Search

Accurate diagnosis of heart arrhythmias requires the interpretation of electrocardiograms (ECG), which capture the electrical activity of the heart. Automating this process through machine learning is challenging due to the need for large annotated datasets, which are difficult and costly to collect. To address this issue, transfer learning is often employed, where models are pre-trained on large datasets and fine-tuned for specific ECG classification tasks with limited labeled data. Self-supervised learning has become a widely adopted pre-training method, enabling models to learn meaningful representations from unlabeled datasets. In this work, we explore the joint-embedding predictive architecture (JEPA) for self-supervised learning from ECG data. Unlike invariance-based methods, JEPA does not rely on hand-crafted data augmentations, and unlike generative methods, it predicts latent features rather than reconstructing input data. We create a large unsupervised pre-training dataset by combining ten public ECG databases, amounting to over one million records. We pre-train Vision Transformers using JEPA on this dataset and fine-tune them on various PTB-XL benchmarks. Our results show that JEPA outperforms existing invariance-based and generative approaches, achieving an AUC of 0.945 on the PTB-XL all statements task. JEPA consistently learns the highest quality representations, as demonstrated in linear evaluations, and proves advantageous for pre-training even in the absence of additional data.

ChatGPT disrupted the application of machine-learning methods and drastically reduced the usage barrier. Chatbots are now widely used in a lot of different situations. They provide advice, assist in writing source code, or assess and summarize information from various sources. However, their scope is not only limited to aiding humans; they can also be used to take on tasks like negotiating or bargaining. To understand the implications of Chatbot usage on bargaining situations, we conduct a laboratory experiment with the ultimatum game. In the ultimatum game, two human players interact: The receiver decides on accepting or rejecting a monetary offer from the proposer. To shed light on the new bargaining situation, we let ChatGPT provide an offer to a human player. In the novel design, we vary the wealth of the receivers. Our results indicate that humans have the same beliefs about other humans and chatbots. However, our results contradict these beliefs in an important point: Humans favor poor receivers as correctly anticipated by the humans, but ChatGPT favors rich receivers which the humans did not expect to happen. These results imply that ChatGPT's answers are not aligned with those of humans and that humans do not anticipate this difference.

We obtain new complexity bounds for computing a triangular integral basis of a number field or a function field. We reach for function fields a softly linear cost with respect to the size of the output when the residual characteristic is zero or big enough. Analogous results are obtained for integral basis of fractional ideals, key ingredients towards fast computation of Riemann-Roch spaces. The proof is based on the recent fast OM algorithm of the authors and on the MaxMin algorithm of Stainsby, together with optimal truncation bounds and a precise complexity analysis., Comment: 22 pages

The Large Array Survey Telescope (LAST) is a wide-field telescope designed to explore the variable and transient sky with a high cadence and to be a test-bed for cost-effective telescope design. A LAST node is composed of 48 (32 already deployed), 28-cm f/2.2 telescopes. A single telescope has a 7.4 deg^2 field of view and reaches a 5-sigma limiting magnitude of 19.6 (21.0) in 20s (20x20s) (filter-less), while the entire system provides a 355 deg^2 field of view. The basic strategy of LAST is to obtain multiple 20-s consecutive exposures of each field (a visit). Each telescope carries a 61 Mpix camera, and the system produces, on average, about 2.2 Gbit/s. This high data rate is analyzed in near real-time at the observatory site, using limited computing resources (about 700 cores). Given this high data rate, we have developed a new, efficient data reduction and analysis pipeline. The data pipeline includes two major parts: (i) Processing and calibration of single images, followed by a coaddition of the visit's exposures. (ii) Building the reference images and performing image subtraction and transient detection. Here we describe in detail the first part of the pipeline. Among the products of this pipeline are photometrically and astrometrically calibrated single and coadded images, 32-bit mask images marking a wide variety of problems and states of each pixel, source catalogs built from individual and coadded images, Point Spread Function (PSF) photometry, merged source catalogs, proper motion and variability indicators, minor planets detection, calibrated light curves, and matching with external catalogs. The entire pipeline code is made public. Finally, we demonstrate the pipeline performance on real data taken by LAST., Comment: Accepted to PASP, 15 pages, 10 figures

We show how to assign labels of size $\tilde O(1)$ to the vertices of a directed planar graph $G$, such that from the labels of any three vertices $s,t,f$ we can deduce in $\tilde O(1)$ time whether $t$ is reachable from $s$ in the graph $G\setminus \{f\}$. Previously it was only known how to achieve $\tilde O(1)$ queries using a centralized $\tilde O(n)$ size oracle [SODA'21].

