Search

Scanning Electron Microscopy (SEM) is a widely used tool for nanoparticle characterization, but long-term directional drift can compromise image quality. We present a novel algorithm for post-imaging drift correction in SEM nanoparticle imaging. Our approach combines multiple rapidly acquired, noisy images to produce a single high-quality overlay through redundant cross-correlation, preventing drift-induced distortions. The preservation of critical geometrical properties and accurate imaging of surface features were verified using Atomic Force Microscopy. On platinum nanoparticles with diameters of 300 to 1000 nm, significant improvements in the mean-based signal-to-noise ratio (SNR) were achieved, increasing from 4.4 dB in single images to 11.3 dB when overlaying five images. This method offers a valuable tool for enhancing SEM image quality in nanoparticle research and metrology, particularly in settings without specialized hardware-based drift correction.

Optical clocks have improved their frequency stability and estimated accuracy by more than two orders of magnitude over the best caesium microwave clocks that realise the SI second. Accordingly, an optical redefinition of the second has been widely discussed, prompting a need for the consistency of optical clocks to be verified worldwide. While satellite frequency links are sufficient to compare microwave clocks, a suitable method for comparing high-performance optical clocks over intercontinental distances is missing. Furthermore, remote comparisons over frequency links face fractional uncertainties of a few $10^{-18}$ due to imprecise knowledge of each clock's relativistic redshift, which stems from uncertainty in the geopotential determined at each distant location. Here, we report a landmark campaign towards the era of optical clocks, where, for the first time, state-of-the-art transportable optical clocks from Japan and Europe are brought together to demonstrate international comparisons that require neither a high-performance frequency link nor information on the geopotential difference between remote sites. Conversely, the reproducibility of the clocks after being transported between countries was sufficient to determine geopotential height offsets at the level of 4 cm. Our campaign paves the way for redefining the SI second and has a significant impact on various applications, including tests of general relativity, geodetic sensing for geosciences, precise navigation, and future timing networks., Comment: 29 pages, 5 figures

Reinforcement learning (RL) shows promise in control problems, but its practical application is often hindered by the complexity arising from intricate reward functions with constraints. While the reward hypothesis suggests these competing demands can be encapsulated in a single scalar reward function, designing such functions remains challenging. Building on existing work, we start by formulating preferences over trajectories to derive a realistic reward function that balances goal achievement with constraint satisfaction in the application of mobile robotics with dynamic obstacles. To mitigate reward exploitation in such complex settings, we propose a novel two-stage reward curriculum combined with a flexible replay buffer that adaptively samples experiences. Our approach first learns on a subset of rewards before transitioning to the full reward, allowing the agent to learn trade-offs between objectives and constraints. After transitioning to a new stage, our method continues to make use of past experiences by updating their rewards for sample-efficient learning. We investigate the efficacy of our approach in robot navigation tasks and demonstrate superior performance compared to baselines in terms of true reward achievement and task completion, underlining its effectiveness.

We establish a simple yet general parameterization of Higgs-muon interactions within the effective field theory frameworks, including both the Higgs Effective Field Theory (HEFT) and the Standard Model Effective Field Theory (SMEFT). We investigate the potential of a muon collider, operating at center-of-mass energies of 3 and 10 TeV, to probe Higgs-muon interactions. All possible processes involving the direct production of multiple electroweak bosons ($W$, $Z$, and $H$) with up to five final-state particles are considered. Our findings indicate that a muon collider can achieve greater sensitivity than the high-luminosity LHC, especially considering the independence of the Higgs decay branching fraction to muons. Notably, a 10 TeV muon collider offers exceptional sensitivity to muon-Higgs interactions, surpassing the 3 TeV option. In particular, searches based on multi-Higgs production prove highly effective for probing these couplings., Comment: 6 pages, 4 figures

Deep learning is actively being used in biometrics to develop efficient identification and verification systems. Handwritten signatures are a common subset of biometric data for authentication purposes. Generative adversarial networks (GANs) learn from original and forged signatures to generate forged signatures. While most GAN techniques create a strong signature verifier, which is the discriminator, there is a need to focus more on the quality of forgeries generated by the generator model. This work focuses on creating a generator that produces forged samples that achieve a benchmark in spoofing signature verification systems. We use CycleGANs infused with Inception model-like blocks with attention heads as the generator and a variation of the SigCNN model as the base Discriminator. We train our model with a new technique that results in 80% to 100% success in signature spoofing. Additionally, we create a custom evaluation technique to act as a goodness measure of the generated forgeries. Our work advocates generator-focused GAN architectures for spoofing data quality that aid in a better understanding of biometric data generation and evaluation.

