Search

Background Advanced adaptive randomised clinical trials are increasingly used. Compared to their conventional counterparts, their flexibility may make them more efficient, increase the probability of obtaining conclusive results without larger samples than necessary, and increase the probability that individual participants are allocated to more promising interventions. However, limited guidance is available on designing and evaluating the performance of advanced adaptive trials. Methods We summarise the methodological considerations and provide practical guidance on the entire workflow of planning and evaluating advanced adaptive trials using adaptive stopping, adaptive arm dropping, and response-adaptive randomisation within a Bayesian statistical framework. Results This comprehensive practical guide covers the key methodological decisions for advanced adaptive trials and their specification and evaluation using statistical simulation. These considerations include interventions and common control use; outcome type and generation; analysis timing and outcome-data lag; allocation rules; analysis model; adaptation rules for stopping and arm dropping; clinical scenarios assessed; performance metrics; calibration; sensitivity analyses; and reporting. The considerations are covered in the context of realistic examples, along with simulation code using the adaptr R package. Conclusions This practical guide will help clinical trialists, methodologists, and biostatisticians design and evaluate advanced adaptive trials., Comment: 63 pages (30 without appendices), 5 figures (3 without appendices), 3 tables

Machine learning (ML) systems have become vital in the mobile gaming industry. Companies like King have been using them in production to optimize various parts of the gaming experience. One important area is in-app purchases: purchases made in the game by players in order to enhance and customize their gameplay experience. In this work we describe how we developed an ML system in order to predict when a player is expected to make their next in-app purchase. These predictions are used to present offers to players. We briefly describe the problem definition, modeling approach and results and then, in considerable detail, outline the end-to-end ML system. We conclude with a reflection on challenges encountered and plans for future work.

Influenced by the complexity of volumetric imaging, there is a shortage of established datasets useful for benchmarking volumetric deep-learning models. As a consequence, new and existing models are not easily comparable, limiting the development of architectures optimized specifically for volumetric data. To counteract this trend, we introduce MozzaVID - a large, clean, and versatile volumetric classification dataset. Our dataset contains X-ray computed tomography (CT) images of mozzarella microstructure and enables the classification of 25 cheese types and 149 cheese samples. We provide data in three different resolutions, resulting in three dataset instances containing from 591 to 37,824 images. While being general-purpose, the dataset also facilitates investigating mozzarella structure properties. The structure of food directly affects its functional properties and thus its consumption experience. Understanding food structure helps tune the production and mimicking it enables sustainable alternatives to animal-derived food products. The complex and disordered nature of food structures brings a unique challenge, where a choice of appropriate imaging method, scale, and sample size is not trivial. With this dataset we aim to address these complexities, contributing to more robust structural analysis models. The dataset can be downloaded from: https://archive.compute.dtu.dk/files/public/projects/MozzaVID/.

Distributed data analysis is a large and growing field driven by a massive proliferation of user devices, and by privacy concerns surrounding the centralised storage of data. We consider two \emph{adaptive} algorithms for estimating one quantile (e.g.~the median) when each user holds a single data point lying in a domain $[B]$ that can be queried once through a private mechanism; one under local differential privacy (LDP) and another for shuffle differential privacy (shuffle-DP). In the adaptive setting we present an $\varepsilon$-LDP algorithm which can estimate any quantile within error $\alpha$ only requiring $O(\frac{\log B}{\varepsilon^2\alpha^2})$ users, and an $(\varepsilon,\delta)$-shuffle DP algorithm requiring only $\widetilde{O}((\frac{1}{\varepsilon^2}+\frac{1}{\alpha^2})\log B)$ users. Prior (nonadaptive) algorithms require more users by several logarithmic factors in $B$. We further provide a matching lower bound for adaptive protocols, showing that our LDP algorithm is optimal in the low-$\varepsilon$ regime. Additionally, we establish lower bounds against non-adaptive protocols which paired with our understanding of the adaptive case, proves a fundamental separation between these models.

The field of algorithms with predictions incorporates machine learning advice in the design of online algorithms to improve real-world performance. While this theoretical framework often assumes uniform reliability across all predictions, modern machine learning models can now provide instance-level uncertainty estimates. In this paper, we propose calibration as a principled and practical tool to bridge this gap, demonstrating the benefits of calibrated advice through two case studies: the ski rental and online job scheduling problems. For ski rental, we design an algorithm that achieves optimal prediction-dependent performance and prove that, in high-variance settings, calibrated advice offers more effective guidance than alternative methods for uncertainty quantification. For job scheduling, we demonstrate that using a calibrated predictor leads to significant performance improvements over existing methods. Evaluations on real-world data validate our theoretical findings, highlighting the practical impact of calibration for algorithms with predictions., Comment: Replacement: updated acknowledgments

Parameter shift rules (PSRs) are key techniques for efficient gradient estimation in variational quantum eigensolvers (VQEs). In this paper, we propose its Bayesian variant, where Gaussian processes with appropriate kernels are used to estimate the gradient of the VQE objective. Our Bayesian PSR offers flexible gradient estimation from observations at arbitrary locations with uncertainty information and reduces to the generalized PSR in special cases. In stochastic gradient descent (SGD), the flexibility of Bayesian PSR allows the reuse of observations in previous steps, which accelerates the optimization process. Furthermore, the accessibility to the posterior uncertainty, along with our proposed notion of gradient confident region (GradCoRe), enables us to minimize the observation costs in each SGD step. Our numerical experiments show that the VQE optimization with Bayesian PSR and GradCoRe significantly accelerates SGD and outperforms the state-of-the-art methods, including sequential minimal optimization., Comment: 8 pages, 5 figures. arXiv admin note: text overlap with arXiv:2502.01704

In this work, we demonstrate how extended Lagrangian Born-Oppenheimer quantum-based molecular dynamics (XL-BOMD) can be used to simulate heterogeneous electrocatalytic reactions. In particular, we apply our framework to study the oxygen reduction reaction (ORR) mechanism on nitrogen-doped graphene in an aqueous solution. The electronic ground state and total energy of XL-BOMD are stabilized through nuclear and electronic equations of motion assisted by an integral kernel updated with low-rank approximations. A species charge analysis reveals that the XL-BOMD simulations can be used to detect electron transfer between the catalyst surface and the solvated molecular oxygen mediated by water molecules, resulting in the molecular dissociation of O$_2$. Simulations of the ORR under high electrochemical biases contribute to a larger body of work elucidating an outer-sphere ORR mechanism in which reduction occurs through water networks without direct adsorption of molecular oxygen.

