Search

For the first time the neutron-rich hydrogen isotope $\rm ^{6}H$ was produced in an electron scattering experiment in the reaction $\rm ^{7}Li(e,~e'p\pi^{+})^{6}H$ using the spectrometer facility of the A1 Collaboration at the Mainz Microtron accelerator. By measuring the triple coincidence between the scattered electron, the produced proton, and $\pi^{+}$, the missing mass spectrum of $\rm ^{6}H$ was obtained. A clear peak above $^3$H+n+n+n energy threshold was seen resulting in a ground state energy of $\rm ^{6}H$ at $2.3\pm0.5({\rm stat.})\pm0.4({\rm syst.})$ MeV with a width of $1.9\pm1.0({\rm stat.})\pm0.4({\rm syst.})$ MeV. This work challenges the understandings of multi-nucleon interactions and presents a new method to study light neutron-rich nuclei with electron scattering experiments., Comment: 6 pages, 4 figures

Bayesian inference in function space has gained attention due to its robustness against overparameterization in neural networks. However, approximating the infinite-dimensional function space introduces several challenges. In this work, we discuss function space inference via particle optimization and present practical modifications that improve uncertainty estimation and, most importantly, make it applicable for large and pretrained networks. First, we demonstrate that the input samples, where particle predictions are enforced to be diverse, are detrimental to the model performance. While diversity on training data itself can lead to underfitting, the use of label-destroying data augmentation, or unlabeled out-of-distribution data can improve prediction diversity and uncertainty estimates. Furthermore, we take advantage of the function space formulation, which imposes no restrictions on network parameterization other than sufficient flexibility. Instead of using full deep ensembles to represent particles, we propose a single multi-headed network that introduces a minimal increase in parameters and computation. This allows seamless integration to pretrained networks, where this repulsive last-layer ensemble can be used for uncertainty aware fine-tuning at minimal additional cost. We achieve competitive results in disentangling aleatoric and epistemic uncertainty for active learning, detecting out-of-domain data, and providing calibrated uncertainty estimates under distribution shifts with minimal compute and memory.

Randomization is a powerful tool that endows algorithms with remarkable properties. For instance, randomized algorithms excel in adversarial settings, often surpassing the worst-case performance of deterministic algorithms with large margins. Furthermore, their success probability can be amplified by simple strategies such as repetition and majority voting. In this paper, we enhance deep neural networks, in particular transformer models, with randomization. We demonstrate for the first time that randomized algorithms can be instilled in transformers through learning, in a purely data- and objective-driven manner. First, we analyze known adversarial objectives for which randomized algorithms offer a distinct advantage over deterministic ones. We then show that common optimization techniques, such as gradient descent or evolutionary strategies, can effectively learn transformer parameters that make use of the randomness provided to the model. To illustrate the broad applicability of randomization in empowering neural networks, we study three conceptual tasks: associative recall, graph coloring, and agents that explore grid worlds. In addition to demonstrating increased robustness against oblivious adversaries through learned randomization, our experiments reveal remarkable performance improvements due to the inherently random nature of the neural networks' computation and predictions.

The emergence of spatial coherence in a confined two-dimensional Bose gas of exciton-polaritons with tuneable interactions offers a unique opportunity to explore the role of interactions in a phase transition in a driven-dissipative quantum system, where both the phase transition and thermalisation are mediated by interactions. We investigate, experimentally and numerically, the phase correlations and steady-state properties of the gas over a wide range of interaction strengths by varying the photonic/excitonic fraction of the polaritons and their density. We find that the first order spatial coherence function exhibits algebraic decay consistent with the Berezinskii-Kosterlitz-Thouless (BKT) phase transition. Surprisingly, the exponent of the algebraic decay is inversely proportional to the coherent density of polaritons, in analogy to equilibrium superfluids above the BKT transition, but with a different proportionality constant. Our work paves the way for future investigations of the phenomenon of phase transitions and superfluidity in a driven-dissipative setting, Comment: 11 pages, 6 figures

We present a careful calibration of the exciton fraction of polaritons in high-$Q$ ($\sim 300,000$), long-lifetime ($\sim 300$ ps), GaAs/AlGaAs microcavities.This is a crucial parameter for many-body theories which include the polariton-polariton interactions.It is much harder to establish this number in high-$Q$ structures compared to low-$Q$ structures, because the upper polariton is nearly invisible in high-$Q$ cavities.We present a combination of photoluminescence, photoluminescence excitation, and reflectivity measurements to highly constrain the fit model, and compare the results of this model to the results from low-$Q$ structures.We present a fitted curve of exciton fraction as a function of the lower polariton energy for multiple samples which have been used in prior experiments., Comment: 16 pages, 11 figures

Exciton-polariton laser is a promising source of coherent light for low-energy applications due to its low-threshold operation. However, a detailed experimental study of its spectral purity, which directly affects its coherence properties is still missing. Here}, we present a high-resolution spectroscopic investigation of the energy and linewidth of an exciton-polariton laser in the single-mode regime, which derives its coherent emission from an optically pumped and confined exciton-polariton condensate. We report an ultra-narrow linewidth of 56~MHz or 0.24~$\mu$eV, corresponding to a coherence time of 5.7~ns. The narrow linewidth is consistently achieved by using an exciton-polariton condensate with a high photonic content confined in an optically induced trap. Contrary to previous studies, we show that the excitonic reservoir created by the pump and responsible for creating the trap does not strongly affect the emission linewidth as long as the condensate is trapped and the pump power is well above the condensation (lasing) threshold. \red{The long coherence time of the exciton-polariton system uncovered here opens up opportunities for manipulating its macroscopic quantum state, which is essential for applications in classical and quantum computing.

