arXiv subjects

Philipp Schmidt

Publications and source records attributed to Philipp Schmidt.

30 records · Page 2Linked to original sources

arXiv2024

Virtual Reality for Understanding Artificial-Intelligence-driven Scientific Discovery with an Application in Quantum Optics

Generative Artificial Intelligence (AI) models can propose solutions to scientific problems beyond human capability. To truly make conceptual contributions, researchers need to be capable of understanding the AI-generated structures and extracting the underlying concepts and ideas. When algorithms provide little explanatory reasoning alongside the output, scientists have to reverse-engineer the fundamental insights behind proposals based solely on examples. This task can be challenging as the output is often highly complex and thus not immediately accessible to humans. In this work we show how transferring part of the analysis process into an immersive Virtual Reality (VR) environment can assist researchers in developing an understanding of AI-generated solutions. We demonstrate the usefulness of VR in finding interpretable configurations of abstract graphs, representing Quantum Optics experiments. Thereby, we can manually discover new generalizations of AI-discoveries as well as new understanding in experimental quantum optics. Furthermore, it allows us to customize the search space in an informed way - as a human-in-the-loop - to achieve significantly faster subsequent discovery iterations. As concrete examples, with this technology, we discover a new resource-efficient 3-dimensional entanglement swapping scheme, as well as a 3-dimensional 4-particle Greenberger-Horne-Zeilinger-state analyzer. Our results show the potential of VR for increasing a human researcher's ability to derive knowledge from graph-based generative AI that, which is a common abstract data representation used in diverse fields of science.

cs.HC↗

arXiv2023

Tuning the supercurrent distribution in parallel ballistic graphene Josephson junctions

We report on a ballistic and fully tunable Josephson junction system consisting of two parallel ribbons of graphene in contact with superconducting MoRe. By electrostatic gating of the two individual graphene ribbons we gain control over the real space distribution of the superconducting current density, which can be continuously tuned between both ribbons. We extract the respective gate dependent spatial distributions of the real space current density by employing Fourier- and Hilbert transformations of the magnetic field induced modulation of the critical current. This approach is fast and does not rely on a symmetric current profile. It is therefore a universally applicable tool, potentially useful for carefully adjusting Josephson junctions.

cond-mat.mes-hall↗

arXiv2023

Multiple-core-hole resonance spectroscopy with ultraintense X-ray pulses

Understanding the interaction of intense, femtosecond X-ray pulses with heavy atoms is crucial for gaining insights into the structure and dynamics of matter. One key aspect of nonlinear light-matter interaction was, so far, not studied systematically at free-electron lasers -- its dependence on the photon energy. Using resonant ion spectroscopy, we map out the transient electronic structures occurring during the complex charge-up pathways. Massively hollow atoms featuring up to six simultaneous core holes determine the spectra at specific photon energies and charge states. We also illustrate how the influence of different X-ray pulse parameters that are usually intertwined can be partially disentangled. The extraction of resonance spectra is facilitated by the fact that the ion yields become independent of the peak fluence beyond a saturation point. Our study lays the groundwork for novel spectroscopies of transient atomic species in exotic, multiple-core-hole states that have not been explored previously.

physics.atom-ph↗

arXiv2023

Variational Boosted Soft Trees

Gradient boosting machines (GBMs) based on decision trees consistently demonstrate state-of-the-art results on regression and classification tasks with tabular data, often outperforming deep neural networks. However, these models do not provide well-calibrated predictive uncertainties, which prevents their use for decision making in high-risk applications. The Bayesian treatment is known to improve predictive uncertainty calibration, but previously proposed Bayesian GBM methods are either computationally expensive, or resort to crude approximations. Variational inference is often used to implement Bayesian neural networks, but is difficult to apply to GBMs, because the decision trees used as weak learners are non-differentiable. In this paper, we propose to implement Bayesian GBMs using variational inference with soft decision trees, a fully differentiable alternative to standard decision trees introduced by Irsoy et al. Our experiments demonstrate that variational soft trees and variational soft GBMs provide useful uncertainty estimates, while retaining good predictive performance. The proposed models show higher test likelihoods when compared to the state-of-the-art Bayesian GBMs in 7/10 tabular regression datasets and improved out-of-distribution detection in 5/10 datasets.