We present an explicit solution to the discrete-time Bellman equation for minimax optimal control of positive systems under unconstrained disturbances. The primary contribution of our result relies on deducing a bound for the disturbance penalty, which characterizes the existence of a finite solution to the problem class. Moreover, this constraint on the disturbance penalty reveals that, in scenarios where a solution is feasible, the problem converges to its equivalent minimization problem in the absence of disturbances., Comment: Presented at the 26th International Symposium on Mathematical Theory of Networks and Systems (Cambridge, UK)

The objective to be minimized in the variational quantum eigensolver (VQE) has a restricted form, which allows a specialized sequential minimal optimization (SMO) that requires only a few observations in each iteration. However, the SMO iteration is still costly due to the observation noise -- one observation at a point typically requires averaging over hundreds to thousands of repeated quantum measurement shots for achieving a reasonable noise level. In this paper, we propose an adaptive cost control method, named subspace in confident region (SubsCoRe), for SMO. SubsCoRe uses the Gaussian process (GP) surrogate, and requires it to have low uncertainty over the subspace being updated, so that optimization in each iteration is performed with guaranteed accuracy. The adaptive cost control is performed by first setting the required accuracy according to the progress of the optimization, and then choosing the minimum number of measurement shots and their distribution such that the required accuracy is satisfied. We demonstrate that SubsCoRe significantly improves the efficiency of SMO, and outperforms the state-of-the-art methods., Comment: 9 pages, 6 figures, 41st International Conference on Machine Learning (ICML 2024)

Despite exceptional achievements, training neural networks remains computationally expensive and is often plagued by instabilities that can degrade convergence. While learning rate schedules can help mitigate these issues, finding optimal schedules is time-consuming and resource-intensive. This work explores theoretical issues concerning training stability in the constant-learning-rate (i.e., without schedule) and small-batch-size regime. Surprisingly, we show that the order of gradient updates affects stability and convergence in gradient-based optimizers. We illustrate this new line of thinking using backward-SGD, which processes batch gradient updates like SGD but in reverse order. Our theoretical analysis shows that in contractive regions (e.g., around minima) backward-SGD converges to a point while the standard forward-SGD generally only converges to a distribution. This leads to improved stability and convergence which we demonstrate experimentally. While full backward-SGD is computationally intensive in practice, it highlights opportunities to exploit reverse training dynamics (or more generally alternate iteration orders) to improve training. To our knowledge, this represents a new and unexplored avenue in deep learning optimization.

Pathology detection in Wireless Capsule Endoscopy (WCE) using deep learning has been explored in the recent past. However, deep learning models can be influenced by the color quality of the dataset used to train them, impacting detection, segmentation and classification tasks. In this work, we evaluate the impact of color correction on pathology detection using two prominent object detection models: Retinanet and YOLOv5. We first generate two color corrected versions of a popular WCE dataset (i.e., SEE-AI dataset) using two different color correction functions. We then evaluate the performance of the Retinanet and YOLOv5 on the original and color corrected versions of the dataset. The results reveal that color correction makes the models generate larger bounding boxes and larger intersection areas with the ground truth annotations. Furthermore, color correction leads to an increased number of false positives for certain pathologies. However, these effects do not translate into a consistent improvement in performance metrics such as F1-scores, IoU, and AP50. The code is available at https://github.com/agossouema2011/WCE2024. Keywords: Wireless Capsule Endoscopy, Color correction, Retinanet, YOLOv5, Detection

We present an all-digital programmable machine learning accelerator chip for image classification, underpinning on the Tsetlin machine (TM) principles. The TM is a machine learning algorithm founded on propositional logic, utilizing sub-pattern recognition expressions called clauses. The accelerator implements the coalesced TM version with convolution, and classifies booleanized images of 28$\times$28 pixels with 10 categories. A configuration with 128 clauses is used in a highly parallel architecture. Fast clause evaluation is obtained by keeping all clause weights and Tsetlin automata (TA) action signals in registers. The chip is implemented in a 65 nm low-leakage CMOS technology, and occupies an active area of 2.7mm$^2$. At a clock frequency of 27.8 MHz, the accelerator achieves 60.3k classifications per second, and consumes 8.6 nJ per classification. The latency for classifying a single image is 25.4 $\mu$s which includes system timing overhead. The accelerator achieves 97.42%, 84.54% and 82.55% test accuracies for the datasets MNIST, Fashion-MNIST and Kuzushiji-MNIST, respectively, matching the TM software models., Comment: 10 pages, 6 figures. This work has been submitted to the IEEE for possible publication

To investigate spectra of kilonovae in the NLTE phase (t >=1week), we perform atomic calculations for dielectronic (DR) and radiative (RR) recombination rates for the light r-process elements: Se (Z = 34), Rb (Z = 37), Sr (Z = 38), Y (Z = 39), and Zr (Z = 40) using the HULLAC code. For the different elements, our results for the total rate coefficients for recombining from the ionization states of II to I, III to II, and IV to III vary between 10^{-13}-10^{-9} cm^3/s, 10^{-12}-10^{-10} cm^3/s, and 10^{-13}-10^{-10} cm^3/s, respectively, at a temperature of T = 10,000 K. We also provide fits to the ground state photoionization cross sections of the various ions, finding larger and more slowly declining values with energy in comparison to the hydrogenic approximation. Using this new atomic data, we study the impact on kilonova model spectra at phases of t = 10 days and t = 25 days using the spectral synthesis code SUMO. Compared to models using the previous treatment of recombination as a constant rate, the new models show significant changes in ionization and temperature, and correspondingly, in emergent spectra. With the new rates, we find that Zr (Z = 40) plays a yet more dominant role in kilonova spectra for light r-process compositions. Further, we show that previously predicted mid-infrared (e.g. [Se III] 4.55 \mum) and optical (e.g. [Rb I] 7802, 7949 A) lines disappear in the new model. Instead a strong [Se I] line is seen to be emerging at \lambda=5.03 \mum. These results demonstrate the importance of considering the detailed microphysics for modelling and interpreting the late-time kilonova spectra., Comment: Submitted to ApJ

