Search

Research into the detection of human activities from wearable sensors is a highly active field, benefiting numerous applications, from ambulatory monitoring of healthcare patients via fitness coaching to streamlining manual work processes. We present an empirical study that evaluates and contrasts four commonly employed annotation methods in user studies focused on in-the-wild data collection. For both the user-driven, in situ annotations, where participants annotate their activities during the actual recording process, and the recall methods, where participants retrospectively annotate their data at the end of each day, the participants had the flexibility to select their own set of activity classes and corresponding labels. Our study illustrates that different labeling methodologies directly impact the annotations' quality, as well as the capabilities of a deep learning classifier trained with the data. We noticed that in situ methods produce less but more precise labels than recall methods. Furthermore, we combined an activity diary with a visualization tool that enables the participant to inspect and label their activity data. Due to the introduction of such a tool were able to decrease missing annotations and increase the annotation consistency, and therefore the F1-Score of the deep learning model by up to 8% (ranging between 82.1 and 90.4% F1-Score). Furthermore, we discuss the advantages and disadvantages of the methods compared in our study, the biases they could introduce, and the consequences of their usage on human activity recognition studies as well as possible solutions.

Ensuring the safety of the equipment, its environment and most importantly, the operator during robot operations is of paramount importance. Robots and complex robotic systems are appearing in more and more industrial and professional service applications. However, while mechanical components and control systems are advancing rapidly, the legislation background and standards framework for such systems and machinery are lagging behind. As part of a fundamental research work targeting industrial robots and industry 4.0 solutions for completely automated slaughtering, it was revealed that there are no particular standards addressing robotics systems applied to the agrifood domain. More specifically, within the agrifood sector, the only standards existing for the meat industry and the red meat sector are hygienic standards related to machinery. None of the identified standards or regulations consider the safety of autonomous robot operations or human robot collaborations in the abattoirs. The goal of this paper is to provide a general overview of the regulations and standards (and similar guiding documents) relevant for such applications, that could possibly be used as guidelines during the development of inherently safe robotic systems for abattoirs. Reviewing and summarizing the relevant standard and legislation landscape should also offer some instrumental help regarding the foreseen certification procedure of meat processing robots and robot cells for slaughterhouses in the near future.

The Internet of Things (IoT) is rapidly changing the number of connected devices and the way they interact with each other. This increases the need for an automated and secure onboarding procedure for IoT devices, systems and services. Device manufacturers are entering the market with internet connected devices, ranging from small sensors to production devices, which are subject of security threats specific to IoT. The onboarding procedure is required to introduce a new device in a System of Systems (SoS) without compromising the already onboarded devices and the underlying infrastructure. Onboarding is the process of providing access to the network and registering the components for the first time in an IoT/SoS framework, thus creating a chain of trust from the hardware device to its hosted software systems and their provided services. The large number and diversity of device hardware, software systems and running services raises the challenge to establish a generic onboarding procedure. In this paper, we present an automated and secure onboarding procedure for SoS. We have implemented the onboarding procedure in the Eclipse Arrowhead framework. However, it can be easily adapted for other IoT/SoS frameworks that are based on Service-oriented Architecture (SoA) principles. The automated onboarding procedure ensures a secure and trusted communication between the new IoT devices and the Eclipse Arrowhead framework. We show its application in a smart charging use case and perform a security assessment., Comment: IEEE Access

We apply artificial neural networks (ANNs) to nowcast quarterly GDP growth for the U.S. economy. Using the monthly FRED-MD database, we compare the nowcasting performance of five different ANN architectures: the multilayer perceptron (MLP), the one-dimensional convolutional neural network (1D CNN), the Elman recurrent neural network (RNN), the long short-term memory network (LSTM), and the gated recurrent unit (GRU). The empirical analysis presents results from two distinctively different evaluation periods. The first (2012:Q1 -- 2019:Q4) is characterized by balanced economic growth, while the second (2012:Q1 -- 2022:Q4) also includes periods of the COVID-19 recession. According to our results, longer input sequences result in more accurate nowcasts in periods of balanced economic growth. However, this effect ceases above a relatively low threshold value of around six quarters (eighteen months). During periods of economic turbulence (e.g., during the COVID-19 recession), longer input sequences do not help the models' predictive performance; instead, they seem to weaken their generalization capability. Combined results from the two evaluation periods indicate that architectural features enabling long-term memory do not result in more accurate nowcasts. Comparing network architectures, the 1D CNN has proved to be a highly suitable model for GDP nowcasting. The network has shown good nowcasting performance among the competitors during the first evaluation period and achieved the overall best accuracy during the second evaluation period. Consequently, first in the literature, we propose the application of the 1D CNN for economic nowcasting., Comment: arXiv admin note: text overlap with arXiv:2106.08901 by other authors