Segmentation is a critical step in analyzing the developing human fetal brain. There have been vast improvements in automatic segmentation methods in the past several years, and the Fetal Brain Tissue Annotation (FeTA) Challenge 2021 helped to establish an excellent standard of fetal brain segmentation. However, FeTA 2021 was a single center study, and the generalizability of algorithms across different imaging centers remains unsolved, limiting real-world clinical applicability. The multi-center FeTA Challenge 2022 focuses on advancing the generalizability of fetal brain segmentation algorithms for magnetic resonance imaging (MRI). In FeTA 2022, the training dataset contained images and corresponding manually annotated multi-class labels from two imaging centers, and the testing data contained images from these two imaging centers as well as two additional unseen centers. The data from different centers varied in many aspects, including scanners used, imaging parameters, and fetal brain super-resolution algorithms applied. 16 teams participated in the challenge, and 17 algorithms were evaluated. Here, a detailed overview and analysis of the challenge results are provided, focusing on the generalizability of the submissions. Both in- and out of domain, the white matter and ventricles were segmented with the highest accuracy, while the most challenging structure remains the cerebral cortex due to anatomical complexity. The FeTA Challenge 2022 was able to successfully evaluate and advance generalizability of multi-class fetal brain tissue segmentation algorithms for MRI and it continues to benchmark new algorithms. The resulting new methods contribute to improving the analysis of brain development in utero., Comment: Results from FeTA Challenge 2022, held at MICCAI; Manuscript submitted to IEEE Transactions on Medical Imaging (2024). Supplementary Info (including submission methods descriptions) available here: https://zenodo.org/records/10628648

Many complex tasks can be decomposed into simpler, independent parts. Discovering such underlying compositional structure has the potential to enable compositional generalization. Despite progress, our most powerful systems struggle to compose flexibly. It therefore seems natural to make models more modular to help capture the compositional nature of many tasks. However, it is unclear under which circumstances modular systems can discover hidden compositional structure. To shed light on this question, we study a teacher-student setting with a modular teacher where we have full control over the composition of ground truth modules. This allows us to relate the problem of compositional generalization to that of identification of the underlying modules. In particular we study modularity in hypernetworks representing a general class of multiplicative interactions. We show theoretically that identification up to linear transformation purely from demonstrations is possible without having to learn an exponential number of module combinations. We further demonstrate empirically that under the theoretically identified conditions, meta-learning from finite data can discover modular policies that generalize compositionally in a number of complex environments., Comment: Published as a conference paper at ICLR 2024; Code available at https://github.com/smonsays/modular-hyperteacher

Battery management systems (BMS) are becoming increasingly important in the modern age, where clean energy awareness is getting more prominent. They are responsible for controlling large battery packs in modern electric vehicles. However, conventional solutions rely only on a wired design, which adds manufacturing cost and complexity. Recent research has considered wireless solutions for the BMS. However, it is still challenging to develop a solution that considers both the active in-vehicle and the external second-life applications. The battery passport initiative aims to keep track of the batteries, both during active and inactive use cases. There is a need to provide a secure design while considering energy and cost-efficient solutions. We aim to fill this gap by proposing a wireless solution based on near-field communication (NFC) that extends previous work and provides a unified architecture for both use cases. To provide protection against common wireless threats, an advanced security analysis is performed, as well as a system design analysis for the wake-up process that reduces the daily power consumption of the stored battery packs from milli- to microwatts., Comment: Accepted copy for Publication at the 8th IEEE International Conference on Smart and Sustainable Technologies, 2023

In modern systems that rely on the use of Battery Management Systems (BMS), longevity and the re-use of battery packs have always been important topics of discussion. These battery packs would be stored inside warehouses where they would need to be properly monitored and configured before their re-integration into the new systems. Traditional use of wired connections can be very cumbersome, and sometimes even impossible, due to the outer layers and packaging. To circumvent these issues, we propose an extension to the conventional BMS design that incorporates the use of Near Field Communication (NFC) for the purpose of wireless battery pack status readout. Additionally, to ensure that these packs are only managed by authenticated devices and that the data that is communicated with is protected against outside eavesdropping and tampering, we present a solution in the form of a lightweight security layer on top of the NFC protocol. To show the feasibility of our design, an accompanying prototype has been implemented and evaluated., Comment: Accepted copy for Publication at the 16th IEEE International Conference on RFID, 2022

Be it in the IoT or automotive domain, implicit certificates are gaining ever more prominence in constrained embedded devices. They present a resource-efficient security solution against common threat concerns. The computational requirements are not the main issue anymore. The focus is now placed on determining a good balance between the provided security level and the derived threat model. A security aspect that often gets overlooked is the establishment of secure communication sessions, as most design solutions are based only on the use of static key derivation, and therefore, lack the perfect forward secrecy. This leaves the transmitted data open for potential future exposures by having keys tied to the certificates rather than the communication sessions. We aim to patch this gap, by presenting a design that utilizes the Station to Station (STS) protocol with implicit certificates. In addition, we propose potential protocol optimization implementation steps and run a comprehensive study on the performance and security level between the proposed design and the state-of-the-art key derivation protocols. In our comparative study, we show that with a slight computational increase of 20\% compared to a static ECDSA key derivation, we are able to mitigate many session-related security vulnerabilities that would otherwise remain open., Comment: Accepted copy for Publication at the Design, Automation and Test in Europe (DATE) Conference 2023