cs.LG↗

arXiv2022

Electronic Quantum Coherence in Glycine Molecules Probed with Ultrashort X-ray Pulses in Real Time

Quantum coherence between electronic states of a photoionized molecule and the resulting process of ultrafast electron-hole migration have been put forward as a possible quantum mechanism of charge-directed reactivity governing the photoionization-induced molecular decomposition. Attosecond experiments based on the indirect (fragment ion-based) characterization of the proposed electronic phenomena suggest that the photoionization-induced electronic coherence can survive for tens of femtoseconds, while some theoretical studies predict much faster decay of the coherence due to the quantum uncertainty in the nuclear positions and the nuclear-motion effects. The open questions are: do long-lived electronic quantum coherences exist in complex molecules and can they be probed directly, i.e. via electronic observables? Here, we use x-rays both to create and to directly probe quantum coherence in the photoionized amino acid glycine. The outgoing photoelectron wave leaves behind a positively charged ion that is in a coherent superposition of quantum mechanical eigenstates lying within the ionizing pulse spectral bandwidth. Delayed x-ray pulses track the induced coherence through resonant x-ray absorption that induces Auger decay and by the photoelectron emission from sequential double photoionization. Sinusoidal temporal modulation of the detected signal at early times (0 - 25 fs) is observed in both measurements. Advanced ab initio many-electron simulations, taking into account the quantum uncertainty in the nuclear positions, allow us to explain the first 25 fs of the detected coherent quantum evolution in terms of the electronic coherence.

physics.chem-ph↗

arXiv2021

More Than Words: Towards Better Quality Interpretations of Text Classifiers

The large size and complex decision mechanisms of state-of-the-art text classifiers make it difficult for humans to understand their predictions, leading to a potential lack of trust by the users. These issues have led to the adoption of methods like SHAP and Integrated Gradients to explain classification decisions by assigning importance scores to input tokens. However, prior work, using different randomization tests, has shown that interpretations generated by these methods may not be robust. For instance, models making the same predictions on the test set may still lead to different feature importance rankings. In order to address the lack of robustness of token-based interpretability, we explore explanations at higher semantic levels like sentences. We use computational metrics and human subject studies to compare the quality of sentence-based interpretations against token-based ones. Our experiments show that higher-level feature attributions offer several advantages: 1) they are more robust as measured by the randomization tests, 2) they lead to lower variability when using approximation-based methods like SHAP, and 3) they are more intelligible to humans in situations where the linguistic coherence resides at a higher granularity level. Based on these findings, we show that token-based interpretability, while being a convenient first choice given the input interfaces of the ML models, is not the most effective one in all situations.

cs.CL↗

arXiv2021

Resonance-enhanced multiphoton ionization in the x-ray regime

Here, we report on the nonlinear ionization of argon atoms in the short wavelength regime using ultraintense x rays from the European XFEL. After sequential multiphoton ionization, high charge states are obtained. For photon energies that are insufficient to directly ionize a $1s$ electron, a different mechanism is required to obtain ionization to Ar$^{17+}$. We propose this occurs through a two-color process where the second harmonic of the FEL pulse resonantly excites the system via a $1s \rightarrow 2p$ transition followed by ionization by the fundamental FEL pulse, which is a type of x-ray resonance-enhanced multiphoton ionization (REMPI). This resonant phenomenon occurs not only for Ar$^{16+}$, but through multiple lower charge states, where multiple ionization competes with decay lifetimes, making x-ray REMPI distinctive from conventional REMPI. With the aid of state-of-the-art theoretical calculations, we explain the effects of x-ray REMPI on the relevant ion yields and spectral profile.

physics.atom-ph↗

arXiv2021

Suppression of X-Ray-Induced Radiation Damage to Biomolecules in Aqueous Environments by Immediate Intermolecular Decay of Inner-Shell Vacancies

The predominant reason for the damaging power of high-energy radiation is multiple ionization of a molecule, either direct or via the decay of highly excited intermediates, as e.g., in the case of X-ray irradiation. Consequently, the molecule is irreparably damaged by the subsequent fragmentation in a Coulomb explosion. In an aqueous environment, however, it has been observed that irradiated molecules may be saved from fragmentation presumably by charge and energy dissipation mechanisms. Here, we show that the protective effect of the environment sets in even earlier than hitherto expected, namely immediately after single inner-shell ionization. By combining coincidence measurements of the fragmentation of X-ray-irradiated microsolvated pyrimidine molecules with theoretical calculations, we identify direct intermolecular electronic decay as the protective mechanism, outrunning the usually dominant Auger decay. Our results demonstrate that such processes play a key role in charge delocalization and have to be considered in investigations and models on high-energy radiation damage in realistic environments.