Autonomous systems for software engineering are now capable of fixing bugs and developing features. These systems are commonly evaluated on SWE-bench (Jimenez et al., 2024a), which assesses their ability to solve software issues from GitHub repositories. However, SWE-bench uses only Python repositories, with problem statements presented predominantly as text and lacking visual elements such as images. This limited coverage motivates our inquiry into how existing systems might perform on unrepresented software engineering domains (e.g., front-end, game development, DevOps), which use different programming languages and paradigms. Therefore, we propose SWE-bench Multimodal (SWE-bench M), to evaluate systems on their ability to fix bugs in visual, user-facing JavaScript software. SWE-bench M features 617 task instances collected from 17 JavaScript libraries used for web interface design, diagramming, data visualization, syntax highlighting, and interactive mapping. Each SWE-bench M task instance contains at least one image in its problem statement or unit tests. Our analysis finds that top-performing SWE-bench systems struggle with SWE-bench M, revealing limitations in visual problem-solving and cross-language generalization. Lastly, we show that SWE-agent's flexible language-agnostic features enable it to substantially outperform alternatives on SWE-bench M, resolving 12% of task instances compared to 6% for the next best system.

We propose a scalable encoding of combinatorial optimization problems with arbitrary connectivity, including higher-order terms, on arrays of trapped neutral atoms requiring only a global laser drive. Our approach relies on modular arrangements of a small number of problem-independent gadgets. These gadgets represent maximum-weight independent set (MWIS) problems on unit-disk graphs, which are native to such devices. Instead of programming MWIS weights with site-dependent laser detunings, the scheme relies on systematic placements of auxiliary atoms. We show, that these auxiliary atoms can be simultaneously used for both problem-specific programming and the mitigation of unwanted effects originating from the tails of long-range interactions., Comment: 11 pages, 5 figures

Accurate 3D object detection in real-world environments requires a huge amount of annotated data with high quality. Acquiring such data is tedious and expensive, and often needs repeated effort when a new sensor is adopted or when the detector is deployed in a new environment. We investigate a new scenario to construct 3D object detectors: learning from the predictions of a nearby unit that is equipped with an accurate detector. For example, when a self-driving car enters a new area, it may learn from other traffic participants whose detectors have been optimized for that area. This setting is label-efficient, sensor-agnostic, and communication-efficient: nearby units only need to share the predictions with the ego agent (e.g., car). Naively using the received predictions as ground-truths to train the detector for the ego car, however, leads to inferior performance. We systematically study the problem and identify viewpoint mismatches and mislocalization (due to synchronization and GPS errors) as the main causes, which unavoidably result in false positives, false negatives, and inaccurate pseudo labels. We propose a distance-based curriculum, first learning from closer units with similar viewpoints and subsequently improving the quality of other units' predictions via self-training. We further demonstrate that an effective pseudo label refinement module can be trained with a handful of annotated data, largely reducing the data quantity necessary to train an object detector. We validate our approach on the recently released real-world collaborative driving dataset, using reference cars' predictions as pseudo labels for the ego car. Extensive experiments including several scenarios (e.g., different sensors, detectors, and domains) demonstrate the effectiveness of our approach toward label-efficient learning of 3D perception from other units' predictions.

We prove a Stepanov differentiability type theorem for intrinsic graphs in sub-Riemannian Heisenberg groups., Comment: 17 pages

Quantum computing is currently gaining significant attention, not only from the academic community but also from industry, due to its potential applications across several fields for addressing complex problems. For any practical problem which may be tackled using quantum computing, it is imperative to represent the data used onto a quantum computing system. Depending on the application, many different types of data and data structures occur, including regular numbers, higher-dimensional data structures, e.g., n-dimensional images, up to graphs. This report aims to provide an overview of existing methods for representing these data types on gate-based quantum computers., Comment: 16 pages, 2 figures

Recent advancements in medicine have confirmed that brain disorders often comprise multiple subtypes of mechanisms, developmental trajectories, or severity levels. Such heterogeneity is often associated with demographic aspects (e.g., sex) or disease-related contributors (e.g., genetics). Thus, the predictive power of machine learning models used for symptom prediction varies across subjects based on such factors. To model this heterogeneity, one can assign each training sample a factor-dependent weight, which modulates the subject's contribution to the overall objective loss function. To this end, we propose to model the subject weights as a linear combination of the eigenbases of a spectral population graph that captures the similarity of factors across subjects. In doing so, the learned weights smoothly vary across the graph, highlighting sub-cohorts with high and low predictability. Our proposed sample weighting scheme is evaluated on two tasks. First, we predict initiation of heavy alcohol drinking in young adulthood from imaging and neuropsychological measures from the National Consortium on Alcohol and NeuroDevelopment in Adolescence (NCANDA). Next, we detect Dementia vs. Mild Cognitive Impairment (MCI) using imaging and demographic measurements in subjects from the Alzheimer's Disease Neuroimaging Initiative (ADNI). Compared to existing sample weighting schemes, our sample weights improve interpretability and highlight sub-cohorts with distinct characteristics and varying model accuracy.