With the advent of clean energy awareness and systems that rely on extensive battery usage, the community has seen an increased interest in the development of more complex and secure Battery Management Systems (BMS). In particular, the inclusion of BMS in modern complex systems like electric vehicles and power grids has presented a new set of security-related challenges. A concern is shown when BMS are intended to extend their communication with external system networks, as their interaction can leave many backdoors open that potential attackers could exploit. Hence, it is highly desirable to find a general design that can be used for BMS and its system inclusion. In this work, a security architecture solution is proposed intended for the communication between BMS and other system devices. The aim of the proposed architecture is to be easily applicable in different industrial settings and systems, while at the same time keeping the design lightweight in nature., Comment: Accepted copy for Publication at the 23rd International Conference on Industrial Technology (ICIT), IEEE, 2022

With the advent of modern embedded systems, logging as a process is becoming more and more prevalent for diagnostic and analytic services. Traditionally, storage and managing of the logged data are generally kept as a part of one entity together with the main logic components. In systems that implement network connections, this activity is usually handled over a remote device. However, enabling remote connection is still considered a limiting factor for many embedded devices due to the demanding production cost. A significant challenge is presented to vendors who need to decide how the data will be extracted and handled for an embedded platform during the design concept phase. It is generally desirable that logging memory modules are able to be addressed as separate units. These devices need to be appropriately secured and verifiable on a different system since data compromise can lead to enormous privacy and even financial losses. In this paper, we present two patterns. First, a pattern that allows flexible logging operation design in terms of module and interface responsibility separation. Second, a pattern for the design of secure logging processes during the utilization of constrained embedded devices. The introduced patterns fulfil the following conditions: (i) flexibility, design is independent of the chip vendors making the logging memory modules easily replaceable, (ii) self-sufficiency, every logging controller is maintained as a separate entity in a decentralized topology, (iii) security, through providing authenticity, confidentiality, and integrity by means of using a dedicated security module., Comment: Accepted copy for Publication at the 26th European Conference on Pattern Languages of Programs (EuroPLoP), ACM, 2021

Recent architectural developments have enabled recurrent neural networks (RNNs) to reach and even surpass the performance of Transformers on certain sequence modeling tasks. These modern RNNs feature a prominent design pattern: linear recurrent layers interconnected by feedforward paths with multiplicative gating. Here, we show how RNNs equipped with these two design elements can exactly implement (linear) self-attention, the main building block of Transformers. By reverse-engineering a set of trained RNNs, we find that gradient descent in practice discovers our construction. In particular, we examine RNNs trained to solve simple in-context learning tasks on which Transformers are known to excel and find that gradient descent instills in our RNNs the same attention-based in-context learning algorithm used by Transformers. Our findings highlight the importance of multiplicative interactions in neural networks and suggest that certain RNNs might be unexpectedly implementing attention under the hood.

Wireless Battery Management Systems (BMS) are increasingly being considered for modern applications. The ever-increasing complexity and production costs of BMS modules and wired connections resulted in a necessity for new ideas and approaches. Despite this growing trend, there is a lack of generic solutions focused on battery cells' sensor readout, where wireless communication allows for a more flexible and cost-efficient sensor installation in battery packs. Many wireless technologies, such as those that use the 2.4 GHz frequency band, suffer from interference and other limitations. In this article, we present an alternative approach to communication in BMS that relies on the use of Near Field Communication (NFC) technology for battery sensor readouts. As an answer to the rising concern over the counterfeited battery packs, we consider an authentication schema for battery pack validation. We further consider security measures for the processed and stored BMS status data. To show that a general BMS application can make use of our design, we implement a BMS demonstrator using the targeted components. We further test the demonstrator on the technical and functional level, by also performing evaluation on its performance, energy usage, and a security threat model., Comment: arXiv admin note: text overlap with arXiv:2308.09366

In the last several years, wireless Battery Management Systems (BMS) have slowly become a topic of interest from both academia and industry. It came from a necessity derived from the increased production and use in different systems, including the electrical vehicles. Wireless communication allows for a more flexible and cost-efficient sensor installation in battery packs. However, many wireless technologies, such as those that use the 2.4 GHz frequency band, suffer from the interference limitations that need to be addressed. In this paper we present an alternative approach to communication in BMS that relies on the use of Near Field Communication (NFC) technology for battery sensor readouts. Due to a vital concern over the counterfeited battery pack products, security measures are also considered. To this end, we propose the use of an effective, and easy to integrate, authentication schema that is supported by the dedicated NFC devices. To test the usability of our design, a demonstrator using the targeted devices was implemented and evaluated., Comment: Accepted copy for Publication at the IEEE International Conference on RFID-Technologies and Applications, 2021

