Search

Many important datasets contain samples that are missing one or more feature values. Maintaining the interpretability of machine learning models in the presence of such missing data is challenging. Singly or multiply imputing missing values complicates the model's mapping from features to labels. On the other hand, reasoning on indicator variables that represent missingness introduces a potentially large number of additional terms, sacrificing sparsity. We solve these problems with M-GAM, a sparse, generalized, additive modeling approach that incorporates missingness indicators and their interaction terms while maintaining sparsity through l0 regularization. We show that M-GAM provides similar or superior accuracy to prior methods while significantly improving sparsity relative to either imputation or naive inclusion of indicator variables., Comment: Published in NeurIPS 2024

We evaluate the performance of and level of intergenerational cross subsidy in single-employer and multi-employer collective defined contribution (CDC) schemes which have been designed to be compatible with UK legislation. The single-employer scheme captures the essential features of the Royal Mail CDC scheme, which is currently the only UK CDC scheme. We find that the schemes can be successful in smoothing pension outcomes while outperforming a DC + annuity scheme, but that this outperformance is not guaranteed in a single-employer scheme. There are significant intergenerational cross-subsidies in the single-employer scheme. These qualitatively mirror the cross-subsidies seen in existing defined benefit schemes, but we find the magnitude of the cross-subsidies is much larger in single employer CDC schemes. The multi-employer scheme is intended to minimize such cross-subsidies, but we find that such subsidies still arise due to the approximate pricing methodology implicit in the scheme design. These cross-subsidies tend to cancel out over time, but in any given year they can be large, implying that it is important to use a rigorous pricing methodology when valuing collective pension investments.

We present ProtoViT, a method for interpretable image classification combining deep learning and case-based reasoning. This method classifies an image by comparing it to a set of learned prototypes, providing explanations of the form ``this looks like that.'' In our model, a prototype consists of \textit{parts}, which can deform over irregular geometries to create a better comparison between images. Unlike existing models that rely on Convolutional Neural Network (CNN) backbones and spatially rigid prototypes, our model integrates Vision Transformer (ViT) backbones into prototype based models, while offering spatially deformed prototypes that not only accommodate geometric variations of objects but also provide coherent and clear prototypical feature representations with an adaptive number of prototypical parts. Our experiments show that our model can generally achieve higher performance than the existing prototype based models. Our comprehensive analyses ensure that the prototypes are consistent and the interpretations are faithful.

GPT-4o is an autoregressive omni model that accepts as input any combination of text, audio, image, and video, and generates any combination of text, audio, and image outputs. It's trained end-to-end across text, vision, and audio, meaning all inputs and outputs are processed by the same neural network. GPT-4o can respond to audio inputs in as little as 232 milliseconds, with an average of 320 milliseconds, which is similar to human response time in conversation. It matches GPT-4 Turbo performance on text in English and code, with significant improvement on text in non-English languages, while also being much faster and 50\% cheaper in the API. GPT-4o is especially better at vision and audio understanding compared to existing models. In line with our commitment to building AI safely and consistent with our voluntary commitments to the White House, we are sharing the GPT-4o System Card, which includes our Preparedness Framework evaluations. In this System Card, we provide a detailed look at GPT-4o's capabilities, limitations, and safety evaluations across multiple categories, focusing on speech-to-speech while also evaluating text and image capabilities, and measures we've implemented to ensure the model is safe and aligned. We also include third-party assessments on dangerous capabilities, as well as discussion of potential societal impacts of GPT-4o's text and vision capabilities.

During infectious disease outbreaks, estimates of time-varying pathogen transmissibility, such as the instantaneous reproduction number R(t) or epidemic growth rate r(t), are used to inform decision-making by public health authorities. For directly transmitted infectious diseases, the renewal equation framework is a widely used method for measuring time-varying transmissibility. The framework uses information on the typical time elapsing between an infection and the offspring infections (quantified by the generation time distribution), and R(t), to describe the rate at which currently infected individuals generate new infections. For diseases with transmission cycles involving hosts and vectors, however, renewal equation models have been far less used. This is likely due to difficulties in mechanistically defining generation times that can capture the complexity of multi-stage, human-vector relationships. Here, using dengue as an example, we provide general renewal equations that are derived from first principles using age-structured systems of coupled partial differential equations across human and vector sub-populations. Our framework tracks the multi-stage transmission cycle over calendar time and across stage-specific ages, resulting in governing renewal equations that quantify how the rate at which new infections are generated from existing infections depends on stage-specific processes. The framework provides a foundation on which to base inferential frameworks for estimating R(t) and r(t) for infectious diseases with multiple stages in the transmission cycle, Comment: 14 pages, 3 figures

