arXiv subjects

Johannes Otterbach

Publications and source records attributed to Johannes Otterbach.

14 recordsLinked to original sources

arXiv2024

Anomalous Directed Percolation on a Dynamic Network using Rydberg Facilitation

The facilitation of Rydberg excitations in a gas of atoms provides an ideal model system to study epidemic evolution on (dynamic) networks and self organization of complex systems to the critical point of a non-equilibrium phase transition. Using Monte-Carlo simulations and a machine learning algorithm we show that the universality class of this phase transition can be tuned. The classes include directed percolation (DP), the most common class in short-range spreading models, and mean-field (MF) behavior, but also different types of anomalous directed percolation (ADP), characterized by rare long-range excitation processes. In a frozen gas, ground state atoms that can facilitate each other form a static network, for which we predict DP universality. Atomic motion then turns the network into a dynamic one with long-range (Levy-flight type) excitations. This leads to continuously varying critical exponents corresponding to the ADP universality class, eventually reaching MF behavior. These findings also explain the recently observed critical exponent of Rydberg facilitation in an ultra-cold gas experiment [Helmrich et al., Nature 577, 481 (2020)], which was in between DP and MF values.

cond-mat.quant-gas↗

arXiv2023

Self-Distilled Representation Learning for Time Series

Self-supervised learning for time-series data holds potential similar to that recently unleashed in Natural Language Processing and Computer Vision. While most existing works in this area focus on contrastive learning, we propose a conceptually simple yet powerful non-contrastive approach, based on the data2vec self-distillation framework. The core of our method is a student-teacher scheme that predicts the latent representation of an input time series from masked views of the same time series. This strategy avoids strong modality-specific assumptions and biases typically introduced by the design of contrastive sample pairs. We demonstrate the competitiveness of our approach for classification and forecasting as downstream tasks, comparing with state-of-the-art self-supervised learning methods on the UCR and UEA archives as well as the ETT and Electricity datasets.

cs.LG↗

arXiv2023

Uncovering the Inner Workings of STEGO for Safe Unsupervised Semantic Segmentation

Self-supervised pre-training strategies have recently shown impressive results for training general-purpose feature extraction backbones in computer vision. In combination with the Vision Transformer architecture, the DINO self-distillation technique has interesting emerging properties, such as unsupervised clustering in the latent space and semantic correspondences of the produced features without using explicit human-annotated labels. The STEGO method for unsupervised semantic segmentation contrastively distills feature correspondences of a DINO-pre-trained Vision Transformer and recently set a new state of the art. However, the detailed workings of STEGO have yet to be disentangled, preventing its usage in safety-critical applications. This paper provides a deeper understanding of the STEGO architecture and training strategy by conducting studies that uncover the working mechanisms behind STEGO, reproduce and extend its experimental validation, and investigate the ability of STEGO to transfer to different datasets. Results demonstrate that the STEGO architecture can be interpreted as a semantics-preserving dimensionality reduction technique.

cs.CV↗

arXiv2021

MEAL: Manifold Embedding-based Active Learning

Image segmentation is a common and challenging task in autonomous driving. Availability of sufficient pixel-level annotations for the training data is a hurdle. Active learning helps learning from small amounts of data by suggesting the most promising samples for labeling. In this work, we propose a new pool-based method for active learning, which proposes promising patches extracted from full image, in each acquisition step. The problem is framed in an exploration-exploitation framework by combining an embedding based on Uniform Manifold Approximation to model representativeness with entropy as uncertainty measure to model informativeness. We applied our proposed method to the autonomous driving datasets CamVid and Cityscapes and performed a quantitative comparison with state-of-the-art baselines. We find that our active learning method achieves better performance compared to previous methods.