Biomimetic, dexterous robotic hands have the potential to replicate much of the tasks that a human can do, and to achieve status as a general manipulation platform. Recent advances in reinforcement learning (RL) frameworks have achieved remarkable performance in quadrupedal locomotion and dexterous manipulation tasks. Combined with GPU-based highly parallelized simulations capable of simulating thousands of robots in parallel, RL-based controllers have become more scalable and approachable. However, in order to bring RL-trained policies to the real world, we require training frameworks that output policies that can work with physical actuators and sensors as well as a hardware platform that can be manufactured with accessible materials yet is robust enough to run interactive policies. This work introduces the biomimetic tendon-driven Faive Hand and its system architecture, which uses tendon-driven rolling contact joints to achieve a 3D printable, robust high-DoF hand design. We model each element of the hand and integrate it into a GPU simulation environment to train a policy with RL, and achieve zero-shot transfer of a dexterous in-hand sphere rotation skill to the physical robot hand., Comment: for project website, see https://srl-ethz.github.io/get-ball-rolling/ . for video, see https://youtu.be/YahsMhqNU8o . for code, see https://github.com/srl-ethz/faive_gym_oss . Published to the 2023 IEEE-RAS International Conference on Humanoid Robots

Internet measurements are a crucial foundation of IPv6-related research. Due to the infeasibility of full address space scans for IPv6 however, those measurements rely on collections of reliably responsive, unbiased addresses, as provided e.g., by the IPv6 Hitlist service. Although used for various use cases, the hitlist provides an unfiltered list of responsive addresses, the hosts behind which can come from a range of different networks and devices, such as web servers, customer-premises equipment (CPE) devices, and Internet infrastructure. In this paper, we demonstrate the importance of tailoring hitlists in accordance with the research goal in question. By using PeeringDB we classify hitlist addresses into six different network categories, uncovering that 42% of hitlist addresses are in ISP networks. Moreover, we show the different behavior of those addresses depending on their respective category, e.g., ISP addresses exhibiting a relatively low lifetime. Furthermore, we analyze different Target Generation Algorithms (TGAs), which are used to increase the coverage of IPv6 measurements by generating new responsive targets for scans. We evaluate their performance under various conditions and find generated addresses to show vastly differing responsiveness levels for different TGAs., Comment: Proceedings of the Network Traffic Measurement and Analysis Conference (TMA)

The Kalman filter (KF) is a widely-used algorithm for tracking dynamic systems that are captured by state space (SS) models. The need to fully describe a SS model limits its applicability under complex settings, e.g., when tracking based on visual data, and the processing of high-dimensional signals often induces notable latency. These challenges can be treated by mapping the measurements into latent features obeying some postulated closed-form SS model, and applying the KF in the latent space. However, the validity of this approximated SS model may constitute a limiting factor. In this work, we study tracking from high-dimensional measurements under complex settings using a hybrid model-based/data-driven approach. By gradually tackling the challenges in handling the observations model and the task, we develop Latent-KalmanNet, which implements tracking from high-dimensional measurements by leveraging data to jointly learn the KF along with the latent space mapping. Latent-KalmanNet combines a learned encoder with data-driven tracking in the latent space using the recently proposed-KalmanNet, while identifying the ability of each of these trainable modules to assist its counterpart via providing a suitable prior (by KalmanNet) and by learning a latent representation that facilitates data-aided tracking (by the encoder). Our empirical results demonstrate that the proposed Latent-KalmanNet achieves improved accuracy and run-time performance over both model-based and data-driven techniques by learning a surrogate latent representation that most facilitates tracking, while operating with limited complexity and latency., Comment: Under review for publication in the IEEE

Carbon nanotubes are excellent nano-electromechanical systems, combining high resonance frequency, low mass, and large zero-point motion. At cryogenic temperatures they display high mechanical quality factors. Equally they are outstanding single electron devices with well-known quantum levels and have been proposed for the implementation of charge or spin qubits. The integration of these devices into microwave optomechanical circuits is however hindered by a mismatch of scales, between typical microwave wavelengths, nanotube segment lengths, and nanotube deflections. As experimentally demonstrated recently in [Blien et al., Nat. Comm. 11, 1363 (2020)], coupling enhancement via the quantum capacitance allows to circumvent this restriction. Here we extend the discussion of this experiment. We present the subsystems of the device and their interactions in detail. An alternative approach to the optomechanical coupling is presented, allowing to estimate the mechanical zero point motion scale. Further, the mechanical damping is discussed, hinting at hitherto unknown interaction mechanisms., Comment: 17 pages, 13 figures, 3 tables; v3: minor corrections

Equilibrium Bose-Einstein condensation of exciton-polaritons, demonstrated with a long-lifetime microcavity [Phys. Rev. Lett. 118, 016602 (2017)], has proven that driven-dissipative systems can undergo thermodynamic phase transitions in the limit where the quasiparticle lifetime exceeds the thermalization time. Here, we identify the role of dimensionality and polariton interactions in determining the degree of thermalization in optically generated traps. To distinguish the effect of trapping from interactions and lifetimes, we measured the polariton distribution under four nonresonant Gaussian pumps in a square geometry and compared it with polariton distributions measured with each pump individually. We found that significant redistribution of polaritons arises by trapping and modification of the density of states. Surprisingly efficient polariton-polariton scattering below the condensation threshold is evidenced by the depletion of the inflection-point polaritons. Our work provides a deeper understanding of polariton distributions and their interactions under various geometries of optically generated potentials.