Deception detection is an interdisciplinary field attracting researchers from psychology, criminology, computer science, and economics. We propose a multimodal approach combining deep learning and discriminative models for automated deception detection. Using video modalities, we employ convolutional end-to-end learning to analyze gaze, head pose, and facial expressions, achieving promising results compared to state-of-the-art methods. Due to limited training data, we also utilize discriminative models for deception detection. Although sequence-to-class approaches are explored, discriminative models outperform them due to data scarcity. Our approach is evaluated on five datasets, including a new Rolling-Dice Experiment motivated by economic factors. Results indicate that facial expressions outperform gaze and head pose, and combining modalities with feature selection enhances detection performance. Differences in expressed features across datasets emphasize the importance of scenario-specific training data and the influence of context on deceptive behavior. Cross-dataset experiments reinforce these findings. Despite the challenges posed by low-stake datasets, including the Rolling-Dice Experiment, deception detection performance exceeds chance levels. Our proposed multimodal approach and comprehensive evaluation shed light on the potential of automating deception detection from video modalities, opening avenues for future research., Comment: 29 pages, 17 figures (19 if counting subfigures)

In this paper, we study THz simultaneous wireless information and power transfer (SWIPT) systems. Since coherent information detection is challenging at THz frequencies and Schottky diodes may not be efficient for THz energy harvesting (EH) and information detection, we employ unipolar amplitude shift keying (ASK) modulation at the transmitter (TX) and a resonant tunnelling diode (RTD)-based EH circuit at the receiver (RX) to extract both information and power from the RX signal. We model the dependence of the instantaneous output power at the RX on the instantaneous received power by a non-linear piecewise function, whose parameters are adjusted to fit circuit simulation results. To determine the rate-power tradeoff in THz SWIPT systems, we derive the distribution of the TX signal that maximizes the mutual information between the TX and RX signals subject to constraints on the required average harvested power at the RX and the peak signal amplitude at the TX. Since the computational complexity of maximizing the mutual information may be too high for real-time THz SWIPT systems, for high and low required average harvested powers, we also obtain the suboptimal input signal distribution that maximizes the achievable information rate numerically and in closed form, respectively. Furthermore, based on the obtained results, we propose a suboptimal closed-form TX distribution which also achieves a desired harvested power at the RX. Our simulation results show that a lower reverse current flow and a higher breakdown voltage of the employed RTD are preferable when the input signal power at the RX is low and high, respectively. Finally, we demonstrate that for low and high received signal powers, the rate-power tradeoff of THz SWIPT systems is determined by the peak amplitude of the TX signal and the maximum instantaneous harvested power, respectively., Comment: 32 pages, 10 figures, submitted for a possible journal publication

In this paper, we study terahertz (THz) simultaneous wireless information and power transfer (SWIPT) for future micro-scale 6G Internet-of-Things (IoT) networks. Since Schottky diodes are not efficient for THz energy harvesting (EH), we propose resonant tunneling diodes (RTDs) for EH at the IoT receiver (RX). As the electrical properties of RTDs are different from those of Schottky diodes, we develop a novel closed-form EH model for RTD-based RXs. In particular, we model the dependency of the instantaneous RX output power on the instantaneous received power by a non-linear piecewise function, whose parameters are adjusted to fit circuit simulation results. Furthermore, since coherent information detection is challenging at THz frequencies, we employ unipolar amplitude shift keying (ASK) modulation at the transmitter (TX) and utilize the RTD-based EH circuit at the RX to extract both information and energy from the received signal. We formulate an optimization problem to maximize the mutual information between the TX and RX signals subject to constraints on the peak amplitude of the transmitted signal and the required average harvested power at the RX. Moreover, we determine a feasibility condition for the formulated problem and, for high and low required average harvested powers, we derive the achievable information rate numerically and in closed form, respectively. Our simulation results highlight a tradeoff between the information rate and the average harvested power. Finally, we show that this tradeoff is determined by the peak amplitude of the transmitted signal and the maximum instantaneous harvested power for low and high received signal powers, respectively., Comment: 6 pages, 5 figures, submitted for possible conference publication

