38^{\prime\prime}$ from the primary. This arc is most likely part of an Archimedean spiral caused by an already known G0V companion that is orbiting the primary at a projected distance of 460 au with a period of more than 6200 yr. However, the presence of the elliptical CSE, proper motion variations, and geometric modelling of the VLTI/AMBER observations point towards a third component in the system, with an orbital period shorter than 10 yr, orbiting much closer to the primary than the G0V star. △ Less

Submitted 25 August, 2014; v1 submitted 18 August, 2014; originally announced August 2014.

Comments: 13 pages, 11 figures, accepted for publication in Astronomy & Astrophysics

Journal ref: A&A 570, A113 (2014)

The wind of the M-type AGB star RT Virginis probed by VLTI/MIDI

Authors: Stéphane Sacuto, Sofia Ramstedt, Susanne Höfner, Hans Olofsson, Sara Bladh, Kjell Eriksson, Bernhard Aringer, Daniela Klotz, Matthias Maercker

Abstract: We study the circumstellar environment of the M-type AGB star RT Vir using mid-infrared high spatial resolution observations from the ESO-VLTI focal instrument MIDI. The aim of this study is to provide observational constraints on theoretical prediction that the winds of M-type AGB objects can be driven by photon scattering on iron-free silicate grains located in the close environment (about 2 to… ▽ More We study the circumstellar environment of the M-type AGB star RT Vir using mid-infrared high spatial resolution observations from the ESO-VLTI focal instrument MIDI. The aim of this study is to provide observational constraints on theoretical prediction that the winds of M-type AGB objects can be driven by photon scattering on iron-free silicate grains located in the close environment (about 2 to 3 stellar radii) of the star. We interpreted spectro-interferometric data, first using wavelength-dependent geometric models. We then used a self-consistent dynamic model atmosphere containing a time-dependent description of grain growth for pure forsterite dust particles to reproduce the photometric, spectrometric, and interferometric measurements of RT Vir. Since the hydrodynamic computation needs stellar parameters as input, a considerable effort was first made to determine these parameters. MIDI differential phases reveal the presence of an asymmetry in the stellar vicinity. Results from the geometrical modeling give us clues to the presence of aluminum and silicate dust in the close circumstellar environment (< ~5 stellar radii). Comparison between spectro-interferometric data and a self-consistent dust-driven wind model reveals that silicate dust has to be present in the region between 2 to 3 stellar radii to reproduce the 59 and 63 m baseline visibility measurements around 9.8 micron. This gives additional observational evidence in favor of winds driven by photon scattering on iron-free silicate grains located in the close vicinity of an M-type star. However, other sources of opacity are clearly missing to reproduce the 10-13 micron visibility measurements for all baselines. This study is a first attempt to understand the wind mechanism of M-type AGB stars by comparing photometric, spectrometric, and interferometric measurements with state-of-the-art, self-consistent dust-driven wind models. The agreement of the dynamic model atmosphere with interferometric measurements in the 8-10 micron spectral region gives additional observational evidence that the winds of M-type stars can be driven by photon scattering on iron-free silicate grains. Finally, a larger statistical study and progress in advanced self-consistent 3D modeling are still required to solve the remaining problems. △ Less

Submitted 24 January, 2013; originally announced January 2013.

Comments: Astronomy and Astrophysics (2013)

Catching the fish - Constraining stellar parameters for TX Psc using spectro-interferometric observations

Authors: D. Klotz, C. Paladini, J. Hron, B. Aringer, S. Sacuto, P. Marigo, T. Verhoelst