Ionising radiation interactions in matter can trigger a cascade of processes that underpin long-lived damage in the medium. To date, however, a lack of suitable methodologies has precluded our ability to understand the role that material nanostructure plays in this cascade. Here, we use transient photoabsorption to track the lifetime of free electrons (t_c) in bulk and nanostructured SiO2 (aerogel) irradiated by picosecond-scale (10^-12 s) bursts of X-rays and protons from a laser-driven accelerator. Optical streaking reveals a sharp increase in t_c from < 1 ps to > 50 ps over a narrow average density (p_av) range spanning the expected phonon-fracton crossover in aerogels. Numerical modelling suggests that this discontinuity can be understood by a quenching of rapid, phonon-assisted recovery in irradiated nanostructured SiO_2. This is shown to lead to an extended period of enhanced energy density in the excited electron population. Overall, these results open a direct route to tracking how low-level processes in complex systems can underpin macroscopically observed phenomena and, importantly, the conditions that permit them to emerge.

This work is an extension of our past study focused on a covariant representation of the electromagnetic (EM) current of spin-1/2 Dirac particles, specifically, nucleons. In the past study the EM responses that occur in unpolarized electron scattering from unpolarized nucleons were derived; however, scattering of polarized electrons from polarized nucleons was beyond the scope of that earlier work. Here such extensions are studied in detail. While in other work the EM response has already been developed for the double-polarization scattering problem, that effort was focused on high-energy collider physics. In the present study the formalism is recast into a set of EM response functions that have transparent dependencies on the relevant kinematic variables, especially on how these behave with respect to the momentum p of the (moving) struck, polarized nucleon. The motivation for such a reformulation of the problem is the desire to see a clear path to expansions in $p$ of the EM response for use in devising ``prescriptions for nuclear physics''. Results are provided where comparisons of the full (unexpanded) responses with various approximations that are frequently employed in studies of EM nuclear physics are made, demonstrating that under some circumstances such approximations are reasonable, whereas in other circumstances the expanded results are likely to be invalid. In addition, the EM current operators and approximations to them are discussed in detail., Comment: 42 pages, 9 figures

The Rashomon Effect, coined by Leo Breiman, describes the phenomenon that there exist many equally good predictive models for the same dataset. This phenomenon happens for many real datasets and when it does, it sparks both magic and consternation, but mostly magic. In light of the Rashomon Effect, this perspective piece proposes reshaping the way we think about machine learning, particularly for tabular data problems in the nondeterministic (noisy) setting. We address how the Rashomon Effect impacts (1) the existence of simple-yet-accurate models, (2) flexibility to address user preferences, such as fairness and monotonicity, without losing performance, (3) uncertainty in predictions, fairness, and explanations, (4) reliable variable importance, (5) algorithm choice, specifically, providing advanced knowledge of which algorithms might be suitable for a given problem, and (6) public policy. We also discuss a theory of when the Rashomon Effect occurs and why. Our goal is to illustrate how the Rashomon Effect can have a massive impact on the use of machine learning for complex problems in society.

Prototypical-part models are a popular interpretable alternative to black-box deep learning models for computer vision. However, they are difficult to train, with high sensitivity to hyperparameter tuning, inhibiting their application to new datasets and our understanding of which methods truly improve their performance. To facilitate the careful study of prototypical-part networks (ProtoPNets), we create a new framework for integrating components of prototypical-part models -- ProtoPNeXt. Using ProtoPNeXt, we show that applying Bayesian hyperparameter tuning and an angular prototype similarity metric to the original ProtoPNet is sufficient to produce new state-of-the-art accuracy for prototypical-part models on CUB-200 across multiple backbones. We further deploy this framework to jointly optimize for accuracy and prototype interpretability as measured by metrics included in ProtoPNeXt. Using the same resources, this produces models with substantially superior semantics and changes in accuracy between +1.3% and -1.5%. The code and trained models will be made publicly available upon publication.