Physical reservoir computing (PRC) is a computing framework that harnesses the intrinsic dynamics of physical systems for computation. It offers a promising energy-efficient alternative to traditional von Neumann computing for certain tasks, particularly those demanding both memory and nonlinearity. As PRC is implemented across a broad variety of physical systems, the need increases for standardised tools for data processing and model training. In this manuscript, we introduce PRCpy, an open-source Python library designed to simplify the implementation and assessment of PRC for researchers. The package provides a high-level interface for data handling, preprocessing, model training, and evaluation. Key concepts are described and accompanied by experimental data on two benchmark problems: nonlinear transformation and future forecasting of chaotic signals. Throughout this manuscript, which will be updated as a rolling release, we aim to facilitate researchers from diverse disciplines to prioritise evaluating the computational benefits of the physical properties of their systems by simplifying data processing, model training and evaluation., Comment: 19 pages, 5 figures

Deep learning can help uncover patterns in resting-state functional Magnetic Resonance Imaging (rs-fMRI) associated with psychiatric disorders and personal traits. Yet the problem of interpreting deep learning findings is rarely more evident than in fMRI analyses, as the data is sensitive to scanning effects and inherently difficult to visualize. We propose a simple approach to mitigate these challenges grounded on sparsification and self-supervision. Instead of extracting post-hoc feature attributions to uncover functional connections that are important to the target task, we identify a small subset of highly informative connections during training and occlude the rest. To this end, we jointly train a (1) sparse input mask, (2) variational autoencoder (VAE), and (3) downstream classifier in an end-to-end fashion. While we need a portion of labeled samples to train the classifier, we optimize the sparse mask and VAE with unlabeled data from additional acquisition sites, retaining only the input features that generalize well. We evaluate our method - Sparsely Reconstructed Graphs (SpaRG) - on the public ABIDE dataset for the task of sex classification, training with labeled cases from 18 sites and adapting the model to two additional out-of-distribution sites with a portion of unlabeled samples. For a relatively coarse parcellation (64 regions), SpaRG utilizes only 1% of the original connections while improving the classification accuracy across domains. Our code can be found at github.com/yanismiraoui/SpaRG.

Although language model (LM) agents are demonstrating growing potential in many domains, their success in cybersecurity has been limited due to simplistic design and the lack of fundamental features for this domain. We present EnIGMA, an LM agent for autonomously solving Capture The Flag (CTF) challenges. EnIGMA introduces new Agent-Computer Interfaces (ACIs) to improve the success rate on CTF challenges. We establish the novel Interactive Agent Tool concept, which enables LM agents to run interactive command-line utilities essential for these challenges. Empirical analysis of EnIGMA on over 350 CTF challenges from three different benchmarks indicates that providing a robust set of new tools with demonstration of their usage helps the LM solve complex problems and achieves state-of-the-art results on the NYU CTF and Intercode-CTF benchmarks. Finally, we discuss insights on ACI design and agent behavior on cybersecurity tasks that highlight the need to adapt real-world tools for LM agents.

Many longitudinal neuroimaging studies aim to improve the understanding of brain aging and diseases by studying the dynamic interactions between brain function and cognition. Doing so requires accurate encoding of their multidimensional relationship while accounting for individual variability over time. For this purpose, we propose an unsupervised learning model (called \underline{\textbf{Co}}ntrastive Learning-based \underline{\textbf{Gra}}ph Generalized \underline{\textbf{Ca}}nonical Correlation Analysis (CoGraCa)) that encodes their relationship via Graph Attention Networks and generalized Canonical Correlational Analysis. To create brain-cognition fingerprints reflecting unique neural and cognitive phenotype of each person, the model also relies on individualized and multimodal contrastive learning. We apply CoGraCa to longitudinal dataset of healthy individuals consisting of resting-state functional MRI and cognitive measures acquired at multiple visits for each participant. The generated fingerprints effectively capture significant individual differences and outperform current single-modal and CCA-based multimodal models in identifying sex and age. More importantly, our encoding provides interpretable interactions between those two modalities.