A well-known conjecture by Harris states that any triangle-free $d$-degenerate graph has fractional chromatic number at most $O\left(\frac{d}{\ln d}\right)$. This conjecture has gained much attention in recent years, and is known to have many interesting implications, including a conjecture by Esperet, Kang and Thomass\'e that any triangle-free graph with minimum degree $d$ contains a bipartite induced subgraph of minimum degree $\Omega(\log d)$. Despite this attention, Harris' conjecture has remained wide open with no known improvement on the trivial upper bound, until now. In this article, we give an elegant proof of Harris' conjecture. In particular, we show that any triangle-free $d$-degenerate graph has fractional chromatic number at most $(4+o(1))\frac{d}{\ln d}.$ The conjecture of Esperet et al. follows as a direct consequence. We also prove a more general result, showing that for any triangle-free graph $G$, there exists a random independent set in which each vertex $v$ is included with probability $\Omega(p(v))$, where $p:V(G)\rightarrow [0,1]$ is any function that satisfies a natural condition., Comment: 7 pages. Comments are welcome!

Training sequential recommenders such as SASRec with uniform sample weights achieves good overall performance but can fall short on specific user groups. One such example is popularity bias, where mainstream users receive better recommendations than niche content viewers. To improve recommendation quality across diverse user groups, we explore three Distributionally Robust Optimization(DRO) methods: Group DRO, Streaming DRO, and Conditional Value at Risk (CVaR) DRO. While Group and Streaming DRO rely on group annotations and struggle with users belonging to multiple groups, CVaR does not require such annotations and can naturally handle overlapping groups. In experiments on two real-world datasets, we show that the DRO methods outperform standard training, with CVaR delivering the best results. Additionally, we find that Group and Streaming DRO are sensitive to the choice of group used for loss computation. Our contributions include (i) a novel application of CVaR to recommenders, (ii) showing that the DRO methods improve group metrics as well as overall performance, and (iii) demonstrating CVaR's effectiveness in the practical scenario of intersecting user groups., Comment: This paper has been accepted to the IR for Good track at ECIR 2025

Spin engineering of advanced pulse sequences has had a transformative impact on the development of nuclear magnetic resonance (NMR) spectroscopy, to an extending degree also electron paramagnetic resonance (EPR), and the hybrid between the two, dynamic nuclear polarization (DNP). Based on a simple formalism, we demonstrate that (i) single-crystal static-sample optimisations may tremendously ease design of experiments for rotating powders and (ii) pulse sequences may readily be exchanged between these distinct spectroscopies. Specifically, we design broadband heteronuclear solid-state NMR magic-angle-spinning (MAS) dipolar recoupling experiments based on the recently developed PLATO (PoLarizAtion Transfer via non-linear Optimization) microwave (MW) pulse sequence optimized on a single crystal for powder static-sample DNP. Using this concept, we demonstrate design of ultra-broadband 13C-15N and 2H-13C cross-polarization experiments, using PLATO on the 13C radio-frequency (RF) channel and square/ramped or RESPIRATION (Rotor Echo Short Pulse IRrAdiaTION) RF irradiation on the 15N and 2H RF channels, respectively.

Despite the fundamental importance of the star formation history (SFH) and the initial mass function (IMF) in the description of the Milky Way, their consistent and robust derivation is still elusive. Recent and accurate astrometry and photometry collected by the Gaia satellite provide the natural framework to consolidate these ingredients in our local Galactic environment. We aim to simultaneously infer the IMF and the SFH of the Galactic disc comparing Gaia data with the mock catalog resulting from the Besan\c{c}on population synthesis model (BGM). Our goal is also to estimate the impact of the systematics present in current stellar evolutionary models on this inference. We use a new implementation of the BGM Fast Approximate Simulations (BGM FASt) framework to fit the seven million star Gaia DR3 all-sky $G<13$ color-magnitude diagram (CMD) to the most updated dynamically self-consistent BGM. Our derived SFH supports an abrupt decrease of the star formation approximately 1-1.5 Gyr ago followed by a significant enhancement with a wide plateau in the range 2-6 Gyr ago. A remarkable hiatus appears around 5-7 Gyr ago with a $\sim$1 Gyr shift depending on the set of stellar models. A complex and discrepant evolution at ages older than 8 Gyr deserves further investigation. Precise but discrepant values are found for the power-law indices of the IMF. For the range 0.5-1.53 $M_\odot$ the slope takes a value of $\alpha_2 = 1.45^{+0.19}_{-0.12}$, while for masses larger than 1.53 $M_\odot$ we obtain $\alpha_3 = 1.98^{+0.13}_{-0.05}$. The current implementation of the BGM FASt framework is ready to address executions fitting all-sky Gaia data up to 14-17 apparent limiting magnitude. This will naturally allow us to derive both a reliable SFH for the early epochs of the Galactic disc evolution and a precise slope for the IMF at low masses., Comment: 21 pages, 14 figures; submitted to Astronomy and Astrophysics

When treating moving tumors, the precise delivery of proton therapy by pencil beam scanning (PBS) is challenged by the interplay effect. Although several 4D-optimization methods have been proposed, what is the most beneficial motion management technique is still an open question. In this study, we wish to investigate the dosimetric impact of re-optimizing the PBS spot weights during the treatment delivery in response to, and in anticipation of, variations in the patient's breathing pattern. We simulate for PBS the implementation of a real-time adaptive framework based on principles from receding horizon control. We consider the patient motion as characterized by a one-dimensional amplitude signal and a 4DCT, to simulate breathing of variable frequency. The framework tracks the signal and predicts the future motion with uncertainty increasing with the length of the prediction horizon. After each delivered energy layer, the framework re-optimizes the spot weights of the next layer based on the delivered dose and the predicted motion. For three lung patients, we generate 500 variable breathing patterns to evaluate the dosimetric results of the framework and compare them to those of implementations of previously proposed non-adaptive methods. Compared to the best non-adaptive method, the adaptive framework improves the CTV D98 in the near-worst breathing scenario (5th percentile), from 96.4 to 98.9 % of the prescribed dose and considerably reduces the variation as measured by a mean decrease in the inter-quartile range by more than 80 %. The target coverage improvements are achieved without generally compromising target dose homogeneity or OAR dose. The study indicates that a motion-adaptive approach based on re-optimization of spot weights during delivery has the potential to substantially improve the dosimetric performance of PBS given fast and accurate models of patient motion.

