Search

Radio-frequency interference (RFI) is becoming an increasingly significant problem for most radio telescopes. Working with Green Bank Telescope data from PSR J1730+0747 in the form of complex-valued channelized voltages and their respective high-resolution power spectral densities, we evaluate a variety of statistical measures to characterize RFI. As a baseline for performance comparison, we use median absolute deviation (MAD) in complex channelized voltage data and spectral kurtosis (SK) in power spectral density data to characterize and filter out RFI. From a new perspective, we implement the Shapiro-Wilks (SW) test for normality and two information theoretical measures, spectral entropy (SE) and spectral relative entropy (SRE), and apply them to mitigate RFI. The baseline RFI mitigation algorithms are compared against our novel RFI detection algorithms to determine how effective and robust the performance is. Except for MAD, we find significant improvements in signal-to-noise ratio through the application of SE, symmetrical SRE, asymmetrical SRE, SK, and SW. These algorithms also do a good job of characterizing broadband RFI. Time- and frequency-variable RFI signals are best detected by SK and SW tests., Comment: 13 pages, 26 figures, accepted for publication in RAS Techniques and Instruments

The ability of a brain or a neural network to efficiently learn depends crucially on both the task structure and the learning rule. Previous works have analyzed the dynamical equations describing learning in the relatively simplified context of the perceptron under assumptions of a student-teacher framework or a linearized output. While these assumptions have facilitated theoretical understanding, they have precluded a detailed understanding of the roles of the nonlinearity and input-data distribution in determining the learning dynamics, limiting the applicability of the theories to real biological or artificial neural networks. Here, we use a stochastic-process approach to derive flow equations describing learning, applying this framework to the case of a nonlinear perceptron performing binary classification. We characterize the effects of the learning rule (supervised or reinforcement learning, SL/RL) and input-data distribution on the perceptron's learning curve and the forgetting curve as subsequent tasks are learned. In particular, we find that the input-data noise differently affects the learning speed under SL vs. RL, as well as determines how quickly learning of a task is overwritten by subsequent learning. Additionally, we verify our approach with real data using the MNIST dataset. This approach points a way toward analyzing learning dynamics for more-complex circuit architectures.

We consider the stationary measure of the open asymmetric simple exclusion process (ASEP) with light particles. We prove several results on the asymptotic locations of the light particles under the stationary measure. Moreover, in the fan region of the high and low density phases, we study the mixing times for open ASEP with light particles. Our approach involves the matrix product ansatz, Askey-Wilson signed measures, moderate deviation results for second class particles, as well as various coupling arguments. Along the way we obtain the particle densities at single sites for the standard open ASEP without light particles, which are of independent interest., Comment: 30 pages, 5 figures

The rapid deployment of robotics technologies requires dedicated optimization algorithms to manage large fleets of autonomous agents. This paper supports robotic parts-to-picker operations in warehousing by optimizing order-workstation assignments, item-pod assignments and the schedule of order fulfillment at workstations. The model maximizes throughput, while managing human workload at the workstations and congestion in the facility. We solve it via large-scale neighborhood search, with a novel learn-then-optimize approach to subproblem generation. The algorithm relies on an offline machine learning procedure to predict objective improvements based on subproblem features, and an online optimization model to generate a new subproblem at each iteration. In collaboration with Amazon Robotics, we show that our model and algorithm generate much stronger solutions for practical problems than state-of-the-art approaches. In particular, our solution enhances the utilization of robotic fleets by coordinating robotic tasks for human operators to pick multiple items at once, and by coordinating robotic routes to avoid congestion in the facility.

With increasingly larger and more complex telecommunication networks, there is a need for improved monitoring and reliability. Requirements increase further when working with mission-critical systems requiring stable operations to meet precise design and client requirements while maintaining high availability. This paper proposes a novel methodology for developing a machine learning model that can assist in maintaining availability (through anomaly detection) for client-server communications in mission-critical systems. To that end, we validate our methodology for training models based on data classified according to client performance. The proposed methodology evaluates the use of machine learning to perform anomaly detection of a single virtualized server loaded with simulated network traffic (using SIPp) with media calls. The collected data for the models are classified based on the round trip time performance experienced on the client side to determine if the trained models can detect anomalous client side performance only using key performance indicators available on the server. We compared the performance of seven different machine learning models by testing different trained and untrained test stressor scenarios. In the comparison, five models achieved an F1-score above 0.99 for the trained test scenarios. Random Forest was the only model able to attain an F1-score above 0.9 for all untrained test scenarios with the lowest being 0.980. The results suggest that it is possible to generate accurate anomaly detection to evaluate degraded client-side performance.

A central ingredient in fault-tolerant quantum algorithms is the initialization of a logical state for a given quantum error-correcting code from a set of noisy qubits. A scheme that has demonstrated promising results for small code instances that are realizable on currently available hardware composes a non-fault-tolerant state preparation step with a verification step that checks for spreading errors. Known circuit constructions of this scheme are mostly obtained manually, and no algorithmic techniques for constructing depth- or gate-optimal circuits exist. As a consequence, the current state of the art exploits this scheme only for specific code instances and mostly for the special case of distance 3 codes. In this work, we propose an automated approach for synthesizing fault-tolerant state preparation circuits for arbitrary CSS codes. We utilize methods based on satisfiability solving (SAT) techniques to construct fault-tolerant state preparation circuits consisting of depth- and gate-optimal preparation and verification circuits. We also provide heuristics that can synthesize fault-tolerant state preparation circuits for code instances where no optimal solution can be obtained in an adequate timeframe. Moreover, we give a general construction for non-deterministic state preparation circuits beyond distance 3. Numerical evaluations using $d=3$ and $d=5$ codes confirm that the generated circuits exhibit the desired scaling of the logical error rates. The resulting methods are publicly available as part of the Munich Quantum Toolkit (MQT) at https://github.com/cda-tum/mqt-qecc. Such methods are an important step in providing fault-tolerant circuit constructions that can aid in near-term demonstration of fault-tolerant quantum computing., Comment: 23 pages, 14 figures Version 2: included some references, fixed error in printed circuits