In the field of enumeration of weighted walks confined to the quarter plane, it is known that the generating functions behave very differently depending on the chosen step set; in practice, the techniques used in the literature depend on the complexity of the counting series. In this paper we introduce a unified approach based on the theory of elliptic functions, which allows us to have a common proof of the characterisation of the algebraicity and D-finiteness of the generating functions.

Deep learning models generating structural brain MRIs have the potential to significantly accelerate discovery of neuroscience studies. However, their use has been limited in part by the way their quality is evaluated. Most evaluations of generative models focus on metrics originally designed for natural images (such as structural similarity index and Frechet inception distance). As we show in a comparison of 6 state-of-the-art generative models trained and tested on over 3000 MRIs, these metrics are sensitive to the experimental setup and inadequately assess how well brain MRIs capture macrostructural properties of brain regions (i.e., anatomical plausibility). This shortcoming of the metrics results in inconclusive findings even when qualitative differences between the outputs of models are evident. We therefore propose a framework for evaluating models generating brain MRIs, which requires uniform processing of the real MRIs, standardizing the implementation of the models, and automatically segmenting the MRIs generated by the models. The segmentations are used for quantifying the plausibility of anatomy displayed in the MRIs. To ensure meaningful quantification, it is crucial that the segmentations are highly reliable. Our framework rigorously checks this reliability, a step often overlooked by prior work. Only 3 of the 6 generative models produced MRIs, of which at least 95% had highly reliable segmentations. More importantly, the assessment of each model by our framework is in line with qualitative assessments, reinforcing the validity of our approach., Comment: Accepted MICCAI 20224

When the memory parameter of the elephant random walk is above a critical threshold, the process becomes superdiffusive and, once suitably normalised, converges to a non-Gaussian random variable. In a recent paper by the three first authors, it was shown that this limit variable has a density and that the associated moments satisfy a nonlinear recurrence relation. In this work, we exploit this recurrence to derive an asymptotic expansion of the moments and the asymptotic behaviour of the density at infinity. In particular, we show that an asymmetry in the distribution of the first step of the random walk leads to an asymmetry of the tails of the limit variable. These results follow from a new, explicit expression of the Stieltjes transformation of the moments in terms of special functions such as hypergeometric series and incomplete beta integrals. We also obtain other results about the random variable, such as unimodality and, for certain values of the memory parameter, log-concavity., Comment: 25 pages

The integration of coherent emitters into low-loss photonic circuits is a key technology for quantum networking. In this context, nanophotonic silicon devices implanted with erbium are a promising hardware platform that combines advanced wafer-scale nanofabrication technology with coherent emission in the minimal-loss band of optical fibers. Recent studies have reported two distinct sites in the silicon lattice in which erbium can be reproducibly integrated with particularly promising properties. Here, for an in-depth analysis of these sites, resonant fluorescence spectroscopy is performed on a nanophotonic waveguide in magnetic fields applied along different orientations. In this way, the site symmetry is determined, the spin Hamiltonian is reconstructed and a partial fit of the crystal field Hamiltonian is performed. The obtained quantitative description of the magnetic interaction allows the optimization of Zeeman splittings, optical branching ratios or microwave driving to the needs of future experiments. Beyond that, the derived site symmetry constrains the location of the erbium dopant in the silicon unit cell. This is a key step towards a detailed microscopic understanding of the erbium sites, which may help to improve the integration yield, thus paving the way to efficient nanophotonic quantum memories based on the Er:Si platform.

The number of samples in structural brain MRI studies is often too small to properly train deep learning models. Generative models show promise in addressing this issue by effectively learning the data distribution and generating high-fidelity MRI. However, they struggle to produce diverse, high-quality data outside the distribution defined by the training data. One way to address the issue is using causal models developed for 3D volume counterfactuals. However, accurately modeling causality in high-dimensional spaces is a challenge so that these models generally generate 3D brain MRIS of lower quality. To address these challenges, we propose a two-stage method that constructs a Structural Causal Model (SCM) within the latent space. In the first stage, we employ a VQ-VAE to learn a compact embedding of the MRI volume. Subsequently, we integrate our causal model into this latent space and execute a three-step counterfactual procedure using a closed-form Generalized Linear Model (GLM). Our experiments conducted on real-world high-resolution MRI data (1mm) demonstrate that our method can generate high-quality 3D MRI counterfactuals.