Research has shown the complementarity of camera- and inertial-based data for modeling human activities, yet datasets with both egocentric video and inertial-based sensor data remain scarce. In this paper, we introduce WEAR, an outdoor sports dataset for both vision- and inertial-based human activity recognition (HAR). Data from 22 participants performing a total of 18 different workout activities was collected with synchronized inertial (acceleration) and camera (egocentric video) data recorded at 11 different outside locations. WEAR provides a challenging prediction scenario in changing outdoor environments using a sensor placement, in line with recent trends in real-world applications. Benchmark results show that through our sensor placement, each modality interestingly offers complementary strengths and weaknesses in their prediction performance. Further, in light of the recent success of single-stage Temporal Action Localization (TAL) models, we demonstrate their versatility of not only being trained using visual data, but also using raw inertial data and being capable to fuse both modalities by means of simple concatenation. The dataset and code to reproduce experiments is publicly available via: mariusbock.github.io/wear/., Comment: accepted at IMWUT; 21 pages, 8 figures, 2 tables

We theoretically discuss electronic transport via Majorana states in magnetic topological insulator-superconductor junctions with an asymmetric split of the applied bias voltage. We study normal-superconductor-normal (NSN) junctions made of narrow (wire-like) or wide (film-like) magnetic topological insulator slabs with a central proximitized superconducting sector. The occurrence of charge non-conserving Andreev processes entails a nonzero conductance related to an electric current flowing to ground from the proximitized sector of the NSN junction. We show that topologically-protected Majorana modes require an antisymmetry of this conductance with respect to the point of equally split bias voltage across the junction.

We report the first direct observation of neutrino interactions at a particle collider experiment. Neutrino candidate events are identified in a 13.6 TeV center-of-mass energy $pp$ collision data set of 35.4 fb${}^{-1}$ using the active electronic components of the FASER detector at the Large Hadron Collider. The candidates are required to have a track propagating through the entire length of the FASER detector and be consistent with a muon neutrino charged-current interaction. We infer $153^{+12}_{-13}$ neutrino interactions with a significance of 16 standard deviations above the background-only hypothesis. These events are consistent with the characteristics expected from neutrino interactions in terms of secondary particle production and spatial distribution, and they imply the observation of both neutrinos and anti-neutrinos with an incident neutrino energy of significantly above 200 GeV., Comment: Submitted to PRL on March 24 2023

The GIREP community on teaching and learning quantum physics and the Education section of the Quantum flagship project of the European Union (QTEdu) have brought together different stakeholders in the field of teaching quantum physics on all levels, including outreach. The goal of QTEdu is to pave the way for the training of the future quantum workforce. To this end, it is necessary to understand the needs of the quantum technology (QT) field, make the general public aware of the existence and importance of QT, and introduce quantum physics already in high school, so that high school students can choose QT as their field of study and career. Finally, new university courses need to be established to support emerging specific profiles such as a quantum engineer. In this symposium, four QTEdu pilot projects were brought together to demonstrate how their complementary approaches have worked towards realising the above goals., Comment: 14 pages, 4 figures

