Search

The use of machine learning (ML) methods for development of robust and flexible visual inspection system has shown promising. However their performance is highly dependent on the amount and diversity of training data. This is often restricted not only due to costs but also due to a wide variety of defects and product surfaces which occur with varying frequency. As such, one can not guarantee that the acquired dataset contains enough defect and product surface occurrences which are needed to develop a robust model. Using parametric synthetic dataset generation, it is possible to avoid these issues. In this work, we introduce a complete pipeline which describes in detail how to approach image synthesis for surface inspection - from first acquisition, to texture and defect modeling, data generation, comparison to real data and finally use of the synthetic data to train a defect segmentation model. The pipeline is in detail evaluated for milled and sandblasted aluminum surfaces. In addition to providing an in-depth view into each step, discussion of chosen methods, and presentation of ML results, we provide a comprehensive dual dataset containing both real and synthetic images., Comment: Initial preprint, 21 pages, 21 figures, 6 tables

The ultralight dark photon is a well-motivated, hypothetical dark matter candidate. In a dilute plasma, they can resonantly convert into photons, and heat up the intergalactic medium between galaxies. In this work, we explore the dark photon dark matter parameter space by comparing synthetic Lyman-$\alpha$ forest data from cosmological hydrodynamical simulations to observational data from VLT/UVES of the quasar HE0940-1050 ($z_{\rm em}=3.09$). We use a novel flux normalization technique that targets under-dense gas, reshaping the flux probability distribution. Not only do we place robust constraints on the kinetic mixing parameter of dark photon dark matter, but notably our findings suggest that this model can still reconcile simulated and observed Doppler parameter distributions of $z\sim0$ Lyman-$\alpha$ lines, as seen by HST/COS. This work opens new pathways for the use of the Lyman-$\alpha$ forest to explore new physics, and can be extended to other scenarios such as primordial black hole evaporation, dark matter decay, and annihilation., Comment: 14 pages, 7 figures, 1 table

In the past, high-z AGNs were given a minor role as possible drivers of reionization, despite initial evidences in favor of their large space densities at low luminosities by Chandra and HST. Recent observations from JWST are finding relatively large numbers of faint AGNs at z>4, convincingly confirming these early results. We present a sample of z~5 AGNs (both from wide, shallow ground-based surveys and from deep, pencil-beam observations from JWST), allowing to estimate their space densities with unprecedented accuracy. The bright end (M1450<-26) of the z~5 AGN luminosity function is well constrained, with a rather steep slope. The faint end (M1450>-22) indicates a high space density, the scatter is significant and the knee (M1450~-24) is mostly undetermined. Comparisons with state-of-the-art models find reasonable agreement with the observed AGN luminosity function at z=5, while the predicted space density evolution at higher redshifts appears to be too fast with respect to observational constraints. Given the large variance at the faint end, we consider different options in fitting the luminosity functions and deriving the ionizing emissivity. Even in the most conservative scenario, the photo-ionization rate produced by z~5 AGNs is consistent with the UV background measurements. A slow evolution of the space density of faint AGNs is observed, indicating that active SMBHs are probably producing large amounts of ionizing photons at z>6, well into the epoch of reionization. This is an important indication that high-z AGNs could be the major contributors to the reionization of the Universe., Comment: Accepted for publication on ApJ, 20 pages, 9 figures, 3 tables

Optical trapping enables precise control of individual particles of different sizes, such as atoms, molecules, or nanospheres. Optical tweezers provide free-space omnidirectional optical trapping of objects in laboratories around the world. As an alternative to standard macroscopic setups based on lenses, which are inherently bound by the diffraction limit, plasmonic and photonic nanostructures promise trapping by near-field optical effects on the extreme nanoscale. However, the practical design of lossless waveguide-coupled nanostructures capable of trapping sub-wavelength-sized particles in all spatial directions has until now proven insurmountable. In this work, we demonstrate an omnidirectional optical trap realized by inverse-designing fabrication-ready integrated dielectric nanocavities. The sub-wavelength optical trap is designed to rely solely on the gradient force and is thus particle-size agnostic. In particular, we show how a trapped particle with a radius of 15 nm experiences a force strong enough to overcome room-temperature thermal fluctuations. Furthermore, through the robust inverse design framework, we tailor manufacturable devices operating at short-wave infrared and near-infrared wavelengths. Our results open a new regime of levitated optical trapping by achieving a deep trapping potential capable of trapping single sub-wavelength particles in all directions using optical gradient forces. We anticipate potentially groundbreaking applications of the optimized optical trapping system for biomolecular analysis in aqueous environments, levitated cavity-optomechanics, and cold atom physics, constituting an important step towards realizing integrated bio-nanophotonics and mesoscopic quantum mechanical experiments.