The long-running IPv6 Hitlist service is an important foundation for IPv6 measurement studies. It helps to overcome infeasible, complete address space scans by collecting valuable, unbiased IPv6 address candidates and regularly testing their responsiveness. However, the Internet itself is a quickly changing ecosystem that can affect longrunning services, potentially inducing biases and obscurities into ongoing data collection means. Frequent analyses but also updates are necessary to enable a valuable service to the community. In this paper, we show that the existing hitlist is highly impacted by the Great Firewall of China, and we offer a cleaned view on the development of responsive addresses. While the accumulated input shows an increasing bias towards some networks, the cleaned set of responsive addresses is well distributed and shows a steady increase. Although it is a best practice to remove aliased prefixes from IPv6 hitlists, we show that this also removes major content delivery networks. More than 98% of all IPv6 addresses announced by Fastly were labeled as aliased and Cloudflare prefixes hosting more than 10M domains were excluded. Depending on the hitlist usage, e.g., higher layer protocol scans, inclusion of addresses from these providers can be valuable. Lastly, we evaluate different new address candidate sources, including target generation algorithms to improve the coverage of the current IPv6 Hitlist. We show that a combination of different methodologies is able to identify 5.6M new, responsive addresses. This accounts for an increase by 174% and combined with the current IPv6 Hitlist, we identify 8.8M responsive addresses.

Neural networks trained with stochastic gradient descent (SGD) starting from different random initialisations typically find functionally very similar solutions, raising the question of whether there are meaningful differences between different SGD solutions. Entezari et al.\ recently conjectured that despite different initialisations, the solutions found by SGD lie in the same loss valley after taking into account the permutation invariance of neural networks. Concretely, they hypothesise that any two solutions found by SGD can be permuted such that the linear interpolation between their parameters forms a path without significant increases in loss. Here, we use a simple but powerful algorithm to find such permutations that allows us to obtain direct empirical evidence that the hypothesis is true in fully connected networks. Strikingly, we find that two networks already live in the same loss valley at the time of initialisation and averaging their random, but suitably permuted initialisation performs significantly above chance. In contrast, for convolutional architectures, our evidence suggests that the hypothesis does not hold. Especially in a large learning rate regime, SGD seems to discover diverse modes., Comment: NeurIPS 2022, 14th Annual Workshop on Optimization for Machine Learning (OPT2022)

Ultrathin solar cells reduce material usage and allow the use of lower-quality materials thanks to their one order of magnitude smaller thickness than their conventional counterparts. However, efficient photonic light-trapping is required to harvest the incident light efficiently for an otherwise insufficient absorber thickness. Quasi-random photonic crystals are predicted to have high efficient light-trapping while being more robust under angle and thickness variations than simple photonic crystals. Here we experimentally demonstrate a light-trapping solution based on quasi-random photonic crystals fabricated by polymer blend lithography. We control the average lattice parameter by modifying the spin-coating speed. We demonstrate an ultrathin GaAs cell of 260 nm with a rear quasi-random pattern with submicron features, and a Jsc =26.4 mA/cm2 and an efficiency of 22.35% under the global solar spectrum.

Annotating objects with 3D bounding boxes in LiDAR pointclouds is a costly human driven process in an autonomous driving perception system. In this paper, we present a method to semi-automatically annotate real-world pointclouds collected by deployment vehicles using simulated data. We train a 3D object detector model on labeled simulated data from CARLA jointly with real world pointclouds from our target vehicle. The supervised object detection loss is augmented with a CORAL loss term to reduce the distance between labeled simulated and unlabeled real pointcloud feature representations. The goal here is to learn representations that are invariant to simulated (labeled) and real-world (unlabeled) target domains. We also provide an updated survey on domain adaptation methods for pointclouds., Comment: Accepted at IMPROVE 2022

We connect the problem of semi-supervised clustering to constrained Markov aggregation, i.e., the task of partitioning the state space of a Markov chain. We achieve this connection by considering every data point in the dataset as an element of the Markov chain's state space, by defining the transition probabilities between states via similarities between corresponding data points, and by incorporating semi-supervision information as hard constraints in a Hartigan-style algorithm. The introduced Constrained Markov Clustering (CoMaC) is an extension of a recent information-theoretic framework for (unsupervised) Markov aggregation to the semi-supervised case. Instantiating CoMaC for certain parameter settings further generalizes two previous information-theoretic objectives for unsupervised clustering. Our results indicate that CoMaC is competitive with the state-of-the-art., Comment: 13 pages, 6 figures; this is an extended version of a short paper accepted at ACM SAC 2022 (minor changes to the text; error in source code corrected)

We introduce the first comprehensive 3D dataset for the task of unsupervised anomaly detection and localization. It is inspired by real-world visual inspection scenarios in which a model has to detect various types of defects on manufactured products, even if it is trained only on anomaly-free data. There are defects that manifest themselves as anomalies in the geometric structure of an object. These cause significant deviations in a 3D representation of the data. We employed a high-resolution industrial 3D sensor to acquire depth scans of 10 different object categories. For all object categories, we present a training and validation set, each of which solely consists of scans of anomaly-free samples. The corresponding test sets contain samples showing various defects such as scratches, dents, holes, contaminations, or deformations. Precise ground-truth annotations are provided for every anomalous test sample. An initial benchmark of 3D anomaly detection methods on our dataset indicates a considerable room for improvement., Comment: Accepted for presentation at VISAPP 2022