Dual-encoder-based audio retrieval systems are commonly optimized with contrastive learning on a set of matching and mismatching audio-caption pairs. This leads to a shared embedding space in which corresponding items from the two modalities end up close together. Since audio-caption datasets typically only contain matching pairs of recordings and descriptions, it has become common practice to create mismatching pairs by pairing the audio with a caption randomly drawn from the dataset. This is not ideal because the randomly sampled caption could, just by chance, partly or entirely describe the audio recording. However, correspondence information for all possible pairs is costly to annotate and thus typically unavailable; we, therefore, suggest substituting it with estimated correspondences. To this end, we propose a two-staged training procedure in which multiple retrieval models are first trained as usual, i.e., without estimated correspondences. In the second stage, the audio-caption correspondences predicted by these models then serve as prediction targets. We evaluate our method on the ClothoV2 and the AudioCaps benchmark and show that it improves retrieval performance, even in a restricting self-distillation setting where a single model generates and then learns from the estimated correspondences. We further show that our method outperforms the current state of the art by 1.6 pp. mAP@10 on the ClothoV2 benchmark., Comment: In Proceedings of the 9th Workshop on Detection and Classification of Acoustic Scenes and Events, DCASE, Tokyo, Japan, 2024. Implementation available on GitHub: https://github.com/OptimusPrimus/salsa

Query-by-Vocal Imitation (QBV) is about searching audio files within databases using vocal imitations created by the user's voice. Since most humans can effectively communicate sound concepts through voice, QBV offers the more intuitive and convenient approach compared to text-based search. To fully leverage QBV, developing robust audio feature representations for both the vocal imitation and the original sound is crucial. In this paper, we present a new system for QBV that utilizes the feature extraction capabilities of Convolutional Neural Networks pre-trained with large-scale general-purpose audio datasets. We integrate these pre-trained models into a dual encoder architecture and fine-tune them end-to-end using contrastive learning. A distinctive aspect of our proposed method is the fine-tuning strategy of pre-trained models using an adapted NT-Xent loss for contrastive learning, creating a shared embedding space for reference recordings and vocal imitations. The proposed system significantly enhances audio retrieval performance, establishing a new state of the art on both coarse- and fine-grained QBV tasks., Comment: Accepted to the DCASE Workshop 2024. Source code available: https://github.com/Jonathan-Greif/QBV

Surfaces of cubic perovskite oxides attract significant attention for their physical tunability and high potential for technical applications. Bulk-terminated surfaces are desirable for theoretical modelling and experimental reproducibility, yet there is a lack of methods for preparing such well-defined surfaces. We discuss a method for strain-assisted cleaving of perovskite single crystals, using a setup easily transferable between different experimental systems. The details of the cleaving device and the procedure were optimized in a systematic study on the model SrTiO$_3$. The large-area morphology and typical distribution of surface terminations on cleaved SrTiO$_3$(001) is presented, with specific guidelines on how to distinguish well-cleaved surfaces from conchoidally fractured ones. The cleaving is applicable to other cubic perovskites, as demonstrated on KTaO$_3$(001) and BaTiO$_3$(001). This approach opens up a pathway for obtaining high-quality surfaces of this promising class of materials.

In ground-based astronomy, the ability to couple light into single-mode fibers (SMFs) is limited by atmospheric turbulence, which prohibits the use of many astrophotonic instruments. We propose a silicon-on-insulator photonic chip capable of coherently coupling the out-of-phase beamlets from the subapertures of a telescope pupil into an SMF. The photonic integrated circuit (PIC) consists of an array of grating couplers that are used to inject light from free space into single-mode waveguides on the chip. Metallic heaters modulate the refractive index of a coiled section of the waveguides, facilitating the co-phasing of the propagating modes. The phased beamlets can then be coherently combined to efficiently deliver the light to an output SMF. In an adaptive optics (AO) system, the phase corrector acts as a deformable mirror (DM) commanded by a controller that takes phase measurements from a wavefront sensor (WFS). We present experimental results for the PIC tested on an AO testbed and compare the performance to simulations., Comment: 13 pages, 11 figures, 2 tables, submitted to SPIE Adaptive Optics Systems IX

We study iterative blind symbol detection for block-fading linear inter-symbol interference channels. Based on the factor graph framework, we design a joint channel estimation and detection scheme that combines the expectation maximization (EM) algorithm and the ubiquitous belief propagation (BP) algorithm. Interweaving the iterations of both schemes significantly reduces the EM algorithm's computational burden while retaining its excellent performance. To this end, we apply simple yet effective model-based learning methods to find a suitable parameter update schedule by introducing momentum in both the EM parameter updates as well as in the BP message passing. Numerical simulations verify that the proposed method can learn efficient schedules that generalize well and even outperform coherent BP detection in high signal-to-noise scenarios., Comment: Accepted for presentation at Asilomar Conference on Signals, Systems, and Computers 2024

Emerging network applications ranging from video streaming to virtual/augmented reality need to provide stringent quality-of-service (QoS) guarantees in complex and dynamic environments with shared resources. A promising approach to meeting these requirements is to automate complex network operations and create self-adjusting networks. These networks should automatically gather contextual information, analyze how to efficiently ensure QoS requirements, and adapt accordingly. This paper presents ReactNET, a self-adjusting networked system designed to achieve this vision by leveraging emerging network programmability and machine learning techniques. Programmability empowers ReactNET by providing fine-grained telemetry information, while machine learning-based classification techniques enable the system to learn and adjust the network to changing conditions. Our preliminary implementation of ReactNET in P4 and Python demonstrates its effectiveness in video streaming applications.