Modern machine learning relies nearly exclusively on dedicated electronic hardware accelerators. Photonic approaches, with low consumption and high operation speed, are increasingly considered for inference but, to date, remain mostly limited to relatively basic tasks. Simultaneously, the problem of training deep and complex neural networks, overwhelmingly performed through backpropagation, remains a significant limitation to the size and, consequently, the performance of current architectures and a major compute and energy bottleneck. Here, we experimentally implement a versatile and scalable training algorithm, called direct feedback alignment, on a hybrid electronic-photonic platform. An optical processing unit performs large-scale random matrix multiplications, which is the central operation of this algorithm, at speeds up to 1500 TeraOps. We perform optical training of one of the most recent deep learning architectures, including Transformers, with more than 1B parameters, and obtain good performances on both language and vision tasks. We study the compute scaling of our hybrid optical approach, and demonstrate a potential advantage for ultra-deep and wide neural networks, thus opening a promising route to sustain the exponential growth of modern artificial intelligence beyond traditional von Neumann approaches., Comment: 19 pages, 4 figures

This work introduces Sample-Efficient Speech Diffusion (SESD), an algorithm for effective speech synthesis in modest data regimes through latent diffusion. It is based on a novel diffusion architecture, that we call U-Audio Transformer (U-AT), that efficiently scales to long sequences and operates in the latent space of a pre-trained audio autoencoder. Conditioned on character-aware language model representations, SESD achieves impressive results despite training on less than 1k hours of speech - far less than current state-of-the-art systems. In fact, it synthesizes more intelligible speech than the state-of-the-art auto-regressive model, VALL-E, while using less than 2% the training data., Comment: Interspeech 2024

We present a formalism based on tracking the flow of parity quantum information to implement algorithms on devices with limited connectivity without qubit overhead, SWAP operations or shuttling. Instead, we leverage the fact that entangling gates not only manipulate quantum states but can also be exploited to transport quantum information. We demonstrate the effectiveness of this method by applying it to the quantum Fourier transform (QFT) and the Quantum Approximate Optimization Algorithm (QAOA) with $n$ qubits. This improves upon all state-of-the-art implementations of the QFT on a linear nearest-neighbor architecture, resulting in a total circuit depth of ${5n-3}$ and requiring ${n^2-1}$ CNOT gates. For the QAOA, our method outperforms SWAP networks, which are currently the most efficient implementation of the QAOA on a linear architecture. We further demonstrate the potential to balance qubit count against circuit depth by implementing the QAOA on twice the number of qubits using bi-linear connectivity, which approximately halves the circuit depth., Comment: 10 pages, 5 figures

Current language models demonstrate remarkable proficiency in text generation. However, for many applications it is desirable to control attributes, such as sentiment, or toxicity, of the generated language -- ideally tailored towards each specific use case and target audience. For auto-regressive language models, existing guidance methods are prone to decoding errors that cascade during generation and degrade performance. In contrast, text diffusion models can easily be guided with, for example, a simple linear sentiment classifier -- however they do suffer from significantly higher perplexity than auto-regressive alternatives. In this paper we use a guided diffusion model to produce a latent proposal that steers an auto-regressive language model to generate text with desired properties. Our model inherits the unmatched fluency of the auto-regressive approach and the plug-and-play flexibility of diffusion. We show that it outperforms previous plug-and-play guidance methods across a wide range of benchmark data sets. Further, controlling a new attribute in our framework is reduced to training a single logistic regression classifier., Comment: ACL Findings 2024

The dominant contribution to the energy relaxation of state-of-the-art superconducting qubits is often attributed to their coupling to an ensemble of material defects which behave as two-level systems. These defects have varying microscopic characteristics which result in a large range of observable defect properties such as resonant frequencies, coherence times and coupling rates to qubits $g$. Here, we investigate strategies to mitigate losses to the family of defects that strongly couple to qubits ($g/2\pi\ge$ 0.5 MHz). Such strongly coupled defects are particularly detrimental to the coherence of qubits and to the fidelities of operations relying on frequency excursions, such as flux-activated two-qubit gates. To assess their impact, we perform swap spectroscopy on 92 frequency-tunable qubits and quantify the spectral density of these strongly coupled modes. We show that the frequency configuration of the defects is rearranged by warming up the sample to room temperature, whereas the total number of defects on a processor tends to remain constant. We then explore methods for fabricating qubits with a reduced number of strongly coupled defect modes by systematically measuring their spectral density for decreasing Josephson junction dimensions and for various surface cleaning methods. Our results provide insights into the properties of strongly coupled defect modes and show the benefits of minimizing Josephson junction dimensions to improve qubit properties.