The Voronoi diagrams technique was introduced by Cabello to compute the diameter of planar graphs in subquadratic time. We present novel applications of this technique in static, fault-tolerant, and partially-dynamic undirected unweighted planar graphs, as well as some new limitations. 1. In the static case, we give $n^{3+o(1)}/D^2$ and $\tilde{O}(n\cdot D^2)$ time algorithms for computing the diameter of a planar graph $G$ with diameter $D$. These are faster than the state of the art $\tilde{O}(n^{5/3})$ when $Dn^{2/3}$. 2. In the fault-tolerant setting, we give an $n^{7/3+o(1)}$ time algorithm for computing the diameter of $G\setminus \{e\}$ for every edge $e$ in $G$ the replacement diameter problem. Compared to the naive $\tilde{O}(n^{8/3})$ time algorithm that runs the static algorithm for every edge. 3. In the incremental setting, where we wish to maintain the diameter while while adding edges, we present an algorithm with total running time $n^{7/3+o(1)}$. Compared to the naive $\tilde{O}(n^{8/3})$ time algorithm that runs the static algorithm after every update. 4. We give a lower bound (conditioned on the SETH) ruling out an amortized $O(n^{1-\varepsilon})$ update time for maintaining the diameter in *weighted* planar graph. The lower bound holds even for incremental or decremental updates. Our upper bounds are obtained by novel uses and manipulations of Voronoi diagrams. These include maintaining the Voronoi diagram when edges of the graph are deleted, allowing the sites of the Voronoi diagram to lie on a BFS tree level (rather than on boundaries of $r$-division), and a new reduction from incremental diameter to incremental distance oracles that could be of interest beyond planar graphs. Our lower bound is the first lower bound for a dynamic planar graph problem that is conditioned on the SETH.

The Large Array Survey Telescope (LAST) is a wide-field visible-light telescope array designed to explore the variable and transient sky with a high cadence. LAST will be composed of 48, 28-cm f/2.2 telescopes (32 already installed) equipped with full-frame backside-illuminated cooled CMOS detectors. Each telescope provides a field of view (FoV) of 7.4 deg^2 with 1.25 arcsec/pix, while the system FoV is 355 deg^2 in 2.9 Gpix. The total collecting area of LAST, with 48 telescopes, is equivalent to a 1.9-m telescope. The cost-effectiveness of the system (i.e., probed volume of space per unit time per unit cost) is about an order of magnitude higher than most existing and under-construction sky surveys. The telescopes are mounted on 12 separate mounts, each carrying four telescopes. This provides significant flexibility in operating the system. The first LAST system is under construction in the Israeli Negev Desert, with 32 telescopes already deployed. We present the system overview and performances based on the system commissioning data. The Bp 5-sigma limiting magnitude of a single 28-cm telescope is about 19.6 (21.0), in 20 s (20x20 s). Astrometric two-axes precision (rms) at the bright-end is about 60 (30)\,mas in 20\,s (20x20 s), while absolute photometric calibration, relative to GAIA, provides ~10 millimag accuracy. Relative photometric precision, in a single 20 s (320 s) image, at the bright-end measured over a time scale of about 60 min is about 3 (1) millimag. We discuss the system science goals, data pipelines, and the observatory control system in companion publications., Comment: Submitted to PASP, 15pp

The Large Array Survey Telescope (LAST) is designed to survey the variable and transient sky at high temporal cadence. The array is comprised of 48 F/2.2 telescopes of 27.9cm aperture, coupled to full-frame backside-illuminated cooled CMOS detectors with $3.76$$\mu$m pixels, resulting in a pixel scale of $1.25\mathrm{arcsec}$. A single telescope with a field of view of $7.4\mathrm{deg}^2$ reaches a $5\sigma$ limiting magnitude of $19.6$ in $20$s. LAST 48 telescopes are mounted on 12 independent mounts -- a modular design which allows us to conduct optimized parallel surveys. Here we provide a detailed overview of the LAST survey strategy and its key scientific goals. These include the search for gravitational-wave (GW) electromagnetic counterparts with a system that can cover the uncertainty regions of the next-generation GW detectors in a single exposure, the study of planetary systems around white dwarfs, and the search for near-Earth objects. LAST is currently being commissioned, with full scientific operations expected in mid 2023. This paper is accompanied by two complementary publications in this issue, giving an overview of the system (Ofek et al. 2023a) and of the dedicated data reduction pipeline (Ofek et al. 2023b).