The classic Bogoliubov theory of weakly interacting Bose gases rests upon the assumption that nearly all the bosons condense into the lowest quantum state at sufficiently low temperatures. Here we develop a generalized version of Bogoliubov theory for the case of a driven-dissipative exciton-polariton condensate with a large incoherent uncondensed component, or excitonic reservoir. We argue that such a reservoir can consist of both excitonic high-momentum polaritons and optically dark superpositions of excitons across different optically active layers, such as multiple quantum wells in a microcavity. In particular, we predict interconversion between the dark and bright (light-coupled) excitonic states that can lead to a dynamical equilibrium between the condensate and reservoir populations. We show that the presence of the reservoir fundamentally modifies both the energy and the amplitudes of the Bogoliubov quasiparticle excitations due to the non-Galilean-invariant nature of polaritons. Our theoretical findings are supported by our experiment, where we directly detect the Bogoliubov excitation branches of an optically trapped polariton condensate in the high-density regime. By analyzing the measured occupations of the excitation branches, we extract the Bogoliubov amplitudes across a range of momenta and show that they agree with our generalized theory., Comment: 16 pages, 7 figures

The polariton-polariton interaction strength is an important parameter for all kinds of applications using the nonlinear properties of polaritons, such as optical switching and single-photon blockade devices. In this paper, we review and compare the results of a series of experiments on polariton-polariton interactions in GaAs/Al$_x$Ga$_{-1x}$As microcavity polariton structures, and present new theoretical analysis of these experiments. We show that not just the energy shift of the spectral lines, but also the results of measurements sensitive to the polariton scattering rate are important for the calibration of the interaction parameter at low excitation density. We find that when adjustments are made to correct for new understanding of the experiments, the value of the interaction parameter at low density is lower than previous reported, but still significantly higher than theoretically predicted., Comment: Review article with 13 figures

Permutation Mastermind is a version of the classical mastermind game in which the number of positions $n$ is equal to the number of colors $k$, and repetition of colors is not allowed, neither in the codeword nor in the queries. In this paper we solve the main open question from Glazik, J\"ager, Schiemann and Srivastav (2021), who asked whether their bound of $O(n^{1.525})$ for the static version can be improved to $O(n \log n)$, which would be best possible. By using a simple probabilistic argument we show that this is indeed the case., Comment: 6 pages

Bitcoin and Ethereum transactions present one of the largest real-world complex networks that are publicly available for study, including a detailed picture of their time evolution. As such, they have received a considerable amount of attention from the network science community, beside analysis from an economic or cryptography perspective. Among these studies, in an analysis on the early instance of the Bitcoin network, we have shown the clear presence of the preferential attachment, or "rich-get-richer" phenomenon. Now, we revisit this question, using a recent version of the Bitcoin network that has grown almost 100-fold since our original analysis. Furthermore, we additionally carry out a comparison with Ethereum, the second most important cryptocurrency. Our results show that preferential attachment continues to be a key factor in the evolution of both the Bitcoin and Ethereum transactoin networks. To facilitate further analysis, we publish a recent version of both transaction networks, and an efficient software implementation that is able to evaluate linking statistics necessary for learn about preferential attachment on networks with several hundred million edges.

Topological insulators are a class of electronic materials exhibiting robust edge states immune to perturbations and disorder. This concept has been successfully adapted in photonics, where topologically nontrivial waveguides and topological lasers were developed. However, the exploration of topological properties in a given photonic system is limited to a fabricated sample, without the flexibility to reconfigure the structure in-situ. Here, we demonstrate an all-optical realization of the orbital Su-Schrieffer-Heeger (SSH) model in a microcavity exciton-polariton system, whereby a cavity photon is hybridized with an exciton in a GaAs quantum well. We induce a zigzag potential for exciton polaritons all-optically, by shaping the nonresonant laser excitation, and measure directly the eigenspectrum and topological edge states of a polariton lattice in a nonlinear regime of bosonic condensation. Furthermore, taking advantage of the tunability of the optically induced lattice we modify the intersite tunneling to realize a topological phase transition to a trivial state. Our results open the way to study topological phase transitions on-demand in fully reconfigurable hybrid photonic systems that do not require sophisticated sample engineering., Comment: 7 pages, 4 figures

We design a randomized algorithm that finds a Hamilton cycle in $\mathcal{O}(n)$ time with high probability in a random graph $G_{n,p}$ with edge probability $p\ge C \log n / n$. This closes a gap left open in a seminal paper by Angluin and Valiant from 1979.

Collective (elementary) excitations of quantum bosonic condensates, including condensates of exciton polaritons in semiconductor microcavities, are a sensitive probe of interparticle interactions. In anisotropic microcavities with momentum-dependent TE-TM splitting of the optical modes, the excitations dispersions are predicted to be strongly anisotropic, which is a consequence of the synthetic magnetic gauge field of the cavity, as well as the interplay between different interaction strengths for polaritons in the singlet and triplet spin configurations. Here, by directly measuring the dispersion of the collective excitations in a high-density optically trapped exciton-polariton condensate, we observe excellent agreement with the theoretical predictions for spinor polariton excitations. We extract the inter- and intra-spin polariton interaction constants and map out the characteristic spin textures in an interacting spinor condensate of exciton polaritons.