We formulate a general hybrid quantum-classical technique to describe the interaction of diatomic molecules with XUV pulses. We demonstrate the accuracy of our model in the context of the interaction of the O$_2$ molecule with an XUV pulse with photon energy ranging from 20 eV to 42 eV. We account for the electronic structure and electron ionization quantum mechanically employing accurate molecular continuum wavefunctions. We account for the motion of the nuclei using classical equations of motion. However, the force of the nuclei is computed by obtaining accurate potential-energy curves of O$_2$ up to O$_2^{2+}$, relevant to the 20 eV-42 eV photon-energy range, using advanced quantum-chemistry techniques. We find the dissociation limits of these states and the resulting atomic fragments and employ the Velocity Verlet algorithm to compute the velocities of these fragments. We incorporate both electron ionization and nuclear motion in a stochastic Monte-Carlo simulation and identify the ionization and dissociation pathways when O$_2$ interacts with an XUV pulse. Focusing on the O$^+$ + O$^+$ dissociation pathway, we obtain the kinetic-energy release distributions of the atomic fragments and find very good agreement with experimental results. Also, we explain the main features of the KER in terms of ionization sequences consisting of two sequential single-photon absorptions resulting in different O$^+$ and O$^{2+}$ electronic state configurations involved in the two transitions., Comment: 15 pages, 10 Figures

We introduce Bonito, an open-source model for conditional task generation that converts unannotated text into task-specific training datasets for instruction tuning. We aim to enable zero-shot task adaptation of large language models on users' specialized, private data. We train Bonito by fine-tuning a pretrained large language model on a new large-scale dataset with 1.65M examples created by remixing existing instruction tuning datasets into meta-templates. The meta-templates for a dataset produce training examples where the input is the unannotated text and the task attribute and the output consists of the instruction and the response. We use Bonito to generate synthetic tasks for seven datasets from specialized domains with unannotated text across three task types -- yes-no question answering, extractive question answering, and natural language inference -- and adapt language models. We show that Bonito significantly improves the average performance of pretrained and instruction tuned models over the de facto self supervised baseline. For example, adapting Mistral-Instruct-v2 and instruction tuned variants of Mistral and Llama2 with Bonito improves the strong zero-shot performance by 22.1 F1 points whereas the next word prediction objective undoes some of the benefits of instruction tuning and reduces the average performance by 0.8 F1 points. We conduct additional experiments with Bonito to understand the effects of the domain, the size of the training set, and the choice of alternative synthetic task generators. Overall, we show that learning with synthetic instruction tuning datasets is an effective way to adapt language models to new domains. The model, dataset, and code are available at https://github.com/BatsResearch/bonito., Comment: ACL Findings 2024

Primordial abundances of light elements are sensitive to the physics of the early Universe and can directly constrain cosmological quantities, such as the baryon-to-photon ratio $\eta_{10}$, the baryon density and the number of neutrino families. Deuterium is especially suited for these studies: its primordial abundance is sensitive and monotonically dependent on $\eta_{10}$, allowing an independent measurement of the cosmic baryon density that can be compared, for instance, against the Planck satellite data. The primordial deuterium abundance can be measured in high $H_I$ column density absorption systems towards distant quasars. We report here a new measurement, based on high-resolution ESPRESSO data, of the primordial $D_I$ abundance of a system at redshift $z \sim 3.572$, towards PKS1937-101. Using only ESPRESSO data, we find a D/H ratio of $2.638\pm0.128 \times 10^{-5}$, while including the available UVES data improves the precision, leading to a ratio of $2.608 \pm 0.102 \times 10^{-5}$. The results of this analysis agree with those of the most precise existing measurements. We find that the relatively low column density of this system ($\log{N_{\rm H_I}/ {\rm cm}^{-2}}\sim18 $) introduces modelling uncertainties, which become the main contributor to the error budget., Comment: Accepted for publication in MNRAS. 16 pages, 10 figures and 6 tables (4 in main text, 2 in Appendix). Full model is available at the end of the manuscript

The gravitationally lensed quasar J014516.6-094517 at z=2.719 has been observed with the ESPRESSO instrument at the ESO VLT to obtain high-fidelity spectra of the two images A and B with a resolving power R=70000. At the redshifts under investigation (2.1 < z < 2.7), the Lyman forests along the two sightlines are separated by sub-kiloparsec physical distances and exhibit a strong correlation. We find that the two forests are indistinguishable at the present level of signal-to-noise ratio and do not show any global velocity shift, with the cross-correlation peaking at $\Delta v = 12 \pm 48$ m/s. The distribution of the difference in velocity of individual Lyman-$\alpha$ features is compatible with a null average and a mean absolute deviation of 930 m/s. Significant differences in NHI column density are not detected, putting a limit to the RMS fluctuation in the baryon density on $\leq 1$ proper kpc scales of $\Delta \rho / \rho < 3$%. On the other hand, metal lines show significant differences both in velocity structure and in column density. A toy model shows that the difference in velocity of the metal features between the two sightlines is compatible with the the motions of the baryonic component associated to dark matter halos of typical mass $M\simeq 2\times 10^{10} M_\odot$, also compatible with the observed incidence of the metal systems. The present observations confirm the feasibility of the Sandage test of the cosmic redshift drift with high-fidelity spectroscopy of the Lyman forest of distant, bright quasars, but also provide an element of caution about the intrinsic noise associated to the usage of metal features for the same purpose., Comment: 17 pages, 10 figures, 5 tables, accepted for publication in MNRAS