Dust growth is often indirectly inferred observationally in star-forming environments, theoretically predicted to produce mm-sized particles in circumstellar discs, and also presumably witnessed by the predecessors of the terrestrial meteoritic record. For those reasons it is believed that young gas giants under formation in protoplanetary discs with putative circumplanetary discs (CPDs) surrounding them, such as PDS 70c, should be containing mm-sized particles. We model the spectra of a set of CPDs, which we obtained from radiation hydrodynamic simulations at varying Rosseland opacities kappa_R. The kappa_R from the hydrodynamic simulations are matched with consistent opacity sets of ISM-like composition, but grown to larger sizes. Our high kappa_R hydro data nominally corresponds to 10 mum-sized particles, and our low kappa_R-cases correspond to mm-sized particles. We investigate the resulting broad spectral features at first while keeping the overall optical depth in the planetary envelope constant. Dust growth to size distributions dominated by millimeter particles generally results in broad, featureless spectra with black-body like slopes in the far-infrared, while size distributions dominated by small dust develop steeper slopes in the far-infrared and maintain some features stemming from individual minerals. We find that significant dust growth from microns to millimeters can explain the broad features of the PDS 70c data, when upscaling the dust masses from our simulations by x100. Furthermore our results indicate that the spectral range of 30-500 mum is an ideal hunting ground for broadband features arising from the CPD, but that longer wavelengths observed with ALMA can also be used for massive circumplanetary discs, Comment: 12 pages, 8 figures, Accepted to A&A

Isolation levels, consistency guarantees among concurrently execution transactions in local- and distributed systems, have been formalized in a number of models. Thus far, no model can reason about executable implementations of databases or local transaction libraries providing weak isolation levels. Weak isolation levels are characterized by being highly concurrent and, unlike their stronger counterpart serializability, they are not equivalent to the consistency guarantees provided by a transaction library implemented using a global lock. In this paper, we formalize three weak isolation levels in separation logic, namely read uncommitted, read committed, and snapshot isolation. We define modular separation logic specifications that are independent of the underlying transaction library implementation. Historically, isolation levels have been specified using examples of executions between concurrent transactions that are not allowed to occur, and we demonstrate that our specifications correctly prohibit such examples. To show that our specifications are realizable, we formally verify that an executable implementation of a key-value database running the multi-version concurrency control algorithm from the original snapshot isolation paper satisfies our specification of snapshot isolation. Moreover, we prove implications between the specifications -- snapshot isolation implies read committed and read committed implies read uncommitted -- and thus the verification effort of the database serves as proof that all of our specifications are realizable. All results are mechanised in the Coq proof assistant on top of the Iris separation logic framework.

We propose a nanodevice based on multi-terminal Josephson junctions for the creation and detection of poor man's Majorana (PMM) modes. The device consists of a double three-terminal Josephson junction (3TJJ) embedded in a planar semiconductor that engineers a two-site Kitaev chain. We identify the conditions necessary for the emergence of PMM modes by precisely tuning the superconducting phases of the terminals and the inter-junction coupling between the 3TJJs. To validate our findings, we perform numerical simulations that account for disorder in a more general geometry, and test experimentally feasible protocols based on conductance measurements to reliably detect these modes. Unlike traditional approaches that rely on magnetic fields, our design eliminates the need for such fields, potentially increasing the energy-level resolution and significantly expanding the range of compatible semiconductor materials, such as Ge-based heterostructures., Comment: 19 pages, 9 figures

These notes contain -- apart from some physics -- scattered reminiscences of everyday life at the Institute for Theoretical Physics in G\"oteborg in the early eighties. The text has been in the making for many years. Some of it was written inspired by the Lars Brink 70-fest at The Solvay Institute in Brussels in 2014, but was never finished or published. Parts of the old text has now been replaced by a review of the quite unexpected recent renaissance for light-front higher spin theory. Lars sadly did not live to experience more than the very beginnings of this quite remarkable development. The text was finalized in the spring of 2023., Comment: Contribution to a memorial volume for Lars Brink, to be published by World Scientific

This papers consider a two terminal problem, where Alice and Bob jointly want to perform a measurement on a bipartite quantum system $\rho^{AB}$. Alice can transmit the results of her measurements to Bob on a classical channel, and Alice and Bob have common randomness. The question is what is the minimum amount of communications and common randomness needed for faithful simulation. The paper derives an achievable rate region., Comment: This is a version of an ISIT'25 submission with a complete proof

Machine learning (ML) has the potential to revolutionize various domains, but its adoption is often hindered by the disconnect between the needs of domain experts and translating these needs into robust and valid ML tools. Despite recent advances in LLM-based co-pilots to democratize ML for non-technical domain experts, these systems remain predominantly focused on model-centric aspects while overlooking critical data-centric challenges. This limitation is problematic in complex real-world settings where raw data often contains complex issues, such as missing values, label noise, and domain-specific nuances requiring tailored handling. To address this we introduce CliMB-DC, a human-guided, data-centric framework for LLM co-pilots that combines advanced data-centric tools with LLM-driven reasoning to enable robust, context-aware data processing. At its core, CliMB-DC introduces a novel, multi-agent reasoning system that combines a strategic coordinator for dynamic planning and adaptation with a specialized worker agent for precise execution. Domain expertise is then systematically incorporated to guide the reasoning process using a human-in-the-loop approach. To guide development, we formalize a taxonomy of key data-centric challenges that co-pilots must address. Thereafter, to address the dimensions of the taxonomy, we integrate state-of-the-art data-centric tools into an extensible, open-source architecture, facilitating the addition of new tools from the research community. Empirically, using real-world healthcare datasets we demonstrate CliMB-DC's ability to transform uncurated datasets into ML-ready formats, significantly outperforming existing co-pilot baselines for handling data-centric challenges. CliMB-DC promises to empower domain experts from diverse domains -- healthcare, finance, social sciences and more -- to actively participate in driving real-world impact using ML., Comment: Saveliev, Liu & Seedat contributed equally

Bootstrapping is often applied to get confidence limits for semiparametric inference of a target parameter in the presence of nuisance parameters. Bootstrapping with replacement can be computationally expensive and problematic when cross-validation is used in the estimation algorithm due to duplicate observations in the bootstrap samples. We provide a valid, fast, easy-to-implement subsampling bootstrap method for constructing confidence intervals for asymptotically linear estimators and discuss its application to semiparametric causal inference. Our method, inspired by the Cheap Bootstrap (Lam, 2022), leverages the quantiles of a t-distribution and has the desired coverage with few bootstrap replications. We show that the method is asymptotically valid if the subsample size is chosen appropriately as a function of the sample size. We illustrate our method with data from the LEADER trial (Marso et al., 2016), obtaining confidence intervals for a longitudinal targeted minimum loss-based estimator (van der Laan and Gruber, 2012). Through a series of empirical experiments, we also explore the impact of subsample size, sample size, and the number of bootstrap repetitions on the performance of the confidence interval.