We have observed the effect of pseudo magnetic field originating from the polaritonic analog of spin-orbit coupling (TE$-$TM splitting) on a polariton condensate in a ring-shaped microcavity. The effect gives rise to a stable four-leaf pattern around the ring as seen from the linear polarization measurements of the condensate photoluminescence. This pattern is found to originate from the interplay of the cavity potential, energy relaxation, and TE-TM splitting in the ring. Our observations are compared to the dissipative one-dimensional spinor Gross-Pitaevskii equation with the TE-TM splitting energy which shows good qualitative agreement., Comment: 13 pages, 7 figures, preprint version, minor edit v2

Let $\pi$ be an irreducible unitary representation of a finitely generated nonabelian free group $\Gamma$; suppose $\pi$ is weakly contained in the regular representation. In 2001 the first and third authors conjectured that such a representation must be either odd or monotonous or duplicitous. In 2004 they introduced the class of multiplicative representations: this is a large class of representations obtained by looking at the action of $\Gamma$ on its Cayley graph. In the second paper of this series we showed that some of the multiplicative representations were monotonous. Here we show that all the other multiplicative representations are either odd or duplicitous. The conjecture is therefore established for multiplicative representations.

We report the observation of low-energy, low-momenta collective oscillations of an excitonpolariton condensate in a round "box" trap. The oscillations are dominated by the dipole and breathing modes, and the ratio of the frequencies of the two modes is consistent with that of a weakly interacting two-dimensional trapped Bose gas. The speed of sound extracted from the dipole oscillation frequency is smaller than the Bogoliubov sound, which can be partly explained by the influence of the incoherent reservoir. These results pave the way for understanding the effects of reservoir, dissipation, energy relaxation, and finite temperature on the superfluid properties of exciton-polariton condensates and other two-dimensional open-dissipative quantum fluids., Comment: 13 pages, 15 figures

Consider the following simple coloring algorithm for a graph on $n$ vertices. Each vertex chooses a color from $\{1, \dotsc, \Delta(G) + 1\}$ uniformly at random. While there exists a conflicted vertex choose one such vertex uniformly at random and recolor it with a randomly chosen color. This algorithm was introduced by Bhartia et al. [MOBIHOC'16] for channel selection in WIFI-networks. We show that this algorithm always converges to a proper coloring in expected $O(n \log \Delta)$ steps, which is optimal and proves a conjecture of Chakrabarty and Supinski [SOSA'20].

Optical trapping and manipulation of microcavity exciton polaritons rely on effective potentials induced by the interaction of polaritons with a reservoir of high energy excitonic particles injected by an off-resonant optical pump. Here, we experimentally investigate possible mechanisms responsible for reshaping of these effective potentials in the low-density exciton-polariton regime. We infer the spatial distribution of the reservoir from the spatially resolved energy of exciton-polariton emission measured at zero momentum (zero kinetic energy). Power-dependent shape analysis of the potential barrier induced by a focused continuous wave laser pump shows a monotonic decrease of the barrier width with increasing excitation power, which is attributed to the local heating of the sample at the pump spot. In addition, we observe the significant influence of the reservoir on the zero-momentum emission tens of micrometers away from the laser pump spot, which is in line with the previously reported enhanced transport of high-momentum excitonic polaritons from the bottleneck region. Our work presents evidence for a complex spatial reshaping of the reservoir with the pump power, contrary to the common assumption of a static reservoir distribution fixed by the intensity profile of the pump., Comment: There was a substantial error in the analysis of the experimental data presented in the paper. The figures as well as the conclusions of the manuscript are incorrect and should not be cited

The Higgs amplitude mode is a collective excitation studied and observed in a broad class of matter, including superconductors, charge density waves, antiferromagnets, 3He p-wave superfluid, and ultracold atomic condensates. In all the observations reported thus far, the amplitude mode was excited by perturbing the condensate out of equilibrium. Studying an exciton-polariton condensate, here we report the first observation of this mode purely driven by intrinsic quantum fluctuations without such perturbations. By using an ultrahigh quality microcavity and a Raman spectrometer to maximally reject photoluminescence from the condensate, we observe weak but distinct photoluminescence at energies below the condensate emission. We identify this as the so-called ghost branches of the amplitude mode arising from quantum depletion of the condensate into this mode. These energies, as well as the overall structure of the photoluminescence spectra, are in good agreement with our theoretical analysis.

Research collaborations are continuously emerging catalyzed by online platforms, where people can share their codes, calculations, data and results. These virtual research platforms are innovative, community oriented, flexible and secure as required by modern scientific approaches. A wide range of open source and commercial solutions are available in this field emphasizing the relevant aspects of such a platform differently. In this paper we present our open source and modular platform, KOOPLEX, which combines the key concepts of dynamic collaboration, customizable research environment, data sharing, access to datahubs, reproducible research and reporting. It is easily deployable and scalable to serve more users or access large computational resources., Comment: 22 pages, 4 figures, 6 tables

We introduce a powerful student-teacher framework for the challenging problem of unsupervised anomaly detection and pixel-precise anomaly segmentation in high-resolution images. Student networks are trained to regress the output of a descriptive teacher network that was pretrained on a large dataset of patches from natural images. This circumvents the need for prior data annotation. Anomalies are detected when the outputs of the student networks differ from that of the teacher network. This happens when they fail to generalize outside the manifold of anomaly-free training data. The intrinsic uncertainty in the student networks is used as an additional scoring function that indicates anomalies. We compare our method to a large number of existing deep learning based methods for unsupervised anomaly detection. Our experiments demonstrate improvements over state-of-the-art methods on a number of real-world datasets, including the recently introduced MVTec Anomaly Detection dataset that was specifically designed to benchmark anomaly segmentation algorithms., Comment: Accepted to CVPR 2020