Digital mammography is essential to breast cancer detection, and deep learning offers promising tools for faster and more accurate mammogram analysis. In radiology and other high-stakes environments, uninterpretable ("black box") deep learning models are unsuitable and there is a call in these fields to make interpretable models. Recent work in interpretable computer vision provides transparency to these formerly black boxes by utilizing prototypes for case-based explanations, achieving high accuracy in applications including mammography. However, these models struggle with precise feature localization, reasoning on large portions of an image when only a small part is relevant. This paper addresses this gap by proposing a novel multi-scale interpretable deep learning model for mammographic mass margin classification. Our contribution not only offers an interpretable model with reasoning aligned with radiologist practices, but also provides a general architecture for computer vision with user-configurable prototypes from coarse- to fine-grained prototypes., Comment: 8 pages, 6 figures, Accepted for oral presentation at the 2024 CVPR Workshop on Domain adaptation, Explainability, Fairness in AI for Medical Image Analysis (DEF-AI-MIA)

This study examined which variables predicted the sense of presence (being there) in an immersive audio experience, with a focus on the impacts of immersion technology (headphones with spatialised sound versus speaker with 2D stereo sound), the nature of the audio experience, the narrative content, participants emotional and cognitive engagement and personal characteristics such as age and gender. Museum visitors listened to a story on the One Story, Many Voices immersive audio installation. A convergent mixed-methods design was used, including multiple regression analysis of survey data (n 185) and relational analysis of interview data (n 9). This study found mixed methods support from both surveys and interviews to suggest that presence was predicted by the audio quality (especially the audio being perceived as different from other audio stories they had listened to in the past), cognitive engagement (especially being totally absorbed), the narrative (the story and how it was told) and some support was found for emotional engagement (especially feeling a connection with the storyteller). Effects of the environment and immersion technology on presence received only either quantitative or qualitative support. No influences of personal characteristics were found. Findings are relevant for academics, immersive experience (XR) commissioners and developers.

We use deep Spitzer mid-infrared spectroscopic maps of radial strips across three nearby galaxies with well-studied metallicity gradients (M101, NGC 628, and NGC 2403) to explore the physical origins of the observed deficit of polycyclic aromatic hydrocarbons (PAHs) at sub-solar metallicity (i.e. the PAH-metallicity relation or PZR). These maps allow us to trace the evolution of all PAH features from 5-18 $\mu$m as metallicity decreases continuously from solar ($Z_\odot$) to 0.2 $Z_\odot$. The total PAH to dust luminosity ratio remains relatively constant until reaching a threshold of $\sim$$\frac{2}{3}$$Z_\odot$, below which it declines smoothly but rapidly. The PZR has been attributed to preferential destruction of the smallest grains in the hard radiation environments found at low metallicity. In this scenario, a decrease in emission from the shortest wavelength PAH features is expected. In contrast, we find a steep decline in long wavelength power below $Z_\odot$, especially in the 17 $\mu$m feature, with the shorter wavelength PAH bands carrying an increasingly large fraction of power at low metallicity. We use newly developed grain models to reproduce the observed PZR trends, including these variations in fractional PAH feature strengths. The model that best reproduces the data employs an evolving grain size distribution that shifts to smaller sizes as metallicity declines. We interpret this as a result of inhibited grain growth at low metallicity, suggesting continuous replenishment in the interstellar medium is the dominant process shaping the PAH grain population in galaxies., Comment: 23 pages, 13 figures, accepted for publication in ApJ