We address the selection and evaluation of uncertain segmentation methods in medical imaging and present two case studies: prostate segmentation, illustrating that for minimal annotator variation simple deterministic models can suffice, and lung lesion segmentation, highlighting the limitations of the Generalized Energy Distance (GED) in model selection. Our findings lead to guidelines for accurately choosing and developing uncertain segmentation models, that integrate aleatoric and epistemic components. These guidelines are designed to aid researchers and practitioners in better developing, selecting, and evaluating uncertain segmentation methods, thereby facilitating enhanced adoption and effective application of segmentation uncertainty in practice., Comment: Published in the conference proceedings of the 21st IEEE International Symposium on Biomedical Imaging (ISBI 2024)

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

Since language models (LMs) now outperform average humans on many challenging tasks, it has become increasingly difficult to develop challenging, high-quality, and realistic evaluations. We address this issue by examining LMs' capabilities to generate code for solving real scientific research problems. Incorporating input from scientists and AI researchers in 16 diverse natural science sub-fields, including mathematics, physics, chemistry, biology, and materials science, we created a scientist-curated coding benchmark, SciCode. The problems in SciCode naturally factorize into multiple subproblems, each involving knowledge recall, reasoning, and code synthesis. In total, SciCode contains 338 subproblems decomposed from 80 challenging main problems. It offers optional descriptions specifying useful scientific background information and scientist-annotated gold-standard solutions and test cases for evaluation. Claude3.5-Sonnet, the best-performing model among those tested, can solve only 4.6% of the problems in the most realistic setting. We believe that SciCode demonstrates both contemporary LMs' progress towards becoming helpful scientific assistants and sheds light on the development and evaluation of scientific AI in the future., Comment: 25 pages, 9 figures, 7 tables

The International Muon Collider Collaboration (IMCC) [1] was established in 2020 following the recommendations of the European Strategy for Particle Physics (ESPP) and the implementation of the European Strategy for Particle Physics-Accelerator R&D Roadmap by the Laboratory Directors Group [2], hereinafter referred to as the the European LDG roadmap. The Muon Collider Study (MuC) covers the accelerator complex, detectors and physics for a future muon collider. In 2023, European Commission support was obtained for a design study of a muon collider (MuCol) [3]. This project started on 1st March 2023, with work-packages aligned with the overall muon collider studies. In preparation of and during the 2021-22 U.S. Snowmass process, the muon collider project parameters, technical studies and physics performance studies were performed and presented in great detail. Recently, the P5 panel [4] in the U.S. recommended a muon collider R&D, proposed to join the IMCC and envisages that the U.S. should prepare to host a muon collider, calling this their "muon shot". In the past, the U.S. Muon Accelerator Programme (MAP) [5] has been instrumental in studies of concepts and technologies for a muon collider., Comment: This document summarises the International Muon Collider Collaboration (IMCC) progress and status of the Muon Collider R&D programme

Models with vector-like leptons can strongly modify the lepton mass generation mechanism and lead to correlated effects in lepton-Higgs couplings and lepton dipole moments. Here we begin an analysis of higher-order corrections in such models by setting up a renormalization scheme with full on-shell conditions on the lepton self energies, masses and fields. A minimal set of fundamental parameters is renormalized in the $\overline{\text{MS}}$ scheme. We provide a detailed discussion of lepton mixing and redundancies at higher orders, show how the relevant counterterms can be obtained from the renormalization conditions, and determine the $\beta$-functions corresponding to the scheme. As a first application we calculate the one-loop effective muon--Higgs coupling and analyse its correlation with the muon anomalous magnetic moment $\Delta a_\mu^{\text{VLL}}$. In the interesting case of large masses and opposite-sign coupling, the lowest-order correlation implies a fixed value of $\Delta a_\mu^{\text{VLL}}$ around $22.5\times 10^{-10}$, while the higher-order corrections significantly reduce this value to the interval $(10...18)\times 10^{-10}$., Comment: 28 pages, 8 figures

This work explores the in-context learning capabilities of State Space Models (SSMs) and presents, to the best of our knowledge, the first theoretical explanation of a possible underlying mechanism. We introduce a novel weight construction for SSMs, enabling them to predict the next state of any dynamical system after observing previous states without parameter fine-tuning. This is accomplished by extending the HiPPO framework to demonstrate that continuous SSMs can approximate the derivative of any input signal. Specifically, we find an explicit weight construction for continuous SSMs and provide an asymptotic error bound on the derivative approximation. The discretization of this continuous SSM subsequently yields a discrete SSM that predicts the next state. Finally, we demonstrate the effectiveness of our parameterization empirically. This work should be an initial step toward understanding how sequence models based on SSMs learn in context., Comment: ICML 2024, Next Generation Sequence Modeling Architectures Workshop