Using a focused He$^+$ beam for nanopatterning and writing of Josephson barriers we fabricated specially shaped Josephson junctions of in-line geometry in YBa$_2$Cu$_3$O$_7$ thin film microbridges with an asymmetry ratio of critical currents of opposite polarities (non-reciprocity ratio) $\approx 7$ at optimum magnetic field. Those Josephson diodes were subsequently used as ratchets to rectify an applied ac current into a dc voltage. We also demonstrate the operation of such a ratchet in the loaded regime, where it produces a nonzero dc output power and yields a thermodynamic efficiency of up to $75\,\mathrm{\%}$. The ratchet shows record figures of merit: an output dc voltage of up to $212\,\mathrm{\mu V}$ and an output power of up to $0.2\,\mathrm{nW}$. The device has an essential area $\approx 1\,\mathrm{\mu m^2}$. For rectification of quasistatic Gaussian noise, the figures of merit are more modest, however the efficiency can be as high as for the deterministic ac drives within some regimes. Since the device is based on YBa$_2$Cu$_3$O$_7$, it can operate at temperatures up to $\sim40\,\mathrm{K}$, where more noise is available for rectification.

Tomographic locality is a principle commonly used in the program of finding axioms that pick out quantum theory within the landscape of possible theories. The principle asserts the sufficiency of local measurements for achieving a tomographic characterization of any bipartite state. In this work, we explore the meaning of the principle of tomographic locality by developing a simple scheme for generating a wide variety of theories that violate the principle. In this scheme, one starts with a tomographically local theory -- which can be classical, quantum or post-quantum -- and a physical symmetry, and one restricts the processes in the theory to all and only those that are covariant with respect to that symmetry. We refer to the resulting theories as twirled worlds. We show that failures of tomographic locality are ubiquitous in twirled worlds. From the possibility of such failures in classical twirled worlds, we argue that the failure of tomographic locality (i.e., tomographic nonlocality) does not imply ontological holism. Our results also demonstrate the need for researchers seeking to axiomatize quantum theory to take a stand on the question of whether there are superselection rules that have a fundamental status., Comment: 5+12 pages, 2 figures. Comments welcome!

Based on our companion paper [Krieger-Schmid, 2024], we show that the 4D energy critical Zakharov system admits finite time type II blow up solutions. The main new difficulty this work deals with is the appearance of a term in the linearization around the approximate solution, which is non-local with respect to both space and time. In particular this cannot be handled by straightforward adaptation of the methods developed in [Krieger-Schlag-Tataru, 2008/09]. The key new ingredients we use are a type of approximate modulation theory, taking advantage of frequency localisations, and the exploitation of an inhomogeneous wave equation with both a non-local, as well as a local potential term. These terms arise for the main non-perturbative component of the ion density $n$ and can be solved via inversion of a certain Fredholm type operator, as well as by using distorted Fourier methods. Our result relies on a number of numerical non-degeneracy assumptions., Comment: 169 pages

We construct approximate solutions $ (\psi_*, n_*)$ of the critical 4D Zakharov system which collapse in finite time to a singular renormalization of the solitary bulk solutions $ (\lambda e^{i \theta}W, \lambda^2 W^2)$ . To be precise for $ N \in \mathbb{Z}_+,\;N \gg1 $ we obtain a magnetic envelope/ion density pair of the form $$ \psi_*(t, x)= e^{i\alpha(t)}\lambda(t) W(\lambda(t)x) + \eta(t, x), \;n_*(t,x) = \lambda^2(t) W^2(\lambda(t) x) + \chi(t,x), $$ where $ W(x) = (1 + \frac{|x|^2}{8})^{-1}$, $\alpha(t) = \alpha_0 \log(t)$, $\lambda(t)= t^{-\frac{1}{2}-\nu}$ with large $\nu > 1 $ and further $$ i \partial_t \psi_* + \Delta \psi_* + n_* \psi_* = \mathcal{O}(t^N),\; \Box n_* - \Delta (|\psi_*|^2) = \mathcal{O}(t^N),\;\;\eta(t) \to \eta_0, \chi(t) \to \chi_0, $$ as $ t \to 0^+$ in a suitable sense. The method of construction is inspired by matched asymptotic regions and approximation procedures in the context of blow up solutions introduced by the first author jointly with W. Schlag and D. Tataru, as well as the subsequently developed methods in the Schr\"odinger context by G. Perelman et al., Comment: 149 pages

Machine learning (ML) has seen significant growth in both popularity and importance. The high prediction accuracy of ML models is often achieved through complex black-box architectures that are difficult to interpret. This interpretability problem has been hindering the use of ML in fields like medicine, ecology and insurance, where an understanding of the inner workings of the model is paramount to ensure user acceptance and fairness. The need for interpretable ML models has boosted research in the field of interpretable machine learning (IML). Here we propose a novel approach for the functional decomposition of black-box predictions, which is considered a core concept of IML. The idea of our method is to replace the prediction function by a surrogate model consisting of simpler subfunctions. Similar to additive regression models, these functions provide insights into the direction and strength of the main feature contributions and their interactions. Our method is based on a novel concept termed stacked orthogonality, which ensures that the main effects capture as much functional behavior as possible and do not contain information explained by higher-order interactions. Unlike earlier functional IML approaches, it is neither affected by extrapolation nor by hidden feature interactions. To compute the subfunctions, we propose an algorithm based on neural additive modeling and an efficient post-hoc orthogonalization procedure.