Motivated by the algorithmic study of 3-dimensional manifolds, we explore the structural relationship between the JSJ decomposition of a given 3-manifold and its triangulations. Building on work of Bachman, Derby-Talbot and Sedgwick, we show that a "sufficiently complicated" JSJ decomposition of a 3-manifold enforces a "complicated structure" for all of its triangulations. More concretely, we show that, under certain conditions, the treewidth (resp. pathwidth) of the graph that captures the incidences between the pieces of the JSJ decomposition of an irreducible, closed, orientable 3-manifold M yields a linear lower bound on its treewidth tw(M) (resp. pathwidth pw(M)), defined as the smallest treewidth (resp. pathwidth) of the dual graph of any triangulation of M. We present several applications of this result. We give the first example of an infinite family of bounded-treewidth 3-manifolds with unbounded pathwidth. We construct Haken 3-manifolds with arbitrarily large treewidth; previously the existence of such 3-manifolds was only known in the non-Haken case. We also show that the problem of providing a constant-factor approximation for the treewidth (resp. pathwidth) of bounded-degree graphs efficiently reduces to computing a constant-factor approximation for the treewidth (resp. pathwidth) of 3-manifolds., Comment: 22 pages, 19 figures. The appendix has been extended

Human-Robot collaboration in home and industrial workspaces is on the rise. However, the communication between robots and humans is a bottleneck. Although people use a combination of different types of gestures to complement speech, only a few robotic systems utilize gestures for communication. In this paper, we propose a gesture pseudo-language and show how multiple types of gestures can be combined to express human intent to a robot (i.e., expressing both the desired action and its parameters - e.g., pointing to an object and showing that the object should be emptied into a bowl). The demonstrated gestures and the perceived table-top scene (object poses detected by CosyPose) are processed in real-time) to extract the human's intent. We utilize behavior trees to generate reactive robot behavior that handles various possible states of the world (e.g., a drawer has to be opened before an object is placed into it) and recovers from errors (e.g., when the scene changes). Furthermore, our system enables switching between direct teleoperation of the end-effector and high-level operation using the proposed gesture sentences. The system is evaluated on increasingly complex tasks using a real 7-DoF Franka Emika Panda manipulator. Controlling the robot via action gestures lowered the execution time by up to 60%, compared to direct teleoperation., Comment: 7 pages, 9 figures

A Round Robin study has been carried out to estimate the impact of the human element in small-angle scattering data analysis. Four corrected datasets were provided to participants ready for analysis. All datasets were measured on samples containing spherical scatterers, with two datasets in dilute dispersions, and two from powders. Most of the 46 participants correctly identified the number of populations in the dilute dispersions, with half of the population mean entries within 1.5% and half of the population width entries within 40%, respectively. Due to the added complexity of the structure factor, much fewer people submitted answers on the powder datasets. For those that did, half of the entries for the means and widths were within 44% and 86% respectively. This Round Robin experiment highlights several causes for the discrepancies, for which solutions are proposed., Comment: 23 pages, 10 figures. For the original information sent to RR participants, see https://zenodo.org/record/7506365 . For the anonymized results and Jupyter notebook for analysis, see https://zenodo.org/record/7509710

Fast and accurate climate simulations and weather predictions are critical for understanding and preparing for the impact of climate change. Real-world weather and climate modeling consist of complex compound stencil kernels that do not perform well on conventional architectures. Horizontal diffusion is one such important compound stencil found in many climate and weather prediction models. Recent works propose using FPGAs as an alternative to traditional CPU and GPU-based systems to accelerate compound stencil kernels. However, we observe that compound stencil computations cannot leverage the bit-level flexibility available on an FPGA because of its complex memory access patterns, leading to high hardware resource utilization and low peak performance. We introduce SPARTA, a novel spatial accelerator for horizontal diffusion weather stencil computation. We exploit the two-dimensional spatial architecture to efficiently accelerate horizontal diffusion stencil by designing the first scaled-out spatial accelerator using MLIR (Multi-Level Intermediate Representation) compiler framework. We evaluate its performance on a real cutting-edge AMD-Xilinx Versal AI Engine spatial architecture. Our real-system evaluation results demonstrate that SPARTA outperforms the state-of-the-art CPU, GPU, and FPGA implementations by 17.1x, 1.2x, and 2.1x, respectively. Our results reveal that balancing workload across the available processing resources is crucial in achieving high performance on spatial architectures. We also implement and evaluate five elementary stencils that are commonly used as benchmarks for stencil computation research. We freely open-source all our implementations to aid future research in stencil computation and spatial computing systems at https://github.com/CMU-SAFARI/SPARTA.