Developing prompt-based methods with Large Language Models (LLMs) requires making numerous decisions, which give rise to a combinatorial search problem. For example, selecting the right pre-trained LLM, prompt, and hyperparameters to attain the best performance for a task typically necessitates evaluating an expoential number of candidates on large validation sets. This exhaustive evaluation can be time-consuming and costly, as both inference and evaluation of LLM-based approaches are resource-intensive. Worse, a lot of computation is wasted: Many hyper-parameter settings are non-competitive, and many samples from the validation set are highly correlated - providing little or no new information. So, if the goal is to identify the best method, it can be done far more efficiently if the validation samples and methods are selected adaptively. In this paper, we propose a novel method to address this challenge. We lean on low-rank matrix factorization to fill in missing evaluations and on multi-armed bandits to sequentially identify the next (method, validation sample)-pair to evaluate. We carefully assess the efficacy of our approach on several competitive benchmark problems and show that it can identify the top-performing method using only 5-15% of the typically needed resources -- resulting in a staggering 85-95% LLM cost savings.

This paper presents a novel approach to aligning large language models (LLMs) with individual human preferences, sometimes referred to as Reinforcement Learning from \textit{Personalized} Human Feedback (RLPHF). Given stated preferences along multiple dimensions, such as helpfulness, conciseness, or humor, the goal is to create an LLM without re-training that best adheres to this specification. Starting from specialized expert LLMs, each trained for one such particular preference dimension, we propose a black-box method that merges their outputs on a per-token level. We train a lightweight Preference Control Model (PCM) that dynamically translates the preference description and current context into next-token prediction weights. By combining the expert models' outputs at the token level, our approach dynamically generates text that optimizes the given preference. Empirical tests show that our method matches or surpasses existing preference merging techniques, providing a scalable, efficient alternative to fine-tuning LLMs for individual personalization.

Noncommutative geometry is a mathematical framework that expresses the structure of space-time in terms of operator algebras. By using the tools of quantum mechanics to describe the geometry, noncommutative space-times are expected to give rise to quantum gravity effects, at least in some regime. This manuscript focuses on the physical aspects of these so-called quantum space-times, in particular through the formalism of field and gauge theories. Scalar field theories are shown to possibly trigger mixed divergences in the infra-red and ultra-violet for the 2-point function at one loop. This phenomenon is generically called UV/IR mixing and stems from a diverging behaviour of the propagator. The analysis of such divergences differs from the commutative case because the momentum space is now also noncommutative. From another perspective, a gauge theory on $\kappa$-Minkowski, a quantum deformation of the Minkowski space-time, is derived. A first perturbative computation is shown to break the gauge invariance, a pathological behaviour common to other quantum space-times. A causality toy model is also developed on $\kappa$-Minkowski, in which an analogue of the speed-of-light limit emerges. The phenomenology of quantum gravity arising from quantum space-times is discussed, together with the actual constraints it imposes. Finally, a toy model for noncommutative gravity is tackled, using the former $\kappa$-Minkowski space-time to describe the tangent space. It necessitates the notion of noncommutative partition of unity specifically defined there., Comment: PhD Thesis defended on 24/06/2024 - 154 pages. Can also be accessed by HAL: https://hal.science/tel-04627007v1

Magnetic skyrmions are promising candidates for information and storage technologies. In the last years, magnetic multilayer systems have been tuned to enable room-temperature skyrmions, stable even in the absence of external magnetic field. There are several models describing the properties of an isolated skyrmion in a homogeneous background for single repetition multilayer stack, however, the description on how the equilibrium skyrmion size in lattices scales with increasing the number of repetitions of the stack remains unaddressed. This question is essential for fundamental and practical perspectives, as the behaviour of an ensemble of skyrmions differs from the isolated case. Based on a multilayer stack hosting a skyrmion lattice, we have carried out a series of imaging experiments scaling up the dipolar interaction by repeating $n$ times the multilayer unit, from $n =1$ up to $n=30$. We have developed an analytical description for the skyrmion radius in the whole multilayer regime, $i.e.$, from thin to thick film limits. Furthermore, we provide insight on how nucleation by an externally applied field can give rise to a lattice with more skyrmions (thus, overfilled) than the predicted by the calculations., Comment: 9 pages, 3 figures

The "ego network of words" model captures structural properties in language production associated with cognitive constraints. While previous research focused on the layer-based structure and its semantic properties, this paper argues that an essential element, the concept of an active network, is missing. The active part of the ego network of words only includes words that are regularly used by individuals, akin to the ego networks in the social domain, where the active part includes relationships regularly nurtured by individuals and hence demanding cognitive effort. In this work, we define a methodology for extracting the active part of the ego network of words and validate it using interview transcripts and tweets. The robustness of our method to varying input data sizes and temporal stability is demonstrated. We also demonstrate that without the active network concept (and a tool for properly extracting the active network from data), the "ego network of words" model is not able to properly estimate the cognitive effort involved and it becomes vulnerable to the amount of data considered (leading to the disappearance of the layered structure in large datasets). Our results are well-aligned with prior analyses of the ego network of words, where the limitation of the data collected led automatically (and implicitly) to approximately consider the active part of the network only. Moreover, the validation on the transcripts dataset (MediaSum) highlights the generalizability of the model across diverse domains and the ingrained cognitive constraints in language usage., Comment: Accepted for publication in IEEE Transactions on Computational Social Systems. Partly supported by projects SoBigData.it (PNRR IR0000013) and ICSC (PNRR CN00000013)