We give an $\tilde O(n^2)$ time algorithm for computing the exact Dynamic Time Warping distance between two strings whose run-length encoding is of size at most $n$. This matches (up to log factors) the known (conditional) lower bound, and should be compared with the previous fastest $O(n^3)$ time exact algorithm and the $\tilde O(n^2)$ time approximation algorithm.

The ability to tune quantum tunneling is key for achieving selectivity in manipulation of individual particles in quantum technology applications. In this work we count electron escape events out of a time-dependent confinement potential, realized as a dynamic quantum dot in a GaAs/AlGaAs heterostructure. A universal scaling relation of the escape probability as a function of potential barrier rise time and depth is established and developed as a method to probe tunneling rates over many orders of magnitude reaching the limit of shallow anharmonic confinement. Crossover to thermally activated transport is used to estimate the single time-energy scale of the universal model. In application to metrological single electron sources, in-situ calibrated control signals greatly extend the accessible dynamical range for probing the quantization mechanism. Validation of the cubic potential approximation sets a foundation for microscopic modeling of quantum tunneling devices in the shallow confinement regime.

We report the levitation of a superconducting lead-tin sphere with 100 micrometer diameter (corresponding to a mass of 5.6 micrograms) in a static magnetic trap formed by two coils in an anti-Helmholtz configuration, with adjustable resonance frequencies up to 240 hertz. The center-of-mass motion of the sphere is monitored magnetically using a dc superconducting quantum interference device as well as optically and exhibits quality factors of up to 2.6e7. We also demonstrate 3D magnetic feedback control of the sphere's motion. The setup is housed in a dilution refrigerator operating at 15 millikelvin. By implementing a cryogenic vibration isolation system we can attenuate environmental vibrations at 200 hertz by approximately seven orders of magnitude. The combination of low temperature, large mass and high quality factor as well as adjustable resonance frequencies provides a promising platform for testing quantum physics in previously unexplored regimes with high mass and long coherence times., Comment: Updated to published version

Recommender systems suffer from the cold-start problem whenever a new user joins the platform or a new item is added to the catalog. To address item cold-start, we propose to replace the embedding layer in sequential recommenders with a dynamic storage that has no learnable weights and can keep an arbitrary number of representations. In this paper, we present FELRec, a large embedding network that refines the existing representations of users and items in a recursive manner, as new information becomes available. In contrast to similar approaches, our model represents new users and items without side information and time-consuming finetuning, instead it runs a single forward pass over a sequence of existing representations. During item cold-start, our method outperforms similar method by 29.50%-47.45%. Further, our proposed model generalizes well to previously unseen datasets in zero-shot settings. The source code is publicly available at https://github.com/kweimann/FELRec .

The non-linear response of a beam splitter to the coincident arrival of interacting particles enables numerous applications in quantum engineering and metrology yet poses considerable challenge to achieve focused interactions on the individual particle level. Here we probe the coincidence correlations at a mesoscopic constriction between individual ballistic electrons in a system with unscreened Coulomb interactions and introduce concepts to quantify the associated parametric non-linearity. The full counting statistics of joint detection allows us to explore the interaction-mediated energy exchange. We observe an increase from 50\% up to 70\% in coincidence counts between statistically indistinguishable on demand sources, and a correlation signature consistent with independent tomography of the electron emission. Analytical modeling and numerical simulations underpin consistency of the experimental results with Coulomb interactions between two electrons counterpropagating in a dispersive quadratic saddle, and demonstrate interactions sufficiently strong, $U/(\hbar \omega) > 10$, to enable single-shot in-flight detection and quantum logic gates.