We present low-temperature magnetotransport measurements on selectively-grown Sb$_2$Te$_3$-based topological insulator ring structures. These topological insulator ring geometries display clear Aharonov-Bohm oscillations in the conductance originating from phase-coherent transport around the ring. The temperature dependence of the oscillation amplitude indicates that the Aharonov-Bohm oscillations originate from ballistic transport along the ring arms. The oscillations can therefore be attributed to topological surface states, which can maintain a quasi-ballistic transport regime in the presence of disorder. Further insight on the phase coherence is gained by comparing with similar Aharonov-Bohm-type oscillations in topological insulator nanoribbons exposed to an axial magnetic field. Here, quasi-ballistic phase-coherent transport is confirmed for closed-loop topological surface states in transverse direction enclosing the cross-section of the nanoribbon. In contrast, the appearance of universal conductance fluctuations indicates phase-coherent transport in the diffuse regime, which is attributed to bulk carrier transport. Thus, it appears that even in the presence of diffusive $p$-type charge carriers in Aharonov-Bohm ring structures, phase-coherent quasi-ballistic transport of topologically protected surface states is maintained over long distances., Comment: 11 pages (including Supplementary Material), 5 figures, 4 supplementary figures

We review the surface flux transport model for the evolution of magnetic flux patterns on the Sun's surface. Our underlying motivation is to understand the model's prediction of the polar field (or axial dipole) strength at the end of the solar cycle. The main focus is on the "classical" model: namely, steady axisymmetric profiles for differential rotation and meridional flow, and uniform supergranular diffusion. Nevertheless, the review concentrates on recent advances, notably in understanding the roles of transport parameters and - in particular - the source term. We also discuss the physical justification for the surface flux transport model, along with efforts to incorporate radial diffusion, and conclude by summarizing the main directions where researchers have moved beyond the classical model., Comment: 35 pages, 11 figures, accepted for publication in Space Science Reviews

Dissolution testing is part of the target product quality that is essential in approving new products in the pharmaceutical industry. The prediction of the dissolution profile based on spectroscopic data is an alternative to the current destructive and time-consuming method. Raman and near-infrared (NIR) spectroscopies are two fast and complementary methods that provide information on the tablets' physical and chemical properties and can help predict their dissolution profiles. This work aims to compare the information collected by these spectroscopy methods to support the decision of which measurements should be used so that the accuracy requirement of the industry is met. Artificial neural network models were created, in which the spectroscopy data and the measured compression curves were used as an input individually and in different combinations in order to estimate the dissolution profiles. Results showed that using only the NIR transmission method along with the compression force data or the Raman and NIR reflection methods, the dissolution profile was estimated within the acceptance limits of the f2 similarity factor. Adding further spectroscopy measurements increased the prediction accuracy., Comment: 11 pagee, 5 figures, 3rd International Conference on Data Science and Machine Learning (DSML 2022)

Wearable pupil detection systems often separate the analysis of the captured wearer's eye images for wirelessly-tethered back-end systems. We argue in this paper that investigating hardware-software co-designs would bring along opportunities to make such systems smaller and more efficient. We introduce an open-source embedded system for wearable, non-invasive pupil detection in real-time, on the wearable, embedded platform itself. Our system consists of a head-mounted eye tracker prototype, which combines two miniature camera systems with Raspberry Pi-based embedded system. Apart from the hardware design, we also contribute a pupil detection pipeline that operates using edge analysis, natively on the embedded system at 30fps and run-time of 54ms at 480x640 and 23ms at 240x320. Average cumulative error of 5.3368px is found on the LPW dataset for a detection rate of 51.9\% with our detection pipeline. For evaluation on our hardware-specific camera frames, we also contribute a dataset of 35000 images, from 20 participants.

In inverse optimization problems, the goal is to modify the costs in an underlying optimization problem in such a way that a given solution becomes optimal, while the difference between the new and the original cost functions, called the deviation vector, is minimized with respect to some objective function. The $\ell_1$- and $\ell_\infty$-norms are standard objectives used to measure the size of the deviation. Minimizing the $\ell_1$-norm is a natural way of keeping the total change of the cost function low, while the $\ell_\infty$-norm achieves the same goal coordinate-wise. Nevertheless, none of these objectives is suitable to provide a balanced or fair change of the costs. In this paper, we initiate the study of a new objective that measures the difference between the largest and the smallest weighted coordinates of the deviation vector, called the weighted span. We give a min-max characterization for the minimum weighted span of a feasible deviation vector, and provide a Newton-type algorithm for finding one that runs in strongly polynomial time in the case of unit weights., Comment: 47 pages, 2 figures, 1 table