LLMs are changing the way humans create and interact with content, potentially affecting citizens' political opinions and voting decisions. As LLMs increasingly shape our digital information ecosystems, auditing to evaluate biases, sycophancy, or steerability has emerged as an active field of research. In this paper, we evaluate and compare the alignment of six LLMs by OpenAI, Anthropic, and Cohere with German party positions and evaluate sycophancy based on a prompt experiment. We contribute to evaluating political bias and sycophancy in multi-party systems across major commercial LLMs. First, we develop the benchmark dataset GermanPartiesQA based on the Voting Advice Application Wahl-o-Mat covering 10 state and 1 national elections between 2021 and 2023. In our study, we find a left-green tendency across all examined LLMs. We then conduct our prompt experiment for which we use the benchmark and sociodemographic data of leading German parliamentarians to evaluate changes in LLMs responses. To differentiate between sycophancy and steerabilty, we use 'I am [politician X], ...' and 'You are [politician X], ...' prompts. Against our expectations, we do not observe notable differences between prompting 'I am' and 'You are'. While our findings underscore that LLM responses can be ideologically steered with political personas, they suggest that observed changes in LLM outputs could be better described as personalization to the given context rather than sycophancy., Comment: 12 pages

This article introduces the hardware-software co-design of the control system for an autonomy-enabled formula-style high-speed racecar that will be utilized as the deployment platform for high-level autonomy in the first ever head-to-head driverless race called the Indy Autonomous Challenge. The embedded control system needs to facilitate autonomous functionality, including perception, localization, and by-wire actuation, at high speeds and dynamic limits of the vehicle. Rapid maneuvering during the race, however, excites transient dynamics of the vehicle and the actuators. Compared to current autonomous driving focused on highway cruising and urban traffic, transient vehicle control imposes new challenges to the algorithm and system design. The presented work introduces the cascaded control structure employed by the IAC prototype to fully exploit the time scale separation between different control tasks. It is demonstrated by example way how the model-based control strategies and simulation are utilized to inform the decisions in the actuation, computation, perception, and software pipeline design decisions for the first-of-its-kind IAC racecar., Comment: 6 pages, 6 figures, 2021 IEEE International Conference on Robotics and Automation, Workshop on Opportunities and Challenges with Autonomous Racing

With the growing prevalence of AI, demand increases for efficient machine learning hardware. Physical systems are sought which combine image feature detection with the essential nonlinearity for tasks such as image classification. Existing physical hardware typically detects features linearly, then employs digital processing for nonlinear activation. Biological vision systems excel at nonlinear image processing. The retina detects features in ganglion cells via lateral inhibition, where cells nonlinearly compete for neuronal firing while supressing neighbouring cells. We present a bio-inspired 'retinomorphic' machine vision platform using an on-chip semiconductor network laser. The system detects multiple features in parallel via spatially-overlapping lasing modes, with integrated nonlinearity provided by antagonistic gain competition between modes - a photonic analogue of retinal inhibition. Parallel feature-detection enhances efficiency relative to feature-detection schemes which operate sequentially or via multiple device copies, with Si-compatible processing and a compact micron-scale footprint relative to existing mm-scale systems. We report 98.05% accuracy on MNIST-digits and 87.85% on Fashion-MNIST, with strong performance on short training datasets.

Strong optomechanical coupling -- a regime where mechanical motion is damped by environmental radiation -- has traditionally required demanding experimental ingredients such as superconducting resonators, high-quality optical cavities, or large magnetic fields. Here we demonstrate a room temperature, cavity-free, all-electric device reaching this regime at radio frequencies, enabled by a mechanically compliant parallel-plate capacitor with a nanoscale plate separation and an aspect ratio exceeding 1,000. The device has four orders of magnitude lower insertion loss than a comparable commercial quartz crystal, and achieves a position imprecision rivaling an optical interferometer. With the help of a back-action isolation scheme, we observe radiative cooling of mechanical motion by a remote cryogenic load. This work provides a technologically accessible route to high-precision sensing, transduction, and signal processing., Comment: 19+13 pages, 5+2 figures

Detecting single photons is essential for applications such as dark matter detection, quantum science and technology, and biomedical imaging. Superconducting nanowire single-photon detectors (SNSPDs) excel in this task due to their near-unity detection efficiency, sub-Hz dark count rates, and picosecond timing jitter. However, a local increase of current density (current crowding) in the bends of meander-shaped SNSPDs limits these performance metrics. By locally irradiating the straight segments of SNSPDs with helium ions while leaving the bends unirradiated, we realize current-crowding-free SNSPDs with simultaneously enhanced sensitivity: after irradiation with 800 ions/nm$\unicode{xB2}$, locally irradiated SNSPDs showed a relative saturation plateau width of 37% while fully irradiated SNSPDs reached only 10%. This larger relative plateau width allows operation at lower relative bias currents, thereby reducing the dark count rate while still detecting single photons efficiently. We achieve an internal detection efficiency of 94% for a wavelength of 780 nm with a dark count rate of 7 mHz near the onset of saturating detection efficiency., Comment: 11 pages, 7 figures

Current multimodal machine translation (MMT) systems rely on fully supervised data (i.e models are trained on sentences with their translations and accompanying images). However, this type of data is costly to collect, limiting the extension of MMT to other language pairs for which such data does not exist. In this work, we propose a method to bypass the need for fully supervised data to train MMT systems, using multimodal English data only. Our method, called ZeroMMT, consists in adapting a strong text-only machine translation (MT) model by training it on a mixture of two objectives: visually conditioned masked language modelling and the Kullback-Leibler divergence between the original and new MMT outputs. We evaluate on standard MMT benchmarks and the recently released CoMMuTE, a contrastive benchmark aiming to evaluate how well models use images to disambiguate English sentences. We obtain disambiguation performance close to state-of-the-art MMT models trained additionally on fully supervised examples. To prove that our method generalizes to languages with no fully supervised training data available, we extend the CoMMuTE evaluation dataset to three new languages: Arabic, Russian and Chinese. We further show that we can control the trade-off between disambiguation capabilities and translation fidelity at inference time using classifier-free guidance and without any additional data. Our code, data and trained models are publicly accessible., Comment: Preprint. Under review