We present a method for augmenting a Large Language Model (LLM) with a combination of text and visual data to enable accurate question answering in visualization of scientific data, making conversational visualization possible. LLMs struggle with tasks like visual data interaction, as they lack contextual visual information. We address this problem by merging a text description of a visualization and dataset with snapshots of the visualization. We extract their essential features into a structured text file, highly compact, yet descriptive enough to appropriately augment the LLM with contextual information, without any fine-tuning. This approach can be applied to any visualization that is already finally rendered, as long as it is associated with some textual description.

We explore scattering amplitudes on the Coulomb branch of maximally supersymmetric Yang-Mills theory. We introduce a particular pattern of scalar vacuum expectation values that allow us to define amplitudes with a different mass pattern compared to what was studied previously. This is motivated by an extension of the Amplituhedron that leads to infrared-finite four-particle amplitudes involving massive particles. We work out the Feynman rules on the Coulomb branch and use them, together with generalized unitarity techniques, to perform consistency checks on the Amplituhedron expectations for the one- and two-loop integrands for the four-particle amplitude. We present details of the computation of the required two-loop four-point integrals via a four-dimensional version of the differential equations method. Finally, we study the Regge limit of the four-point amplitude, including the first power suppressed terms. We find that when organized in terms of a suitable expansion parameter, the subleading power term exponentiates, with the exponent matching the anomalous dimension of a cusped Wilson line with a local operator insertion. The latter is known from integrability, which leads to a prediction at higher loop orders in the Regge limit., Comment: 36 pages, 4 figures, 4 ancillary files + readme

Purpose : Planning a transition towards sustainable carbon neutrality at the organization level raises several accounting challenges. This paper aims to shed light on key challenges, highlight answers from current accounting standards and guidance, point out potential inconsistencies or limits, and outline potential solutions from the industrial ecology community through systemic environmental assessment tools, such as life cycle assessment (LCA) and environmentally-extended input-output (EEIO) analysis. Results and discussion: We propose a Measure-Reduce-Neutralize-Control sequence allowing organizations to plan their sustainable net-zero strategy, and discuss 24 accounting challenges occurring within this sequence. We then outline ways forward for organizations planning their carbon neutrality trajectory, pointing to existing resources, and for guidelines providers and the industrial ecology communities to address current limitations in the development of future accounting methods and guidelines. Overarching solutions to many accounting issues are to develop comprehensive, open-source, and high-quality life cycle inventory databases, to enable improved dynamic assessments and prospective LCA through integrated assessment models, to refine methods for assessing mineral scarcity and environmental impacts, the supply in some metals being expected to be a bottleneck to the energy transition, and to identify the appropriate climate metrics for planning sustainable carbon neutrality pathways at the organizational level.

In this paper, we reschedule the duties of train drivers one day before the operation. Due to absent drivers (e.g., because of sick leave), some trains have no driver. Thus, duties need to be rescheduled for the day of operation. We start with a feasible crew schedule for each of the remaining operating drivers, a set of unassigned tasks originally assigned to the absent drivers, and a group of standby drivers with fixed start time, end time, start depot, and end depot. Our aim is to generate a crew schedule with as few canceled or changed tasks as possible. We present a tabu-search-based approach for crew rescheduling. We also adapt a column-generation approach with the same objective function and equivalent restrictions as the benchmark for comparing the results, computational time, and space usage. Our tabu-search-based approach needs both less computation time and space than the column-generation approach to compute an acceptable result. We further test the performance of our approach under different settings. The data used in the experiments originated from a regional passenger-train system around Stockholm, Sweden and was provided by M\"alart\r{a}g.

The vertical shear instability and the streaming instability are two robust sources of turbulence in protoplanetary disks. The former has been found to induce anisotropic turbulence that is stronger in the vertical than in the radial dimension and to be overall stronger compared to the largely isotropic turbulence caused by the streaming instability. In this study, we shed light on the dust diffusion by the vertical shear instability and the streaming instability separately and together, and in particular on the direction- and scale-dependence of the diffusion. To this end, we employ two-dimensional global models of the two instabilities either in isolation or in combination. The vertical shear instability in isolation diffuses dust more strongly in the vertical direction than the streaming instability in isolation, resulting in a wave-shaped dust layer in our two-dimensional simulations. Compared with this large-scale diffusion, though, our study highlights that the vertical shear instability causes substantially weaker or even negligible small-scale diffusion. We validate this result using previously published three-dimensional simulations. In particular when simulating centimetre-sized dust, the undulating dust layer becomes internally razor-thin. In contrast, the diffusion owing to the streaming instability exhibits only a marginal scale-dependence, with the dust layer possessing a Gaussian shape. In models including both instabilities, the undulating mid-plane layer is broadened to a width set by the intrinsic diffusion level caused by the streaming instability., Comment: 15 pages, 8 figures, 2 tables; accepted for publication in Astronomy & Astrophysics

For any arc-weighted oriented graph $D=(V(D), A(D),w)$, we write ${\rm fas}_w(D)$ to denote the minimum weight of a feedback arc set in $D$. In this paper, we consider upper bounds on ${\rm fas}_w(D)$ for arc-weight oriented graphs $D$ with bounded maximum degrees and directed girth. We obtain such bounds by introducing a new parameter ${\rm fasd}(D)$, which is the maximum integer such that $A(D)$ can be partitioned into ${\rm fasd}(D)$ feedback arc sets. This new parameter seems to be interesting in its own right. We obtain several bounds for both ${\rm fas}_w(D)$ and ${\rm fasd}(D)$ when $D$ has maximum degree $\Delta(D)\le \Delta$ and directed girth $g(D)\geq g$. In particular, we show that if $\Delta(D)\leq~4$ and $g(D)\geq 3$, then ${\rm fasd}(D) \geq 3$ and therefore ${\rm fas}_w(D)\leq \frac{w(D)}{3}$ which generalizes a tight bound for an unweighted oriented graph with maximum degree at most 4. We also show that ${\rm fasd}(D)\geq g$ and ${\rm fas}_w(D) \leq \frac{w(D)}{g}$ if $\Delta(D)\leq 3$ and $g(D)\geq g$ for $g\in \{3,4,5\}$ and these bounds are tight. However, for $g=10$ the bound ${\rm fasd}(D)\geq g$ does not always hold when $\Delta(D)\leq 3$. Finally we give some bounds for the cases when $\Delta$ or $g$ are large.