In minimum-cost inverse optimization problems, we are given a feasible solution to an underlying optimization problem together with a linear cost function, and the goal is to modify the costs by a small deviation vector so that the input solution becomes optimal. The difference between the new and the original cost functions can be measured in several ways. In this paper, we focus on two objectives: the weighted bottleneck Hamming distance and the weighted $\ell_\infty$-norm. We consider a general model in which the coordinates of the deviation vector are required to fall within given lower and upper bounds. For the weighted bottleneck Hamming distance objective, we present a simple, purely combinatorial algorithm that determines an optimal deviation vector in strongly polynomial time. For the weighted $\ell_\infty$-norm objective, we give a min-max characterization for the optimal solution, and provide a pseudo-polynomial algorithm for finding an optimal deviation vector that runs in strongly polynomial time in the case of unit weights. For both objectives, we assume that an algorithm with the same time complexity for solving the underlying combinatorial optimization problem is available. For both objectives, we also show how to extend the results to inverse optimization problems with multiple cost functions., Comment: 18 pages

Magnetic topological insulators in the quantum anomalous Hall regime host ballistic chiral edge channels. When proximitized by an $s$-wave superconductor, these edge states offer the potential for realizing topological superconductivity and Majorana bound states without the detrimental effect of large externally-applied magnetic fields on superconductivity. Realizing well-separated unpaired Majorana bound states requires magnetic topological insulator ribbons with a width of the order of the transverse extent of the edge state, however, which is expected to bring the required ribbon width down to around 100 nm. In this regime, it is known to be extremely difficult to retain the ballistic nature of chiral edge channels and realize a quantized Hall conductance. In this paper, we study the impact of disorder in such magnetic topological insulator nanoribbons and compare the fragility of ballistic chiral edge channels with the stability of Majorana bound states when the ribbon is covered by a superconducting film. We find that the Majorana bound states exhibit greater robustness against disorder than the underlying chiral edge channels., Comment: 6 pages, 5 figures

The problem of covering the ground set of two matroids by a minimum number of common independent sets is notoriously hard even in very restricted settings, i.e.\ when the goal is to decide if two common independent sets suffice or not. Nevertheless, as the problem generalizes several long-standing open questions, identifying tractable cases is of particular interest. Strongly base orderable matroids form a class for which a basis-exchange condition that is much stronger than the standard axiom is met. As a result, several problems that are open for arbitrary matroids can be solved for this class. In particular, Davies and McDiarmid showed that if both matroids are strongly base orderable, then the covering number of their intersection coincides with the maximum of their covering numbers. Motivated by their result, we propose relaxations of strongly base orderability in two directions. First we weaken the basis-exchange condition, which leads to the definition of a new, complete class of matroids with distinguished algorithmic properties. Second, we introduce the notion of covering the circuits of a matroid by a graph, and consider the cases when the graph is ought to be 2-regular or a path. We give an extensive list of results explaining how the proposed relaxations compare to existing conjectures and theorems on coverings by common independent sets., Comment: 21 pages, 3 figures

The phenomenon of $B^0_q$-$\bar{B}^0_q$ mixing ($q=d,s$) provides a sensitive probe for physics beyond the Standard Model. We have a careful look at the determination of the Unitarity Triangle apex, which is needed for the Standard Model predictions of the $B_q$ mixing parameters, and explore how much space for New Physics is left through the current data. We study the impact of tensions between inclusive and exclusive determinations of the CKM matrix elements $|V_{ub}|$ and $|V_{cb}|$, and focus on the $\gamma$ angle extraction. We present various future scenarios and discuss the application of these results for leptonic rare $B$ decays, which allows us to minimise the CKM parameter impact in the New Physics searches. Performing future projections, we explore and illustrate the impact of increased precision on key input quantities. It will be exciting to see how more precise data in the future high-precision era of flavour physics can lead to a much sharper picture., Comment: 8 pages, 2 figures, Contribution to the Proceedings of the 8th Symposium on Prospects in the Physics of Discrete Symmetries (DISCRETE 2022), 7-11 November 2022, Baden-Baden, Germany