Abstract: Stellar parameter determination is a challenging task when dealing with galactic giant stars. The combination of different investigation techniques has proven to be a promising approach. We analyse archive spectra obtained with the Short-Wavelength-Spectrometer (SWS) onboard of ISO, and new interferometric observations from the Very Large Telescope MID-infrared Interferometric instrument (VLTI/MID… ▽ More Stellar parameter determination is a challenging task when dealing with galactic giant stars. The combination of different investigation techniques has proven to be a promising approach. We analyse archive spectra obtained with the Short-Wavelength-Spectrometer (SWS) onboard of ISO, and new interferometric observations from the Very Large Telescope MID-infrared Interferometric instrument (VLTI/MIDI) of a very well studied carbon-rich giant: TX Psc. The aim of this work is to determine stellar parameters using spectroscopy and interferometry. The observations are used to constrain the model atmosphere, and eventually the stellar evolutionary model in the region where the tracks map the beginning of the carbon star sequence. Two different approaches are used to determine stellar parameters: (i) the 'classic' interferometric approach where the effective temperature is fixed by using the angular diameter in the N-band (from interferometry) and the apparent bolometric magnitude; (ii) parameters are obtained by fitting a grid of state-of-the-art hydrostatic models to spectroscopic and interferometric observations. We find a good agreement between the parameters of the two methods. The effective temperature and luminosity clearly place TX Psc in the carbon-rich AGB star domain in the H-R-diagram. Current evolutionary tracks suggest that TX Psc became a C-star just recently, which means that the star is still in a 'quiet' phase compared to the subsequent strong-wind regime. This is in agreement with the C/O ratio being only slightly larger than 1. △ Less

Submitted 3 January, 2013; originally announced January 2013.

Comments: 11 pages, 9 figures, 5 tables

Detection of an asymmetry in the envelope of the carbon Mira R Fornacis using VLTI/MIDI

Authors: C. Paladini, S. Sacuto, D. Klotz, K. Ohnaka, M. Wittkowski, W. Nowotny, A. Jorissen, J. Hron

Abstract: Aims. We present a study of the envelope morphology of the carbon Mira R For with VLTI/MIDI. This object is one of the few asymptotic giant branch (AGB) stars that underwent a dust-obscuration event. The cause of such events is still a matter of discussion. Several symmetric and asymmetric scenarios have been suggested in the literature. Methods. Mid-infrared interferometric observations were obta… ▽ More Aims. We present a study of the envelope morphology of the carbon Mira R For with VLTI/MIDI. This object is one of the few asymptotic giant branch (AGB) stars that underwent a dust-obscuration event. The cause of such events is still a matter of discussion. Several symmetric and asymmetric scenarios have been suggested in the literature. Methods. Mid-infrared interferometric observations were obtained separated by two years. The observations probe different depths of the atmosphere and cover different pulsation phases. The visibilities and the differential phases were interpreted using GEM-FIND, a tool for fitting spectrally dispersed interferometric observations with the help of wavelength-dependent geometric models. Results. We report the detection of an asymmetric structure revealed through the MIDI differential phase. This asymmetry is observed at the same baseline and position angle two years later. The observations are best simulated with a model that includes a uniform-disc plus a Gaussian envelope plus a point-source. The geometric model can reproduce both the visibilities and the differential phase signatures. Conclusions. Our MIDI data favour explanations of the R For obscuration event that are based on an asymmetric geometry. We clearly detect a photocentre shift between the star and the strongly resolved dust component. This might be caused by a dust clump or a substellar companion. However, the available observations do not allow us to distinguish between the two options. The finding has strong implications for future studies of the geometry of the envelope of AGB stars: if this is a binary, are all AGB stars that show an obscuration event binaries as well? Or are we looking at asymmetric mass-loss processes (i.e. dusty clumps) in the inner part of a carbon-rich Mira? △ Less

Submitted 19 July, 2012; v1 submitted 17 July, 2012; originally announced July 2012.

Comments: accepted for publication as A&A letter

Geometrical model fitting for interferometric data: GEM-FIND

Authors: D. Klotz, S. Sacuto, C. Paladini, J. Hron, G. Wachter

Abstract: We developed the tool GEM-FIND that allows to constrain the morphology and brightness distribution of objects. The software fits geometrical models to spectrally dispersed interferometric visibility measurements in the N-band using the Levenberg-Marquardt minimization method. Each geometrical model describes the brightness distribution of the object in the Fourier space using a set of wavelength-i… ▽ More We developed the tool GEM-FIND that allows to constrain the morphology and brightness distribution of objects. The software fits geometrical models to spectrally dispersed interferometric visibility measurements in the N-band using the Levenberg-Marquardt minimization method. Each geometrical model describes the brightness distribution of the object in the Fourier space using a set of wavelength-independent and/or wavelength-dependent parameters. In this contribution we numerically analyze the stability of our nonlinear fitting approach by applying it to sets of synthetic visibilities with statistically applied errors, answering the following questions: How stable is the parameter determination with respect to (i) the number of uv-points, (ii) the distribution of points in the uv-plane, (iii) the noise level of the observations? △ Less

Submitted 9 July, 2012; originally announced July 2012.