Entangled quantum states are essential ingredients for many quantum technologies, but they must be validated before they are used. As a full characterization is prohibitively resource-intensive, recent work has focused on developing methods to efficiently extract a few parameters of interest, in a so-called verification framework. Most existing approaches are based on preparing an ensemble of nominally identical and independent (IID) quantum states, and then measuring each copy of the ensemble. However, this leaves no states left for the intended quantum tasks and the IID assumptions do not always hold experimentally. To overcome these challenges, we experimentally implement quantum state certification (QSC), which measures only a subset of the ensemble, certifying the fidelity of the remaining states. We use active optical switches to randomly sample from sources of two-photon Bell states and three-photon GHZ states, reporting statistically-sound fidelities in real time without destroying the entire ensemble. Additionally, our QSC protocol removes the assumption that the states are identical, is device-independent, and can achieve close $N^{-1}$ scaling, in the number of states measured $N$. Altogether, these benefits make our QSC protocol suitable for benchmarking large-scale quantum computing devices and deployed quantum communication setups relying on entanglement in both standard and adversarial situations., Comment: 8 pages, 4 figures

This technical report describes the CP-JKU team's submission for Task 4 Sound Event Detection with Heterogeneous Training Datasets and Potentially Missing Labels of the DCASE 24 Challenge. We fine-tune three large Audio Spectrogram Transformers, PaSST, BEATs, and ATST, on the joint DESED and MAESTRO datasets in a two-stage training procedure. The first stage closely matches the baseline system setup and trains a CRNN model while keeping the large pre-trained transformer model frozen. In the second stage, both CRNN and transformer are fine-tuned using heavily weighted self-supervised losses. After the second stage, we compute strong pseudo-labels for all audio clips in the training set using an ensemble of all three fine-tuned transformers. Then, in a second iteration, we repeat the two-stage training process and include a distillation loss based on the pseudo-labels, boosting single-model performance substantially. Additionally, we pre-train PaSST and ATST on the subset of AudioSet that comes with strong temporal labels, before fine-tuning them on the Task 4 datasets., Comment: Technical Report describing our system for DCASE2024 Challenge Task 4: https://dcase.community/challenge2024/task-sound-event-detection-with-heterogeneous-training-dataset-and-potentially-missing-labels-results Code: https://github.com/CPJKU/cpjku_dcase24. arXiv admin note: text overlap with arXiv:2407.12997

A central problem in building effective sound event detection systems is the lack of high-quality, strongly annotated sound event datasets. For this reason, Task 4 of the DCASE 2024 challenge proposes learning from two heterogeneous datasets, including audio clips labeled with varying annotation granularity and with different sets of possible events. We propose a multi-iteration, multi-stage procedure for fine-tuning Audio Spectrogram Transformers on the joint DESED and MAESTRO Real datasets. The first stage closely matches the baseline system setup and trains a CRNN model while keeping the pre-trained transformer model frozen. In the second stage, both CRNN and transformer are fine-tuned using heavily weighted self-supervised losses. After the second stage, we compute strong pseudo-labels for all audio clips in the training set using an ensemble of fine-tuned transformers. Then, in a second iteration, we repeat the two-stage training process and include a distillation loss based on the pseudo-labels, achieving a new single-model, state-of-the-art performance on the public evaluation set of DESED with a PSDS1 of 0.692. A single model and an ensemble, both based on our proposed training procedure, ranked first in Task 4 of the DCASE Challenge 2024., Comment: Code: https://github.com/CPJKU/cpjku_dcase24

Optical beams and starlight distorted by atmospheric turbulence can be corrected with adaptive optics systems to enable efficient coupling into single-mode fibers. Deformable mirrors, used to flatten the wavefront in astronomical telescopes, are costly, sensitive, and complex mechanical components that require careful calibration to enable high-quality imaging in astronomy, microscopy, and vision science. They are also impractical to deploy in large numbers for non-imaging applications like free-space optical communication. Here, we propose a photonic integrated c rcuit capable of spatially sampling the wavefront collected by the telescope and co-phasing the subapertures to maximize the flux delivered to an output single-mode fiber as the integrated photonic implementation of a deformable mirror. We present the results of end-to-end simulations to quantify the performance of the proposed photonic solution under varying atmospheric conditions toward realizing an adaptive optics system without a deformable mirror for free-space optical receivers., Comment: 14 pages, 8 figures, 1 table, to be published in Optics Express. Corresponding author: Momen Diab (momen.diab@utoronto.ca)

While text-to-image diffusion models have been shown to achieve state-of-the-art results in image synthesis, they have yet to prove their effectiveness in downstream applications. Previous work has proposed to generate data for image classifier training given limited real data access. However, these methods struggle to generate in-distribution images or depict fine-grained features, thereby hindering the generalization of classification models trained on synthetic datasets. We propose DataDream, a framework for synthesizing classification datasets that more faithfully represents the real data distribution when guided by few-shot examples of the target classes. DataDream fine-tunes LoRA weights for the image generation model on the few real images before generating the training data using the adapted model. We then fine-tune LoRA weights for CLIP using the synthetic data to improve downstream image classification over previous approaches on a large variety of datasets. We demonstrate the efficacy of DataDream through extensive experiments, surpassing state-of-the-art classification accuracy with few-shot data across 7 out of 10 datasets, while being competitive on the other 3. Additionally, we provide insights into the impact of various factors, such as the number of real-shot and generated images as well as the fine-tuning compute on model performance. The code is available at https://github.com/ExplainableML/DataDream., Comment: Accepted to ECCV 2024

We provide the first systematic technique for deriving witnesses of contextuality in prepare-transform-measure scenarios. More specifically, we show how linear quantifier elimination can be used to compute a polytope of correlations consistent with generalized noncontextuality in such scenarios. This polytope is specified as a set of noncontextuality inequalities that are necessary and sufficient conditions for observed data in the scenario to admit of a classical explanation relative to any linear operational identities, if one ignores some constraints from diagram preservation. While including these latter constraints generally leads to tighter inequalities, it seems that nonlinear quantifier elimination would be required to systematically include them. We also provide a linear program which can certify the nonclassicality of a set of numerical data arising in a prepare-transform-measure experiment. We apply our results to get a robust noncontextuality inequality for transformations that can be violated within the stabilizer subtheory. Finally, we give a simple algorithm for computing all the linear operational identities holding among a given set of states, of transformations, or of measurements., Comment: 15 pages + appendices; 3 figures