physics.chem-ph↗

arXiv2020

Site-specific Interrogation of an Ionic Chiral Fragment During Photolysis Using an X-ray Free-Electron Laser

Short-wavelength free-electron lasers with their ultrashort pulses at high intensities have originated new approaches for tracking molecular dynamics from the vista of specific sites. X-ray pump X-ray probe schemes even allow to address individual atomic constituents with a 'trigger'-event that preludes the subsequent molecular dynamics while being able to selectively probe the evolving structure with a time-delayed second X-ray pulse. Here, we use a linearly polarized X-ray photon to trigger the photolysis of a prototypical chiral molecule, namely trifluoromethyloxirane (C$_3$H$_3$F$_3$O), at the fluorine K-edge at around 700 eV. The evolving fluorine-containing fragments are then probed by a second, circularly polarized X-ray pulse of higher photon energy in order to investigate the chemically shifted inner-shell electrons of the ionic motherfragment for their stereochemical sensitivity. We experimentally demonstrate and theoretically support how two-color X-ray pump X-ray probe experiments with polarization control enable XFELs as tools for chiral recognition.

physics.chem-ph↗

arXiv2019

Quantifying Interpretability and Trust in Machine Learning Systems

Decisions by Machine Learning (ML) models have become ubiquitous. Trusting these decisions requires understanding how algorithms take them. Hence interpretability methods for ML are an active focus of research. A central problem in this context is that both the quality of interpretability methods as well as trust in ML predictions are difficult to measure. Yet evaluations, comparisons and improvements of trust and interpretability require quantifiable measures. Here we propose a quantitative measure for the quality of interpretability methods. Based on that we derive a quantitative measure of trust in ML decisions. Building on previous work we propose to measure intuitive understanding of algorithmic decisions using the information transfer rate at which humans replicate ML model predictions. We provide empirical evidence from crowdsourcing experiments that the proposed metric robustly differentiates interpretability methods. The proposed metric also demonstrates the value of interpretability for ML assisted human decision making: in our experiments providing explanations more than doubled productivity in annotation tasks. However unbiased human judgement is critical for doctors, judges, policy makers and others. Here we derive a trust metric that identifies when human decisions are overly biased towards ML predictions. Our results complement existing qualitative work on trust and interpretability by quantifiable measures that can serve as objectives for further improving methods in this field of research.

cs.LG↗

arXiv2013

Long-range last-passage percolation on the line

We consider directed last-passage percolation on the random graph G = (V,E) where V = Z and each edge (i,j), for i < j, is present in E independently with some probability 0 < p <= 1. To every present edge (i,j) we attach i.i.d. random weights v_{i,j} > 0. We are interested in the behaviour of w_{0,n}, which is the maximum weight of all directed paths from 0 to n, as n tends to infinity. We see two very different types of behaviour, depending on whether E[v_{i,j}^2] is finite or infinite. In the case where E[v_{i,j}^2] is finite we show that the process has a certain regenerative structure, and prove a strong law of large numbers and, under an extra assumption, a functional central limit theorem. In the situation where E[v_{i,j}^2] is infinite we obtain scaling laws and asymptotic distributions expressed in terms of a "continuous last-passage percolation" model on [0,1]; these are related to corresponding results for two-dimensional last-passage percolation with heavy-tailed weights obtained by Hambly and Martin.

math.PR↗

arXiv2010

Multi-type TASEP in discrete time

The TASEP (totally asymmetric simple exclusion process) is a basic model for an one-dimensional interacting particle system with non-reversible dynamics. Despite the simplicity of the model it shows a very rich and interesting behaviour. In this paper we study some aspects of the TASEP in discrete time and compare the results to the recently obtained results for the TASEP in continuous time. In particular we focus on stationary distributions for multi-type models, speeds of second-class particles, collision probabilities and the "speed process". In discrete time, jump attempts may occur at different sites simultaneously, and the order in which these attempts are processed is important; we consider various natural update rules.

math.PR↗