Collaborative robots became a popular tool for increasing productivity in partly automated manufacturing plants. Intuitive robot teaching methods are required to quickly and flexibly adapt the robot programs to new tasks. Gestures have an essential role in human communication. However, in human-robot-interaction scenarios, gesture-based user interfaces are so far used rarely, and if they employ a one-to-one mapping of gestures to robot control variables. In this paper, we propose a method that infers the user's intent based on gesture episodes, the context of the situation, and common sense. The approach is evaluated in a simulated table-top manipulation setting. We conduct deterministic experiments with simulated users and show that the system can even handle personal preferences of each user., Comment: 7 pages, 8 figures, accepted for ICRA 2023

Hypergraph Horn functions were introduced as a subclass of Horn functions that can be represented by a collection of circular implication rules. These functions possess distinguished structural and computational properties. In particular, their characterizations in terms of implicate-duality and the closure operator provide extensions of matroid duality and the Mac Lane-Steinitz exchange property of matroid closure, respectively. In the present paper, we introduce a subclass of hypergraph Horn functions that we call matroid Horn functions. We provide multiple characterizations of matroid Horn functions in terms of their canonical and complete CNF representations. We also study the Boolean minimization problem for this class, where the goal is to find a minimum size representation of a matroid Horn function given by a CNF representation. While there are various ways to measure the size of a CNF, we focus on the number of circuits and circuit clauses. We determine the size of an optimal representation for binary matroids, and give lower and upper bounds in the uniform case. For uniform matroids, we show a strong connection between our problem and Tur\'an systems that might be of independent combinatorial interest., Comment: 19 pages, 2 figures. arXiv admin note: text overlap with arXiv:2301.05461

We construct an explicit map from a generic minimal $\delta(2)$-ideal Lagrangian submanifold of $\mathbb{C}^n$ to the quaternionic projective space $\mathbb{H}P^{n-1}$, whose image is either a point or a minimal totally complex surface. A stronger result is obtained for $n=3$, since the above mentioned map then provides a one-to-one correspondence between minimal $\delta(2)$-ideal Lagrangian submanifolds of $\mathbb{C}^3$ and minimal totally complex surfaces in $\mathbb{H}P^2$ which are moreover anti-symmetric. Finally, we also show that there is a one-to-one correspondence between such surfaces in $\mathbb{H}P^2$ and minimal Lagrangian surfaces in $\mathbb{C}P^2$., Comment: 21 pages

Horn functions form a subclass of Boolean functions possessing interesting structural and computational properties. These functions play a fundamental role in algebra, artificial intelligence, combinatorics, computer science, database theory, and logic. In the present paper, we introduce the subclass of hypergraph Horn functions that generalizes matroids and equivalence relations. We provide multiple characterizations of hypergraph Horn functions in terms of implicate-duality and the closure operator, which are respectively regarded as generalizations of matroid duality and Mac Lane-Steinitz exchange property of matroid closure. We also study algorithmic issues on hypergraph Horn functions, and show that the recognition problem (i.e., deciding if a given definite Horn CNF represents a hypergraph Horn function) and key realization (i.e., deciding if a given hypergraph is realized as a key set by a hypergraph Horn function) can be done in polynomial time, while implicate sets can be generated with polynomial delay., Comment: 19 pages

Dense matrix multiply (MM) serves as one of the most heavily used kernels in deep learning applications. To cope with the high computation demands of these applications, heterogeneous architectures featuring both FPGA and dedicated ASIC accelerators have emerged as promising platforms. For example, the AMD/Xilinx Versal ACAP architecture combines general-purpose CPU cores and programmable logic with AI Engine processors optimized for AI/ML. With 400 AIEs, it provides up to 6.4 TFLOPs performance for 32-bit floating-point data. However, machine learning models often contain both large and small MM operations. While large MM operations can be parallelized efficiently across many cores, small MM operations typically cannot. We observe that executing some small MM layers from the BERT natural language processing model on a large, monolithic MM accelerator in Versal ACAP achieved less than 5% of the theoretical peak performance. Therefore, one key question arises: How can we design accelerators to fully use the abundant computation resources under limited communication bandwidth for applications with multiple MM layers of diverse sizes? We identify the biggest system throughput bottleneck resulting from the mismatch of massive computation resources of one monolithic accelerator and the various MM layers of small sizes in the application. To resolve this problem, we propose the CHARM framework to compose multiple diverse MM accelerator architectures working concurrently towards different layers in one application. We deploy the CHARM framework for four different applications, including BERT, ViT, NCF, MLP, on the AMD Versal ACAP VCK190 evaluation board. Our experiments show that we achieve 1.46 TFLOPs, 1.61 TFLOPs, 1.74 TFLOPs, and 2.94 TFLOPs inference throughput for BERT, ViT, NCF and MLP, which obtain 5.40x, 32.51x, 1.00x and 1.00x throughput gains compared to one monolithic accelerator.