As Machine Learning (ML) becomes more prevalent in Industry 4.0, there is a growing need to understand how systematic approaches to bringing ML into production can be practically implemented in industrial environments. Here, MLOps comes into play. MLOps refers to the processes, tools, and organizational structures used to develop, test, deploy, and manage ML models reliably and efficiently. However, there is currently a lack of information on the practical implementation of MLOps in industrial enterprises. To address this issue, we conducted a multiple case study on MLOps in three large companies with dedicated MLOps teams, using established tools and well-defined model deployment processes in the Industry 4.0 environment. This study describes four of the companies' Industry 4.0 scenarios and provides relevant insights into their implementation and the challenges they faced in numerous projects. Further, we discuss MLOps processes, procedures, technologies, as well as contextual variations among companies.

We investigate the properties of a fairly large class of boundary conditions for the linearised Einstein equations in the Riemannian setting, ones which generalise the linearised counterpart of boundary conditions proposed by Anderson. Through the prism of the quest to quantise gravitational waves in curved spacetimes, we study their properties from the point of view of ellipticity, gauge invariance, and the existence of a spectral gap., Comment: 9 pages

This paper proposes small area estimation methods that utilize generalized tree-based machine learning techniques to improve the estimation of disaggregated means in small areas using discrete survey data. Specifically, we present two approaches based on random forests: the Generalized Mixed Effects Random Forest (GMERF) and a Mixed Effects Random Forest (MERF), both tailored to address challenges associated with count outcomes, particularly overdispersion. Our analysis reveals that the MERF, which does not assume a Poisson distribution to model the mean behavior of count data, excels in scenarios of severe overdispersion. Conversely, the GMERF performs best under conditions where Poisson distribution assumptions are moderately met. Additionally, we introduce and evaluate three bootstrap methodologies - one parametric and two non-parametric - designed to assess the reliability of point estimators for area-level means. The effectiveness of these methodologies is tested through model-based (and design-based) simulations and applied to a real-world dataset from the state of Guerrero in Mexico, demonstrating their robustness and potential for practical applications.

Notes on Commutative Alegbra and Algebraic Geometry covering rings, ideals, modules, presheaves, sheaves, schemes, homological algebra, \'etale cohomology and further topics that are more advanced., Comment: Mathematical inaccuracies in the paper

An integrated modeling framework for investigating the application of solid boron powder injection for real-time surface conditioning of plasma-facing components in tokamak environments is presented. Utilizing the DIII-D impurity powder dropper setup, this study simulates B powder injection scenarios ranging from mg to tens of mg per second, corresponding to boron flux rates of $10^{20}-10^{21}$ B/s in standard L-mode conditions. The comprehensive modeling approach combines EMC3-EIRENE for simulating the plasma background and DIS for the ablation and transport of the boron powder particles. The results show substantial transport of B to the inboard lower divertor, predominantly influenced by the main ion plasma flow. The powder particle size (5-250 $\mu$m) was found to be insignificant for the scenario considered. The effects of erosion and redeposition were considered to reconcile the discrepancies with experimental observations, which saw substantial deposition on the outer PFCs. For this purpose, the WallDYN3D code was updated to include B sources within the plasma domain and integrated into the modeling framework. The mixed-material migration modeling shows evolving boron deposition patterns, suggesting the formation of mixed B-C layers or predominantly B coverage depending on the powder mass flow rate. While the modeling outcomes at lower B injection rates tend to align with experimental observations, the prediction of near-pure B layers at higher rates has yet to be experimentally verified in the carbon environment of the DIII-D tokamak. The extensive reach of B layers found in the modeling suggests the need for modeling that encompasses the entire wall geometry for more accurate experimental correlations. This integrated approach sets a precedent for analyzing and applying real-time in-situ boron coating techniques in advanced tokamak scenarios, potentially extendable to ITER.

The tree-structured varying coefficient model (TSVC) is a flexible regression approach that allows the effects of covariates to vary with the values of the effect modifiers. Relevant effect modifiers are identified inherently using recursive partitioning techniques. To quantify uncertainty in TSVC models, we propose a procedure to construct confidence intervals of the estimated partition-specific coefficients. This task constitutes a selective inference problem as the coefficients of a TSVC model result from data-driven model building. To account for this issue, we introduce a parametric bootstrap approach, which is tailored to the complex structure of TSVC. Finite sample properties, particularly coverage proportions, of the proposed confidence intervals are evaluated in a simulation study. For illustration, we consider applications to data from COVID-19 patients and from patients suffering from acute odontogenic infection. The proposed approach may also be adapted for constructing confidence intervals for other tree-based methods.

Determining the local coordination of the active site is a pre-requisite for the reliable modeling of single-atom catalysts (SACs). Obtaining such information is difficult on powder-based systems, so much emphasis is placed on density functional theory-based computations based on idealized low-index surfaces of the support. In this work, we investigate how Pt atoms bind to the (1-102) facet of Fe2O3, a common support material in SAC. Using a combination of scanning tunneling microscopy (STM), x-ray photoelectron spectroscopy (XPS), and an extensive computational evolutionary search, we find that Pt atoms significantly reconfigure the support lattice to facilitate a pseudo-linear coordination to surface oxygen atoms. Despite breaking three surface Fe-O bonds, this geometry is favored by 0.84 eV over the best configuration involving an unperturbed support. We suggest that the linear O-Pt-O configuration is common in reactive Pt-based SAC systems because it balances thermal stability with the ability to adsorb reactants from the gas phase, and that extensive structural searches are likely necessary to determine realistic active site geometry in single-atom catalysis.