cs.CV↗

arXiv2021

DAAIN: Detection of Anomalous and Adversarial Input using Normalizing Flows

Despite much recent work, detecting out-of-distribution (OOD) inputs and adversarial attacks (AA) for computer vision models remains a challenge. In this work, we introduce a novel technique, DAAIN, to detect OOD inputs and AA for image segmentation in a unified setting. Our approach monitors the inner workings of a neural network and learns a density estimator of the activation distribution. We equip the density estimator with a classification head to discriminate between regular and anomalous inputs. To deal with the high-dimensional activation-space of typical segmentation networks, we subsample them to obtain a homogeneous spatial and layer-wise coverage. The subsampling pattern is chosen once per monitored model and kept fixed for all inputs. Since the attacker has access to neither the detection model nor the sampling key, it becomes harder for them to attack the segmentation network, as the attack cannot be backpropagated through the detector. We demonstrate the effectiveness of our approach using an ESPNet trained on the Cityscapes dataset as segmentation model, an affine Normalizing Flow as density estimator and use blue noise to ensure homogeneous sampling. Our model can be trained on a single GPU making it compute efficient and deployable without requiring specialized accelerators.

cs.CV↗

arXiv2021

Chameleon: A Semi-AutoML framework targeting quick and scalable development and deployment of production-ready ML systems for SMEs

Developing, scaling, and deploying modern Machine Learning solutions remains challenging for small- and middle-sized enterprises (SMEs). This is due to a high entry barrier of building and maintaining a dedicated IT team as well as the difficulties of real-world data (RWD) compared to standard benchmark data. To address this challenge, we discuss the implementation and concepts of Chameleon, a semi-AutoML framework. The goal of Chameleon is fast and scalable development and deployment of production-ready machine learning systems into the workflow of SMEs. We first discuss the RWD challenges faced by SMEs. After, we outline the central part of the framework which is a model and loss-function zoo with RWD-relevant defaults. Subsequently, we present how one can use a templatable framework in order to automate the experiment iteration cycle, as well as close the gap between development and deployment. Finally, we touch on our testing framework component allowing us to investigate common model failure modes and support best practices of model deployment governance.

cs.SE↗

arXiv2016

Interacting in-plane molecular dipoles in a zig-zag chain

The system with externally polarized dipole molecules at half-filling moving along a one-dimensional zig-zag chain is studied theoretically, including the ground-state phase diagram. The dipoles are oriented in-plane. Together with the geometry of the chain this gives rise to a bond-alternating nearest neighbor interaction due to simultaneous attractive and repulsive interactions. Because of the quantum Zeno effect due to the reactive nature of molecules the system can be treated as hard-core. By tuning the ratio between the nearest-neighbor interaction and hopping, various phases can be accessed by controlling the polarization angle. In the ultra-strong coupling limit, the system simplifies to a frustrated extended axial Ising model. For the small coupling limit, qualitative discussion of the ordering behavior using effective field theory arguments is provided. We show that when chain angle is small, the system mostly exhibits BKT-type phase transitions, whereas large chain angle would drive the system into a gapped (Ising) dimerized phase, where the hopping strength is closely related to the orientation of dimerized pairs.

cond-mat.quant-gas↗

arXiv2014

Dissipative Preparation of Spatial Order in Rydberg-Dressed Bose-Einstein Condensates

We propose a technique for engineering momentum-dependent dissipation in Bose-Einstein condensates with non-local interactions. The scheme relies on the use of momentum-dependent dark-states in close analogy to velocity-selective coherent population trapping. During the short-time dissipative dynamics, the system is driven into a particular finite-momentum phonon mode, which in real space corresponds to an ordered structure with non-local density-density correlations. Dissipation-induced ordering can be observed and studied in present-day experiments using cold atoms with dipole-dipole or off-resonant Rydberg interactions. Due to its dissipative nature, the ordering does not require artificial breaking of translational symmetry by an opticallattice or harmonic trap. This opens up a perspective of direct cooling of quantum gases into strongly-interacting phases.