Comments: 14 pages, 12 figures; to appear in SPIE Astronomical Telescopes and Instrumentation, SPIE Conference Series 8445

The geometry of the close environment of SV Psc as probed by VLTI/MIDI

Authors: D. Klotz, S. Sacuto, F. Kerschbaum, C. Paladini, H. Olofsson, J. Hron

Abstract: Context. SV Psc is an asymptotic giant branch (AGB) star surrounded by an oxygen-rich dust envelope. The mm-CO line profile of the object's outflow shows a clear double-component structure. Because of the high angular resolution, mid-IR interferometry may give strong constraints on the origin of this composite profile. Aims. The aim of this work is to investigate the morphology of the environmen… ▽ More Context. SV Psc is an asymptotic giant branch (AGB) star surrounded by an oxygen-rich dust envelope. The mm-CO line profile of the object's outflow shows a clear double-component structure. Because of the high angular resolution, mid-IR interferometry may give strong constraints on the origin of this composite profile. Aims. The aim of this work is to investigate the morphology of the environment around SV Psc using high-angular resolution interferometry observations in the mid-IR with the Very Large Telescope MID-infrared Interferometric instrument (VLTI/MIDI). Methods. Interferometric data in the N-band taken at different baseline lengths (ranging from 32-64 m) and position angles (73- 142°) allow a study of the morphology of the circumstellar environment close to the star. The data are interpreted on the basis of 2-dimensional, chromatic geometrical models using the fitting software tool GEM-FIND developed for this purpose. Results. The results favor two scenarios: (i) the presence of a highly inclined, optically thin, dusty disk surrounding the central star; (ii) the presence of an unresolved binary companion at a separation of 13.7 AU and a position angle of 121.8° NE. The derived orbital period of the binary is 38.1 yr. This detection is in good agreement with hydrodynamic simulations showing that a close companion could be responsible for the entrainment of the gas and dust into a circumbinary structure. △ Less

Submitted 23 April, 2012; originally announced April 2012.

Comments: 10 pages, 12 figures

Detection of the 69 μm band of crystalline forsterite in the Herschel MESS-program

Authors: B. L. de Vries, D. Klotz, R. Lombaert, A. Baier, J. A. D. L. Blommaert, L. Decin, F. Kerschbaum, W. Nowotny, T. Posch, H. Van Winckel, M. A. T. Groenewegen, T. Ueta, G. Van de Steene, B. Vandenbussche, P. Royer, C. Waelkens

Abstract: In this article we present the detection of the 69 μm band of the crystalline olivine forsterite within the MESS key program of Herschel. We determine the temperature of the forsterite grains by fitting the 69 μm band. In this article we present the detection of the 69 μm band of the crystalline olivine forsterite within the MESS key program of Herschel. We determine the temperature of the forsterite grains by fitting the 69 μm band. △ Less

Submitted 4 November, 2010; originally announced November 2010.

Comments: Conference proceedings: "Why Galaxies Care About AGB Stars 2", Vienna 2010

Modernising the ESRF control system with GNU/Linux

Authors: A. Gotz, A. Homs, B. Regad, M. Perez, P. Makijarvi, W. D. Klotz

Abstract: he ESRF control system is in the process of being modernised. The present contrsystem is based on VME, 10 MHz Ethernet, OS9, Solaris, HP-UX, NFS/RPC, Motif and C. The new control system will be based on compact PCI, 100 MHz Ethernet, Linux, Windows, Solaris, CORBA/IIOP, C++, Java and Python. The main frontend operating system will be GNU/Linux running on Intel/x86 and Motorola/68k. Linux will al… ▽ More he ESRF control system is in the process of being modernised. The present contrsystem is based on VME, 10 MHz Ethernet, OS9, Solaris, HP-UX, NFS/RPC, Motif and C. The new control system will be based on compact PCI, 100 MHz Ethernet, Linux, Windows, Solaris, CORBA/IIOP, C++, Java and Python. The main frontend operating system will be GNU/Linux running on Intel/x86 and Motorola/68k. Linux will also be used on handheld devices for mobile control. This poster describes how GNU/Linux is being used to modernise the control system and what problems have been encountered so far △ Less

Submitted 9 November, 2001; originally announced November 2001.

Comments: 3 pages

Report number: WEAP023 ACM Class: C.3

Journal ref: eConf C011127 (2001) WEAP023