Low-energy electron diffraction (LEED) is a widely used technique in surface-science. Yet, it is rarely used to its full potential. The quantitative information about the surface structure, contained in the modulation of the intensities of the diffracted beams as a function of incident electron energy, LEED I(V), is underutilized. To acquire these data, minor adjustments would be required in most experimental setups, but existing analysis software is cumbersome to use. ViPErLEED (Vienna package for Erlangen LEED) lowers these barriers, introducing a combined solution for data acquisition, extraction, and computational analysis. These parts are discussed in three separate publications. Here, the focus is on the computational part of ViPErLEED, which performs automated LEED-I(V) calculations and structural optimization. Minimal user input is required, and the functionality is significantly enhanced compared to existing solutions. Computation is performed by embedding the Erlangen tensor-LEED package (TensErLEED). ViPErLEED manages parallelization, monitors convergence, and processes input and output. This makes LEED I(V) more accessible to new users while minimizing the potential for errors and the manual labor. Added functionality includes structure-dependent defaults, automatic detection of bulk and surface symmetries and their relationship, automated symmetry-preserving search procedures, adjustments to the TensErLEED code to handle larger systems, as well as parallelization and optimization. Modern file formats are used as input and output, and there is a direct interface to the Atomic Simulation Environment (ASE) package. The software is implemented primarily in Python (version >=3.7) and provided as an open-source package (GNU GPLv3 or later). A structure determination of the $\alpha$-Fe2O3(1-102)-(1x1) surface is presented as an example for the application of the software.

As part of the ViPErLEED project (Vienna package for Erlangen LEED, low-energy electron diffraction), computer programs have been developed for facile and user-friendly data extraction from movies of LEED images. The programs make use of some concepts from astronomical image processing and analysis. As a first step, flat-field and dark-frame corrections reduce the effects of inhomogeneities of the camera and screen. In a second step, for identifying all diffraction maxima ("spots"), it is sufficient to manually mark and label a single spot or very few spots. Then the program can automatically identify all other spots and determine the distortions of the image. This forms the basis for automatic spot tracking (following the "beams" as they move across the LEED screen) and intensity measurement. Even for complex structures with hundreds to a few thousand diffraction beams, this step takes less than a minute. The package also includes a program for further processing of these I(V) curves (averaging of equivalent beams, manual and/or automatic selection, smoothing) as well as several utilities. The software is implemented as a set of plugins for the public-domain image processing program ImageJ and provided as an open-source package.

Failure mode and effects analysis (FMEA) is a critical tool for mitigating potential failures, particular during ramp-up phases of new products. However, its effectiveness is often limited by the missing reasoning capabilities of the FMEA tools, which are usually tabular structured. Meanwhile, large language models (LLMs) offer novel prospects for fine-tuning on custom datasets for reasoning within FMEA contexts. However, LLMs face challenges in tasks that require factual knowledge, a gap that retrieval-augmented generation (RAG) approaches aim to fill. RAG retrieves information from a non-parametric data store and uses a language model to generate responses. Building on this idea, we propose to advance the non-parametric data store with a knowledge graph (KG). By enhancing the RAG framework with a KG, our objective is to leverage analytical and semantic question-answering capabilities on FMEA data. This paper contributes by presenting a new ontology for FMEA observations, an algorithm for creating vector embeddings from the FMEA KG, and a KG enhanced RAG framework. Our approach is validated through a human study and we measure the performance of the context retrieval recall and precision.

Large language models (LLMs) possess extensive parametric knowledge, but this knowledge is difficult to update with new information because retraining is very expensive and infeasible for closed-source models. Knowledge editing (KE) has emerged as a viable solution for updating the knowledge of LLMs without compromising their overall performance. On-the-fly KE methods, inspired by in-context learning (ICL), have shown great promise and allow LLMs to be treated as black boxes. In the past, KE was primarily employed in English contexts, whereas the potential for cross-lingual KE in current English-centric LLMs has not been fully explored. To foster more research in this direction, we introduce the BMIKE-53 benchmark for evaluating cross-lingual KE on 53 diverse languages across three KE task types. We also propose a gradient-free KE method called Multilingual In-context Knowledge Editing (MIKE) and evaluate it on BMIKE-53. Our evaluation focuses on cross-lingual knowledge transfer in terms of reliability, generality, locality, and portability, offering valuable insights and a framework for future research in cross-lingual KE. Our code and data are publicly accessible via the anonymous repository at https://anonymous.4open.science/r/MIKE., Comment: 12 pages, 4 figures

Understanding optical absorption in silicon nitride is crucial for cutting-edge technologies like photonic integrated circuits, nanomechanical photothermal infrared sensing and spectroscopy, and cavity optomechanics. Yet, the origin of its strong dependence on film deposition and fabrication process is not fully understood. This Letter leverages nanomechanical photothermal sensing to investigate optical extinction $\kappa_{\mathrm{ext}}$ at 632.8 nm wavelength in LPCVD SiN strings across a wide range of deposition-related tensile stresses ($200-850$ MPa). Measurements reveal a reduction in $\kappa_{\mathrm{ext}}$ from 10$^3$ to 10$^1$ ppm with increasing stress, correlated to variations in Si/N content ratio. Within the band-fluctuations framework, this trend indicates an increase of the energy bandgap with the stress, ultimately reducing absorption. Overall, this study showcases the power and simplicity of nanomechanical photothermal sensing for low absorption measurements, offering a sensitive, scattering-free platform for material analysis in nanophotonics and nanomechanics., Comment: Main text: 7 pages, 3 figures, 1 table Supporting Information: 4 pages, 4 figures, 1 table

The explosively growing communication traffic in datacenters imposes increasingly stringent performance requirements on the underlying networks. Over the last years, researchers have developed innovative optical switching technologies that enable reconfigurable datacenter networks (RCDNs) which support very fast topology reconfigurations. This paper presents D3, a novel and feasible RDCN architecture that improves throughput and flow completion time. D3 quickly and jointly adapts its links and packet scheduling toward the evolving demand, combining both demand-oblivious and demand-aware behaviors when needed. D3 relies on a decentralized network control plane supporting greedy, integrated-multihop, IP-based routing, allowing to react, quickly and locally, to topological changes without overheads. A rack-local synchronization and transport layer further support fast network adjustments. Moreover, we argue that D3 can be implemented using the recently proposed Sirius architecture (SIGCOMM 2020). We report on an extensive empirical evaluation using packet-level simulations. We find that D3 improves throughput by up to 15% and preserves competitive flow completion times compared to the state of the art. We further provide an analytical explanation of the superiority of D3, introducing an extension of the well-known Birkhoff-von Neumann decomposition, which may be of independent interest.