The integration of artificial intelligence (AI) across contemporary industries is not just a technological upgrade but a transformation with profound structural implications. This paper explores the concept of structural risks associated with the rapid integration of advanced AI systems across social, economic, and political systems. This framework challenges the conventional perspectives that primarily focus on direct AI threats such as accidents and misuse and suggests that these more proximate risks are interconnected and influenced by a larger sociotechnical system. By analyzing the interactions between technological advancements and social dynamics, this study isolates three primary categories of structural risk: antecedent structural causes, antecedent system causes, and deleterious feedback loops. We present a comprehensive framework to understand the causal chains that drive these risks, highlighting the interdependence between structural forces and the more proximate risks of misuse and system failures. The paper articulates how unchecked AI advancement can reshape power dynamics, trust, and incentive structures, leading to profound and often unpredictable shifts. We introduce a methodological research agenda for mapping, simulating, and gaming these dynamics aimed at preparing policymakers and national security officials for the challenges posed by next-generation AI technologies. The paper concludes with policy recommendations., Comment: 21 Pages, 2 Figures, 2 Tables

We introduce DataComp for Language Models (DCLM), a testbed for controlled dataset experiments with the goal of improving language models. As part of DCLM, we provide a standardized corpus of 240T tokens extracted from Common Crawl, effective pretraining recipes based on the OpenLM framework, and a broad suite of 53 downstream evaluations. Participants in the DCLM benchmark can experiment with data curation strategies such as deduplication, filtering, and data mixing at model scales ranging from 412M to 7B parameters. As a baseline for DCLM, we conduct extensive experiments and find that model-based filtering is key to assembling a high-quality training set. The resulting dataset, DCLM-Baseline enables training a 7B parameter language model from scratch to 64% 5-shot accuracy on MMLU with 2.6T training tokens. Compared to MAP-Neo, the previous state-of-the-art in open-data language models, DCLM-Baseline represents a 6.6 percentage point improvement on MMLU while being trained with 40% less compute. Our baseline model is also comparable to Mistral-7B-v0.3 and Llama 3 8B on MMLU (63% & 66%), and performs similarly on an average of 53 natural language understanding tasks while being trained with 6.6x less compute than Llama 3 8B. Our results highlight the importance of dataset design for training language models and offer a starting point for further research on data curation., Comment: Project page: https://www.datacomp.ai/dclm/

Amid growing concerns over AI's societal risks--ranging from civilizational collapse to misinformation and systemic bias--this study explores the perceptions of AI experts and the general US registered voters on the likelihood and impact of 18 specific AI risks, alongside their policy preferences for managing these risks. While both groups favor international oversight over national or corporate governance, our survey reveals a discrepancy: voters perceive AI risks as both more likely and more impactful than experts, and also advocate for slower AI development. Specifically, our findings indicate that policy interventions may best assuage collective concerns if they attempt to more carefully balance mitigation efforts across all classes of societal-scale risks, effectively nullifying the near-vs-long-term debate over AI risks. More broadly, our results will serve not only to enable more substantive policy discussions for preventing and mitigating AI risks, but also to underscore the challenge of consensus building for effective policy implementation., Comment: 9 pages, 18 page supplementary materials

Radar sensors are low cost, long-range, and weather-resilient. Therefore, they are widely used for driver assistance functions, and are expected to be crucial for the success of autonomous driving in the future. In many perception tasks only pre-processed radar point clouds are considered. In contrast, radar spectra are a raw form of radar measurements and contain more information than radar point clouds. However, radar spectra are rather difficult to interpret. In this work, we aim to explore the semantic information contained in spectra in the context of automated driving, thereby moving towards better interpretability of radar spectra. To this end, we create a radar spectra-language model, allowing us to query radar spectra measurements for the presence of scene elements using free text. We overcome the scarcity of radar spectra data by matching the embedding space of an existing vision-language model. Finally, we explore the benefit of the learned representation for scene retrieval using radar spectra only, and obtain improvements in free space segmentation and object detection merely by injecting the spectra embedding into a baseline model.