Erd{\H o}s (1963) initiated extensive graph discrepancy research on 2-edge-colored graphs. Gishboliner, Krivelevich, and Michaeli (2023) launched similar research on oriented graphs. They conjectured the following generalization of Dirac's theorem: If the minimum degree $\delta$ of an $n$-vertex oriented graph $G$ is greater or equal to $n/2$,then $G$ has a Hamilton oriented cycle with at least $\delta$ forward arcs. This conjecture was proved by Freschi and Lo (2024) who posed an open problem to extend their result to an Ore-type condition. We propose two conjectures for such extensions and prove some results which provide support to the conjectures. For forward arc maximization on Hamilton oriented cycles and paths in semicomplete multipartite digraphs and locally semicomplete digraphs, we obtain characterizations which lead to polynomial-time algorithms.

In this paper we study the evolution of radiative fluxes, flux radii and observable dust masses in protoplanetary discs, in order to understand how these depend on the angular momentum budget and on the assumed heat sources. We use a model that includes the formation and viscous evolution of protoplanetary gas discs, together with the growth and radial drift of the dust component. We find that we are best able to match the observed fluxes and radii of class 0/I discs when we assume (i) an initial total angular momentum budget corresponding to a centrifugal radius of 40 au around solar-like stars, and (ii) inefficient viscous heating. Fluxes and radii of class II discs appear consistent with disc models with angular momentum budgets equivalent to centrifugal radii of both 40 au or 10 au for solar like stars, and with models where viscous heating occurs at either full efficiency or at reduced efficiency. During the first 0.5 Myr of their evolution discs are generally optically thick at a wavelength of 1.3 mm. However, after this discs are optically thin at mm-wavelengths, supporting standard means of dust mass estimates. Using a disc population synthesis model, we then show that the evolution of the cumulative evolution of the observable dust masses agrees well with that observed in young star forming clusters of different ages., Comment: 12 pagers, 12 figures

Carnot famously showed that engine operation is chiefly characterised by the magnitude of the temperature ratio $T_\mathrm{h}/T_\mathrm{c}$ between its hot and cold reservoirs. While temperature ratios ranging between $1.3-2.8$ and $2-10$ are common in macroscopic commercial engines and engines operating in the microscopic regime, respectively, the quest is to test thermodynamics at its extremes. Here we present the hottest engine on earth, with temperature ratios as high as $110$. We achieve this by realising an underdamped single-particle engine using a charged microparticle that is electrically levitated under vacuum conditions. Noisy electric fields are used to synthesise reservoir temperatures in excess of $10^7$ K. As a result, giant fluctuations show up in all thermodynamic quantities of the engine, such as heat exchange and efficiency. Moreover, we find that the particle experiences an effective position dependent temperature, which gives rise to dynamics that drastically deviates from that of standard Brownian motion. We develop a theoretical model accounting for the effects of this multiplicative noise and find excellent agreement with the measured dynamics. The high level of control over the presented experimental platform opens the door to emulate the stochastic dynamics of cellular and biological processes, and provides thermodynamic insight required for the development of nanotechnologies.

Big data presents potential but unresolved value as a source for analysis and inference. However,selection bias, present in many of these datasets, needs to be accounted for so that appropriate inferences can be made on the target population. One way of approaching the selection bias issue is to first estimate the propensity of inclusion in the big dataset for each member of the big dataset, and then to apply these propensities in an inverse probability weighting approach to produce population estimates. In this paper, we provide details of a new variant of existing propensity score estimation methods that takes advantage of the ability to integrate the big data with a probability sample. We compare the ability of this method to produce efficient inferences for the target population with several alternative methods through an empirical study., Comment: Paper presented at the 2024 International Conference on Establishment Statistics in Glasgow, United Kingdom

Charge transport in materials has an impact on a wide range of devices based on semiconductor, battery or superconductor technology. Charge transport in sliding Charge Density Waves (CDW) differs from all others in that the atomic lattice is directly involved in the transport process. To obtain an overall picture of the structural changes associated to the collective transport, the large coherent X-ray beam generated by an X-ray free-electron laser (XFEL) source was used. The CDW phase can be retrieved over the entire sample from diffracted intensities using a genetic algorithm. For currents below threshold, increasing shear deformation is observed in the central part of the sample while longitudinal deformation appears above threshold when shear relaxes. Shear thus precedes longitudinal deformation, with relaxation of one leading to the appearance of the other. Moreover, strain accumulates on surface steps in the sliding regime, demonstrating the strong pinning character of these surface discontinuities. The sliding process of nanometric CDW is based on an impressive spatial coherence involving the macroscopic sample dimensions., Comment: 20 pages, 5 figures

The complexity of modern electro-mechanical systems require the development of sophisticated diagnostic methods like anomaly detection capable of detecting deviations. Conventional anomaly detection approaches like signal processing and statistical modelling often struggle to effectively handle the intricacies of complex systems, particularly when dealing with multi-variate signals. In contrast, neural network-based anomaly detection methods, especially Auto-Encoders, have emerged as a compelling alternative, demonstrating remarkable performance. However, Auto-Encoders exhibit inherent opaqueness in their decision-making processes, hindering their practical implementation at scale. Addressing this opacity is essential for enhancing the interpretability and trustworthiness of anomaly detection models. In this work, we address this challenge by employing a feature selector to select features and counterfactual explanations to give a context to the model output. We tested this approach on the SKAB benchmark dataset and an industrial time-series dataset. The gradient based counterfactual explanation approach was evaluated via validity, sparsity and distance measures. Our experimental findings illustrate that our proposed counterfactual approach can offer meaningful and valuable insights into the model decision-making process, by explaining fewer signals compared to conventional approaches. These insights enhance the trustworthiness and interpretability of anomaly detection models., Comment: 8 pages, 6 figures, 6 tables, conference proceeding