In the random $r$-neighbour bootstrap percolation process on a graph $G$, a set of initially infected vertices is chosen at random by retaining each vertex of $G$ independently with probability $p\in (0,1)$, and "healthy" vertices get infected in subsequent rounds if they have at least $r$ infected neighbours. A graph $G$ \emph{percolates} if every vertex becomes eventually infected. A central problem in this process is to determine the critical probability $p_c(G,r)$, at which the probability that $G$ percolates passes through one half. In this paper, we study random $2$-neighbour bootstrap percolation on the $n$-dimensional Hamming graph $\square_{i=1}^n K_k$, which is the graph obtained by taking the Cartesian product of $n$ copies of the complete graph $K_k$ on $k$ vertices. We extend a result of Balogh and Bollob\'{a}s [Bootstrap percolation on the hypercube, Probab. Theory Related Fields. 134 (2006), no. 4, 624-648. MR2214907] about the asymptotic value of the critical probability $p_c(Q^n,2)$ for random $2$-neighbour bootstrap percolation on the $n$-dimensional hypercube $Q^n=\square_{i=1}^n K_2$ to the $n$-dimensional Hamming graph $\square_{i=1}^n K_k$, determining the asymptotic value of $p_c\left(\square_{i=1}^n K_k,2\right)$, up to multiplicative constants (when $n \rightarrow \infty$), for arbitrary $k \in \mathbb N$ satisfying $2 \leq k\leq 2^{\sqrt{n}}$.

The rise of machine learning methods on heavily resource constrained devices requires not only the choice of a suitable model architecture for the target platform, but also the optimization of the chosen model with regard to execution time consumption for inference in order to optimally utilize the available resources. Random forests and decision trees are shown to be a suitable model for such a scenario, since they are not only heavily tunable towards the total model size, but also offer a high potential for optimizing their executions according to the underlying memory architecture. In addition to the straightforward strategy of enforcing shorter paths through decision trees and hence reducing the execution time for inference, hardware-aware implementations can optimize the execution time in an orthogonal manner. One particular hardware-aware optimization is to layout the memory of decision trees in such a way, that higher probably paths are less likely to be evicted from system caches. This works particularly well when splits within tree nodes are uneven and have a high probability to visit one of the child nodes. In this paper, we present a method to reduce path lengths by rewarding uneven probability distributions during the training of decision trees at the cost of a minimal accuracy degradation. Specifically, we regularize the impurity computation of the CART algorithm in order to favor not only low impurity, but also highly asymmetric distributions for the evaluation of split criteria and hence offer a high optimization potential for a memory architecture-aware implementation. We show that especially for binary classification data sets and data sets with many samples, this form of regularization can lead to an reduction of up to approximately four times in the execution time with a minimal accuracy degradation.

A quick overview of category theory and topos theory including slice categories, monics, epics, isos, diagrams, cones, cocones, limits, colimits, products and coproducts, pushouts and pullbacks, equalizers and coequalizers, initial and terminal objects, exponential objects, subobjects, subobject classifiers, the definition of a topos, algebras of subobjects, functors, natural transformations and adjoint functors. This paper is refashioned and adopted from Richard Pettigrew's university notes., Comment: 37 pages

Scientific Machine Learning is a new class of approaches that integrate physical knowledge and mechanistic models with data-driven techniques for uncovering governing equations of complex processes. Among the available approaches, Universal Differential Equations (UDEs) are used to combine prior knowledge in the form of mechanistic formulations with universal function approximators, like neural networks. Integral to the efficacy of UDEs is the joint estimation of parameters within mechanistic formulations and the universal function approximators using empirical data. The robustness and applicability of resultant models, however, hinge upon the rigorous quantification of uncertainties associated with these parameters, as well as the predictive capabilities of the overall model or its constituent components. With this work, we provide a formalisation of uncertainty quantification (UQ) for UDEs and investigate important frequentist and Bayesian methods. By analysing three synthetic examples of varying complexity, we evaluate the validity and efficiency of ensembles, variational inference and Markov chain Monte Carlo sampling as epistemic UQ methods for UDEs., Comment: Shared last authorship between W.W. and J.H

Multimodal Large Language Models (mLLMs) are trained on a large amount of text-image data. While most mLLMs are trained on caption-like data only, Alayrac et al. [2022] showed that additionally training them on interleaved sequences of text and images can lead to the emergence of in-context learning capabilities. However, the dataset they used, M3W, is not public and is only in English. There have been attempts to reproduce their results but the released datasets are English-only. In contrast, current multilingual and multimodal datasets are either composed of caption-like only or medium-scale or fully private data. This limits mLLM research for the 7,000 other languages spoken in the world. We therefore introduce mOSCAR, to the best of our knowledge the first large-scale multilingual and multimodal document corpus crawled from the web. It covers 163 languages, 315M documents, 214B tokens and 1.2B images. We carefully conduct a set of filtering and evaluation steps to make sure mOSCAR is sufficiently safe, diverse and of good quality. We additionally train two types of multilingual model to prove the benefits of mOSCAR: (1) a model trained on a subset of mOSCAR and captioning data and (2) a model train on captioning data only. The model additionally trained on mOSCAR shows a strong boost in few-shot learning performance across various multilingual image-text tasks and benchmarks, confirming previous findings for English-only mLLMs., Comment: Preprint. Under review