cond-mat.quant-gas↗

arXiv2013

Wigner Crystallization of Single Photons in Cold Rydberg Ensemble

The coupling of weak light fields to Rydberg states of atoms under conditions of electromagnetically induced transparency (EIT) leads to the formation of Rydberg polaritons which are quasi-particles with tunable effective mass and long-range interactions. Confined to one spatial dimension their low energy physics is that of a moving-frame Luttinger liquid which due to the long-range character of the repulsive interaction can form a Wigner crystal. We calculate the Luttinger $K$ parameter using density-matrix renormalization group (DMRG) simulations and find that under typical slow-light conditions kinetic energy contributions are too strong for crystal formation. However, adiabatically increasing the polariton mass by turning a light pulse into stationary spin excitations allows to generate true crystalline order over a finite length. The dynamics of this process and asymptotic correlations are analyzed in terms of a time-dependent Luttinger theory.

quant-ph↗

arXiv2013

Dissipative Preparation of Spin Squeezed Atomic Ensembles in a Steady State

We present and analyze a new approach for the generation of atomic spin squeezed states. Our method involves the collective coupling of an atomic ensemble to a decaying mode of an open optical cavity. We demonstrate the existence of a collective atomic dark-state, decoupled from the radiation field. By explicitly constructing this state we find that it can feature spin squeezing bounded only by the Heisenberg limit. We show that such dark states can be deterministically prepared via dissipative means, thus turning dissipation into a resource for entanglement. The scaling of the phase sensitivity taking realistic imperfections into account is discussed.

quant-ph↗

arXiv2011

Electromagnetically induced transparency with Rydberg atoms

We present a theory of electromagnetically induced transparency in a cold ensemble of strongly interacting Rydberg atoms. Long-range interactions between the atoms constrain the medium to behave as a collection of superatoms, each comprising a blockade volume that can accommodate at most one Rydberg excitation. The propagation of a probe field is affected by its two-photon correlations within the blockade distance, which are strongly damped due to low saturation threshold of the superatoms. Our model is computationally very efficient and is in quantitative agreement with the results of recent experiment of Pritchard et al. [Phys. Rev. Lett. 105, 193603 (2010)].

quant-ph↗

arXiv2011

Photon-Photon Interactions via Rydberg Blockade

We develop the theory of light propagation under the conditions of electromagnetically induced transparency (EIT) in systems involving strongly interacting Rydberg states. Taking into account the quantum nature and the spatial propagation of light, we analyze interactions involving few-photon pulses. We demonstrate that this system can be used for the generation of nonclassical states of light including trains of single photons with an avoided volume between them, for implementing photon-photon quantum gates, as well as for studying many-body phenomena with strongly correlated photons.

quant-ph↗

arXiv2010

Photonic Phase Gate via an Exchange of Fermionic Spin Waves in a Spin Chain

We propose a new protocol for implementing the two-qubit photonic phase gate. In our approach, the pi phase is acquired by mapping two single photons into atomic excitations with fermionic character and exchanging their positions. The fermionic excitations are realized as spin waves in a spin chain, while photon storage techniques provide the interface between the photons and the spin waves. Possible imperfections and experimental systems suitable for implementing the gate are discussed.

quant-ph↗

arXiv2008

Bose-Einstein condensation of stationary-light polaritons

We propose and analyze a mechanism for Bose-Einstein condensation of stationary dark-state polaritons. Dark-state polaritons (DSPs) are formed in the interaction of light with laser-driven 3-level Lambda-type atoms and are the basis of phenomena such as electromagnetically induced transparency (EIT), ultra-slow and stored light. They have long intrinsic lifetimes and in a stationary set-up with two counterpropagating control fields of equal intensity have a 3D quadratic dispersion profile with variable effective mass. Since DSPs are bosons they can undergo a Bose-Einstein condensation at a critical temperature which can be many orders of magnitude larger than that of atoms. We show that thermalization of polaritons can occur via elastic collisions mediated by a resonantly enhanced optical Kerr nonlinearity on a time scale short compared to the decay time. Finally condensation can be observed by turning stationary into propagating polaritons and monitoring the emitted light.

quant-ph↗