We prove the following local strengthening of Shearer's classic bound on the independence number of triangle-free graphs: For every triangle-free graph $G$ there exists a probability distribution on its independent sets such that every vertex $v$ of $G$ is contained in a random independent set drawn from the distribution with probability $(1-o(1))\frac{\ln d(v)}{d(v)}$. This resolves the main conjecture raised by Kelly and Postle (2018) about fractional coloring with local demands, which in turn confirms a conjecture by Cames van Batenburg et al. (2018) stating that every $n$-vertex triangle-free graph has fractional chromatic number at most $(\sqrt{2}+o(1))\sqrt{\frac{n}{\ln(n)}}$. Addressing another conjecture posed by Cames van Batenburg et al., we also establish an analogous upper bound in terms of the number of edges. To prove these results we establish a more general technical theorem that works in a weighted setting. As a further application of this more general result, we obtain a new spectral upper bound on the fractional chromatic number of triangle-free graphs: We show that every triangle-free graph $G$ satisfies $\chi_f(G)\le (1+o(1))\frac{\rho(G)}{\ln \rho(G)}$ where $\rho(G)$ denotes the spectral radius. This improves the bound implied by Wilf's classic spectral estimate for the chromatic number by a $\ln \rho(G)$ factor and makes progress towards a conjecture of Harris on fractional coloring of degenerate graphs., Comment: 11 pages, comments welcome

Ergotropy -- the maximal amount of unitarily extractable work -- measures the ``charge level'' of quantum batteries. We prove that in large many-body batteries ergotropy exhibits a concentration of measure phenomenon. Namely, the ergotropy of such systems is almost constant for almost all states sampled from the Hilbert--Schmidt measure. We establish this by first proving that ergotropy, as a function of the state, is Lipschitz-continuous with respect to the Bures distance, and then applying Levy's measure concentration lemma. In parallel, we showcase the analogous properties of von Neumann entropy, compiling and adapting known results about its continuity and concentration properties. Furthermore, we consider the situation with the least amount of prior information about the state. This corresponds to the quantum version of the Jeffreys prior distribution -- the Bures measure. In this case, there exist no analytical bounds guaranteeing exponential concentration of measure. Nonetheless, we provide numerical evidence that ergotropy, as well as von Neumann entropy, concentrate also in this case., Comment: 8+4 pages, 2 figures; comments are welcome

We study reinforcement learning (RL) for decision processes with non-Markovian reward, in which high-level knowledge of the task in the form of reward machines is available to the learner. We consider probabilistic reward machines with initially unknown dynamics, and investigate RL under the average-reward criterion, where the learning performance is assessed through the notion of regret. Our main algorithmic contribution is a model-based RL algorithm for decision processes involving probabilistic reward machines that is capable of exploiting the structure induced by such machines. We further derive high-probability and non-asymptotic bounds on its regret and demonstrate the gain in terms of regret over existing algorithms that could be applied, but obliviously to the structure. We also present a regret lower bound for the studied setting. To the best of our knowledge, the proposed algorithm constitutes the first attempt to tailor and analyze regret specifically for RL with probabilistic reward machines., Comment: 35 pages

Accurate uncertainty information associated with essential climate variables (ECVs) is crucial for reliable climate modeling and understanding the spatiotemporal evolution of the Earth system. In recent years, geoscience and climate scientists have benefited from rapid progress in deep learning to advance the estimation of ECV products with improved accuracy. However, the quantification of uncertainties associated with the output of such deep learning models has yet to be thoroughly adopted. This survey explores the types of uncertainties associated with ECVs estimated from deep learning and the techniques to quantify them. The focus is on highlighting the importance of quantifying uncertainties inherent in ECV estimates, considering the dynamic and multifaceted nature of climate data. The survey starts by clarifying the definition of aleatoric and epistemic uncertainties and their roles in a typical satellite observation processing workflow, followed by bridging the gap between conventional statistical and deep learning views on uncertainties. Then, we comprehensively review the existing techniques for quantifying uncertainties associated with deep learning algorithms, focusing on their application in ECV studies. The specific need for modification to fit the requirements from both the Earth observation side and the deep learning side in such interdisciplinary tasks is discussed. Finally, we demonstrate our findings with two ECV examples, snow cover and terrestrial water storage, and provide our perspectives for future research.

An adaptive data-driven controller is proposed and analysed for the class of infinite-horizon optimal control of positive linear system problems presented in [1]. This controller is synthesized from the solution of a "data-driven algebraic equation" constructed from the model-free Bellman equation used in Q-learning. This algebraic equation is driven by data correlation matrices that does not scale in the number of data points, making for an efficient online implementation. As a result, a sufficient condition that ensures stability and robustness against unmodeled dynamics in the plant is extracted. The derived results also allows for a quantitative characterization of the interplay between explorations and robustness to unmodeled dynamics. The class of optimal control problems considered here bears equivalence to Stochastic Shortest Path (SSP) problems, allowing for a performance comparison between the proposed adaptive policy and model-free algorithms for learning the stochastic shortest path, as presented in the numerical experiment., Comment: Submitted to L4DC

Longitudinal imaging allows for the study of structural changes over time. One approach to detecting such changes is by non-linear image registration. This study introduces Multi-Session Temporal Registration (MUSTER), a novel method that facilitates longitudinal analysis of changes in extended series of medical images. MUSTER improves upon conventional pairwise registration by incorporating more than two imaging sessions to recover longitudinal deformations. Longitudinal analysis at a voxel-level is challenging due to effects of a changing image contrast as well as instrumental and environmental sources of bias between sessions. We show that local normalized cross-correlation as an image similarity metric leads to biased results and propose a robust alternative. We test the performance of MUSTER on a synthetic multi-site, multi-session neuroimaging dataset and show that, in various scenarios, using MUSTER significantly enhances the estimated deformations relative to pairwise registration. Additionally, we apply MUSTER on a sample of older adults from the Alzheimer's Disease Neuroimaging Initiative (ADNI) study. The results show that MUSTER can effectively identify patterns of neuro-degeneration from T1-weighted images and that these changes correlate with changes in cognition, matching the performance of state of the art segmentation methods. By leveraging GPU acceleration, MUSTER efficiently handles large datasets, making it feasible also in situations with limited computational resources.