We present an optical cavity with deep sub-wavelength confinement of light in a region that simultaneously works as a quantum dot. The design is based on a dielectric membrane with a buried quantum well and restricts the electron and hole wave functions to the area of the optical hotspot in order to overcome the challenge of co-locating an optical cavity with a quantum emitter. Combined with proper surface passivation, this geometry points towards the deterministic fabrication of functional quantum dots in optical cavities by lithographic means.

Recent research shows that Large Language Models (LLMs) exhibit a compelling level of proficiency in Theory of Mind (ToM) tasks. This ability to impute unobservable mental states to others is vital to human social cognition and may prove equally important in principal-agent relations between individual humans and Artificial Intelligences (AIs). In this paper, we explore how a mechanism studied in developmental psychology known as Violation of Expectation (VoE) can be implemented to reduce errors in LLM prediction about users by leveraging emergent ToM affordances. And we introduce a \textit{metacognitive prompting} framework to apply VoE in the context of an AI tutor. By storing and retrieving facts derived in cases where LLM expectation about the user was violated, we find that LLMs are able to learn about users in ways that echo theories of human learning. Finally, we discuss latent hazards and augmentative opportunities associated with modeling user psychology and propose ways to mitigate risk along with possible directions for future inquiry.

Deep reinforcement learning (RL) has shown promising results in robot motion planning with first attempts in human-robot collaboration (HRC). However, a fair comparison of RL approaches in HRC under the constraint of guaranteed safety is yet to be made. We, therefore, present human-robot gym, a benchmark suite for safe RL in HRC. Our benchmark suite provides eight challenging, realistic HRC tasks in a modular simulation framework. Most importantly, human-robot gym includes a safety shield that provably guarantees human safety. We are, thereby, the first to provide a benchmark suite to train RL agents that adhere to the safety specifications of real-world HRC. This bridges a critical gap between theoretic RL research and its real-world deployment. Our evaluation of six tasks led to three key results: (a) the diverse nature of the tasks offered by human-robot gym creates a challenging benchmark for state-of-the-art RL methods, (b) incorporating expert knowledge in RL training in the form of an action-based reward can outperform the expert, and (c) our agents negligibly overfit to training data.

It is shown that semiconductor nanoscale resonators with extreme dielectric confinement accelerate the diffusion of electron-hole pairs excited by nonlinear absorption. These novel cavity designs may lead to optical switches with superior modulation speeds compared to conventional geometries. The response function of the effective carrier density is computed by an efficient eigenmode expansion technique. A few eigenmodes of the diffusion equation conveniently capture the long-timescale carrier decay rate, which is advantageous compared to time-domain simulations. Notably, the eigenmode approach elucidates the contribution to carrier diffusion of the in-plane and out-of-plane cavity geometry, which may guide future designs.

We present a semi-analytical framework for studying interactions between quantum emitters and general electromagnetic resonators. The method relies on the Lippmann-Schwinger equation to calculate the complex resonance frequencies of the coupled system based only on a single calculation for the electromagnetic resonator without the quantum emitter and with no fitting parameters. This is in stark contrast to standard approaches in the literature, in which the properties of the coupled system are fitted from calculated spectra. As an application example, we consider a recent dielectric cavity design featuring deep subwavelength confinement of light. We find the expected anti-crossing of the emitter and cavity resonance frequencies, and comparing to independent reference calculations, we find an extraordinary quantitative agreement with a relative error below one part in ten thousand. In order to unambiguously connect with the Jaynes-Cummings model, we derive an explicit expression relating the classical description of the emitter, as modeled by a spherical inclusion with a Lorentzian material response, to the dipole moment of the corresponding quantum optical model. The combined framework therefore enables classical calculations to be used for evaluating the coupling strength entering quantum optical theories in a transparent way., Comment: 18 pages, 7 figures, research article

We describe a numerical time-domain approach for high-accuracy calculations of Casimir-Polder forces near micro-structured materials. The use of a time-domain formulation enables the investigation of a broad range of materials described by advanced material models, including nonlocal response functions. We validate the method by a number of example calculations for which we thoroughly investigate the convergence properties of the method, and comparing to analytical reference calculations, we find average relative errors as low as a few parts in a million. As an application example, we investigate the anisotropy-induced repulsive behavior of the Casimir-Polder force near a sharp gold wedge described by a hydrodynamic Drude model., Comment: 17 pages, 11 figures