A combinatorial market consists of a set of indivisible items and a set of agents, where each agent has a valuation function that specifies for each subset of items its value for the given agent. From an optimization point of view, the goal is usually to determine a pair of pricing and allocation of the items that provides an efficient distribution of the resources, i.e., maximizes the social welfare, or is as profitable as possible for the seller, i.e., maximizes the revenue. To overcome the weaknesses of mechanisms operating with static prices, a recent line of research has concentrated on dynamic pricing schemes. In this model, agents arrive in an unspecified sequential order, and the prices can be updated between two agent-arrivals. Though the dynamic setting is capable of maximizing social welfare in various scenarios, the assumption that the agents arrive one after the other eliminates the standard concept of fairness. In this paper, we study the existence of optimal dynamic prices under fairness constraints in unit-demand markets. We propose four possible notions of envy-freeness of different strength depending on the time period over which agents compare themselves to others: the entire time horizon, only the past, only the future, or only the present. For social welfare maximization, while the first definition leads to Walrasian equilibria, we give polynomial-time algorithms that always find envy-free optimal dynamic prices in the remaining three cases. In contrast, for revenue maximization, we show that the corresponding problems are APX-hard if the ordering of the agents is fixed. On the positive side, we give polynomial-time algorithms for the setting when the seller can choose the order in which agents arrive., Comment: 19 pages, 3 figures

Topological insulator (TI) nanoribbons with proximity-induced superconductivity are a promising platform for Majorana bound states (MBSs). In this work, we consider a detailed modeling approach for a TI nanoribbon in contact with a superconductor via its top surface, which induces a superconducting gap in its surface-state spectrum. The system displays a rich phase diagram with different numbers of end-localized MBSs as a function of chemical potential and magnetic flux piercing the cross section of the ribbon. These MBSs can be robust or fragile upon consideration of electrostatic disorder. We simulate a tunneling spectroscopy setup to probe the different topological phases of top-proximitized TI nanoribbons. Our simulation results indicate that a top-proximitized TI nanoribbon is ideally suited for realizing fully gapped topological superconductivity, in particular when the Fermi level is pinned near the Dirac point. In this regime, the setup yields a single pair of MBSs, well separated at opposite ends of the proximitized ribbon, which gives rise to a robust quantized zero-bias conductance peak., Comment: 12 pages, 5 figures

Detecting persons in images or video with neural networks is a well-studied subject in literature. However, such works usually assume the availability of a camera of decent resolution and a high-performance processor or GPU to run the detection algorithm, which significantly increases the cost of a complete detection system. However, many applications require low-cost solutions, composed of cheap sensors and simple microcontrollers. In this paper, we demonstrate that even on such hardware we are not condemned to simple classic image processing techniques. We propose a novel ultra-lightweight CNN-based person detector that processes thermal video from a low-cost 32x24 pixel static imager. Trained and compressed on our own recorded dataset, our model achieves up to 91.62% accuracy (F1-score), has less than 10k parameters, and runs as fast as 87ms and 46ms on low-cost microcontrollers STM32F407 and STM32F746, respectively.

SchNetPack is a versatile neural networks toolbox that addresses both the requirements of method development and application of atomistic machine learning. Version 2.0 comes with an improved data pipeline, modules for equivariant neural networks as well as a PyTorch implementation of molecular dynamics. An optional integration with PyTorch Lightning and the Hydra configuration framework powers a flexible command-line interface. This makes SchNetPack 2.0 easily extendable with custom code and ready for complex training task such as generation of 3d molecular structures.