Evolutionary Strategies (ES) are known to be an effective black-box optimization technique for deep neural networks when the true gradients cannot be computed, such as in Reinforcement Learning. We continue a recent line of research that uses surrogate gradients to improve the gradient estimation of ES. We propose a novel method to optimally incorporate surrogate gradient information. Our approach, unlike previous work, needs no information about the quality of the surrogate gradients and is always guaranteed to find a descent direction that is better than the surrogate gradient. This allows to iteratively use the previous gradient estimate as surrogate gradient for the current search point. We theoretically prove that this yields fast convergence to the true gradient for linear functions and show under simplifying assumptions that it significantly improves gradient estimates for general functions. Finally, we evaluate our approach empirically on MNIST and reinforcement learning tasks and show that it considerably improves the gradient estimation of ES at no extra computational cost.

Learning long-term dependencies is a key long-standing challenge of recurrent neural networks (RNNs). Hierarchical recurrent neural networks (HRNNs) have been considered a promising approach as long-term dependencies are resolved through shortcuts up and down the hierarchy. Yet, the memory requirements of Truncated Backpropagation Through Time (TBPTT) still prevent training them on very long sequences. In this paper, we empirically show that in (deep) HRNNs, propagating gradients back from higher to lower levels can be replaced by locally computable losses, without harming the learning capability of the network, over a wide range of tasks. This decoupling by local losses reduces the memory requirements of training by a factor exponential in the depth of the hierarchy in comparison to standard TBPTT.

The property of superfluidity, first discovered in liquid 4He, is closely related to Bose-Einstein condensation (BEC) of interacting bosons. However, even at zero temperature, when one would expect the whole bosonic quantum liquid to become condensed, a fraction of it is excited into higher momentum states via interparticle interactions and quantum fluctuations -- the phenomenon of quantum depletion. Quantum depletion of weakly interacting atomic BECs in thermal equilibrium is well understood theoretically but is difficult to measure. This is even more challenging in driven-dissipative systems such as exciton-polariton condensates(photons coupled to electron-hole pairs in a semiconductor), since their nonequilibrium nature is predicted to suppress quantum depletion. Here, we observe quantum depletion of an optically trapped high-density exciton-polariton condensate by directly detecting the spectral branch of elementary excitations populated by this process. Analysis of the population of this branch in momentum space shows that quantum depletion of an exciton-polariton condensate can closely follow or strongly deviate from the equilibrium Bogoliubov theory, depending on the fraction of matter (exciton) in an exciton-polariton. Our results reveal the effects of exciton-polariton interactions beyond the mean-field description and call for a deeper understanding of the relationship between equilibrium and nonequilibrium BECs., Comment: 18 pages, 5 figures, with supplementary information

We present a powerful theorem for proving the irreducibility of tempered unitary representations of the free group.

Cavity optomechanics allows the characterization of a vibration mode, its cooling and quantum manipulation using electromagnetic fields. Regarding nanomechanical as well as electronic properties, single wall carbon nanotubes are a prototypical experimental system. At cryogenic temperatures, as high quality factor vibrational resonators, they display strong interaction between motion and single-electron tunneling. Here, we demonstrate large optomechanical coupling of a suspended carbon nanotube quantum dot and a microwave cavity, amplified by several orders of magnitude via the nonlinearity of Coulomb blockade. From an optomechanically induced transparency (OMIT) experiment, we obtain a single photon coupling of up to $g_0=2\pi\cdot 95\,\textrm{Hz}$. This indicates that normal mode splitting and full optomechanical control of the carbon nanotube vibration in the quantum limit is reachable in the near future. Mechanical manipulation and characterization via the microwave field can be complemented by the manifold physics of quantum-confined single electron devices., Comment: 6 pages, 4 figures; published version (open access)

One of the central goals of Recurrent Neural Networks (RNNs) is to learn long-term dependencies in sequential data. Nevertheless, the most popular training method, Truncated Backpropagation through Time (TBPTT), categorically forbids learning dependencies beyond the truncation horizon. In contrast, the online training algorithm Real Time Recurrent Learning (RTRL) provides untruncated gradients, with the disadvantage of impractically large computational costs. Recently published approaches reduce these costs by providing noisy approximations of RTRL. We present a new approximation algorithm of RTRL, Optimal Kronecker-Sum Approximation (OK). We prove that OK is optimal for a class of approximations of RTRL, which includes all approaches published so far. Additionally, we show that OK has empirically negligible noise: Unlike previous algorithms it matches TBPTT in a real world task (character-level Penn TreeBank) and can exploit online parameter updates to outperform TBPTT in a synthetic string memorization task. Code availiable on github., Comment: ICML 2019 camera ready version; new version includes additional plots in the appendix

We present a study of the macroscopic dynamics of a polariton condensate formed by non-resonant optical excitation in a quasi-one-dimensional ring-shaped microcavity. The presence of a gradient in the cavity photon energy creates a macroscopic trap for the polaritons in which a single mode condensate is formed. With time- and energy-resolved imaging we show the role of interactions in the motion of the condensate as it undergoes equilibration in the ring. These experiments also give a direct measurement of the polariton-polariton interaction strength above the condensation threshold. Our observations are compared to the open-dissipative one-dimensional Gross-Pitaevskii equation which shows excellent qualitative agreement., Comment: Revised manuscript for Phys Rev. B submission