We study the dynamic structure factor $S(k,\omega)$ of the spin-1/2 chain with long-range, power-law decaying unfrustrated (sign alternating) Heisenberg interactions $J_r \sim (-1)^{r-1} r^{-\alpha}$ by means of stochastic analytic continuation (SAC) of imaginary-time correlations computed by quantum Monte Carlo calculations. We do so in both the long-range antiferromagnetic (AFM, for $\alpha \lesssim 2.23$) and quasi ordered (QLRO, for $\alpha \gtrsim 2.23$) ground-state phases, employing different SAC parametrizations of $S(k,\omega)$ to resolve sharp edges characteristic of fractional quasi-particles and sharp peaks expected with conventional quasi-particles. In order to identify the most statistically accurate parametrization, we apply a newly developed cross-validation method as "model selection" tool. We confirm that the spectral function contains a power-law divergent edge in the QLRO phase and a very sharp (likely $\delta$-function) magnon peak in the AFM phase. From our SAC results, we extract the dispersion relation in the different regimes of the model, and in the AFM phase, we extract the weight of the magnon pole. In the limit where the model reduces to the conventional Heisenberg chain with nearest-neighbor interactions, our $S(k,\omega)$ agrees well with known Bethe ansatz results. In the AFM phase the low-energy dispersion relation is known to be nonlinear, $\omega_k \sim k^z$, and we extract the corresponding dynamic exponent $z(\alpha)$, which in general is somewhat above the form obtained in linear spin-wave theory. We also find a significant continuum above the magnon peak. This study serves as a benchmark for SAC/QMC studies of systems with a transition from conventional to fractionalized quasi-particles., Comment: 22 pages, 28 figures

The continuous scaling of metal-oxide-semiconductor field-effect transistors (MOSFETs) has led to device geometries where charged carriers are increasingly confined to ever smaller channel cross sections. This development is associated with reduced screening of long-range Coulomb interactions. To accurately predict the behavior of such ultra-scaled devices, electron-electron (e-e) interactions must be explicitly incorporated in their quantum transport simulation. In this paper, we present an \textit{ab initio} atomistic simulation framework based on density functional theory, the non-equilibrium Green's function formalism, and the self-consistent GW approximation to perform this task. The implemented method is first validated with a carbon nanotube test structure before being applied to calculate the transfer characteristics of a silicon nanowire MOSFET in a gate-all-around configuration. As a consequence of e-e scattering, the energy and spatial distribution of the carrier and current densities both significantly change, while the on-current of the transistor decreases owing to the Coulomb repulsion between the electrons. Furthermore, we demonstrate how the resulting bandgap modulation of the nanowire channel as a function of the gate-to-source voltage could potentially improve the device performance. To the best of our knowledge, this study is the first one reporting large-scale atomistic quantum transport simulations of nano-devices under non-equilibrium conditions and in the presence of e-e interactions within the GW approximation.

Functional methods can be applied to the quantum effective action to efficiently determine counterterms and matching conditions for effective field theories. We extend the toolbox to two-loop order and beyond and show how to evaluate the expansion of the path integral in a manifestly gauge-covariant manner. We also generalize the method to theories with mixed spin statistics and prove the validity of the hard-region matching formula to all loop orders. The methods are exemplified with a two-loop matching calculation of the Euler-Heisenberg Lagrangian resulting from decoupling the electron in QED., Comment: 63 pages, 5 figures

Mixed integer linear programming (MILP) solvers ship with a staggering number of parameters that are challenging to select a priori for all but expert optimization users, but can have an outsized impact on the performance of the MILP solver. Existing machine learning (ML) approaches to configure solvers require training ML models by solving thousands of related MILP instances, generalize poorly to new problem sizes, and often require implementing complex ML pipelines and custom solver interfaces that can be difficult to integrate into existing optimization workflows. In this paper, we introduce a new LLM-based framework to configure which cutting plane separators to use for a given MILP problem with little to no training data based on characteristics of the instance, such as a natural language description of the problem and the associated LaTeX formulation. We augment these LLMs with descriptions of cutting plane separators available in a given solver, grounded by summarizing the existing research literature on separators. While individual solver configurations have a large variance in performance, we present a novel ensembling strategy that clusters and aggregates configurations to create a small portfolio of high-performing configurations. Our LLM-based methodology requires no custom solver interface, can find a high-performing configuration by solving only a small number of MILPs, and can generate the configuration with simple API calls that run in under a second. Numerical results show our approach is competitive with existing configuration approaches on a suite of classic combinatorial optimization problems and real-world datasets with only a fraction of the training data and computation time.

Structure determination is essential to a mechanistic understanding of diseases and the development of novel therapeutics. Machine-learning-based structure prediction methods have made significant advancements by computationally predicting protein and bioassembly structures from sequences and molecular topology alone. Despite substantial progress in the field, challenges remain to deliver structure prediction models to real-world drug discovery. Here, we present NeuralPLexer3 -- a physics-inspired flow-based generative model that achieves state-of-the-art prediction accuracy on key biomolecular interaction types and improves training and sampling efficiency compared to its predecessors and alternative methodologies. Examined through newly developed benchmarking strategies, NeuralPLexer3 excels in vital areas that are crucial to structure-based drug design, such as physical validity and ligand-induced conformational changes.

Decision trees are a popular machine learning method, known for their inherent explainability. In Explainable AI, decision trees can be used as surrogate models for complex black box AI models or as approximations of parts of such models. A key challenge of this approach is determining how accurately the extracted decision tree represents the original model and to what extent it can be trusted as an approximation of their behavior. In this work, we investigate the use of the Probably Approximately Correct (PAC) framework to provide a theoretical guarantee of fidelity for decision trees extracted from AI models. Based on theoretical results from the PAC framework, we adapt a decision tree algorithm to ensure a PAC guarantee under certain conditions. We focus on binary classification and conduct experiments where we extract decision trees from BERT-based language models with PAC guarantees. Our results indicate occupational gender bias in these models., Comment: Accepted at AAAI 2025