Correlated noise across multi-qubit architectures is known to be highly detrimental to the operation of error correcting codes and the long-term feasibility of quantum processors. The recent discovery of spatially dependent correlated noise in multi-qubit architectures of superconducting qubits arising from the impact of cosmic radiation and high-energy particles giving rise to quasiparticle poisoning within the substrate has led to intense investigations of mitigation strategies to address this. In contrast correlated noise in semiconductor spin qubits as a function of distance has not been reported to date. Here we report the magnitude, frequency and spatial dependence of noise correlations between four silicon quantum dot pairs as a function of inter-dot distance at frequencies from 0.3mHz to 1mHz. We find the magnitude of charge noise correlations, quantified by the magnitude square coherence $C_{xy}$, are significantly suppressed from $>0.5$ to $<0.1$ as the inter-dot distance increases from 75nm to 300nm. Using an analytical model we confirm that, in contrast to superconducting qubits, the dominant source of correlated noise arises from low frequency charge noise from the presence of two level fluctuators (TLFs) at the native silicon-silicon dioxide surface. Knowing this, we conclude with an important and timely discussion of charge noise mitigation strategies., Comment: 9 pages, 4 figures, 1 table

When materials are patterned in three dimensions, there exist opportunities to tailor and create functionalities associated with an increase in complexity, the breaking of symmetries, and the introduction of curvature and non-trivial topologies. For superconducting nanostructures, the extension to the third dimension may trigger the emergence of new physical phenomena, as well as advances in technologies. Here, we harness three-dimensional (3D) nanopatterning to fabricate and control the emergent properties of a 3D superconducting nanostructure. Not only are we able to demonstrate the existence and motion of superconducting vortices in 3D but, with simulations, we show that the confinement leads to a well-defined bending of the vortices within the volume of the structure. Moreover, we experimentally observe a strong geometrical anisotropy of the critical field, through which we achieve the reconfigurable coexistence of superconducting and normal states in our 3D superconducting architecture, and the local definition of weak links. In this way, we uncover an intermediate regime of nanosuperconductivity, where the vortex state is truly three-dimensional and can be designed and manipulated by geometrical confinement. This insight into the influence of 3D geometries on superconducting properties offers a route to local reconfigurable control for future computing devices, sensors, and quantum technologies., Comment: 22 pages, 12 figures

Spin qubits in silicon are strong contenders for realizing a practical quantum computer. This technology has made remarkable progress with the demonstration of single and two-qubit gates above the fault-tolerant threshold and entanglement of up to three qubits. However, maintaining high fidelity operations while executing multi-qubit algorithms has remained elusive, only being achieved for two spin qubits to date due to the small qubit size, which makes it difficult to control qubits without creating crosstalk errors. Here, we use a four-qubit silicon processor with every operation above the fault tolerant limit and demonstrate Grover's algorithm with a ~95% probability of finding the marked state, one of the most successful implementations to date. Our four-qubit processor is made of three phosphorus atoms and one electron spin precision-patterned into 1.5 nm${}^2$ isotopically pure silicon. The strong resulting confinement potential, without additional confinement gates that can increase cross-talk, leverages the benefits of having both electron and phosphorus nuclear spins. Significantly, the all-to-all connectivity of the nuclear spins provided by the hyperfine interaction not only allows for efficient multi-qubit operations, but also provides individual qubit addressability. Together with the long coherence times of the nuclear and electron spins, this results in all four single qubit fidelities above 99.9% and controlled-Z gates between all pairs of nuclear spins above 99% fidelity. The high control fidelities, combined with >99% fidelity readout of all nuclear spins, allows for the creation of a three-qubit Greenberger-Horne-Zeilinger (GHZ) state with 96.2% fidelity, the highest reported for semiconductor spin qubits so far. Such nuclear spin registers can be coupled via electron exchange, establishing a path for larger scale fault-tolerant quantum processors., Comment: 16 pages, 9 figures, 3 tables

Topological defects, or singularities, play a key role in the statics and dynamics of complex systems. In magnetism, Bloch point singularities represent point defects that mediate the nucleation of textures such as skyrmions and hopfions. However, while the textures are typically stabilised in chiral magnets, the influence of chirality on the Bloch point singularities remains relatively unexplored. Here we harness advanced three-dimensional nanofabrication to explore the influence of chirality on Bloch point singularities by introducing curvature-induced symmetry breaking in a ferromagnetic nanowire. Combining X-ray magnetic microscopy with the application of in situ magnetic fields, we demonstrate that Bloch point singularity-containing domain walls are stabilised in straight regions of the sample, and determine that curvature can be used to tune the energy landscape of the Bloch points. Not only are we able to pattern pinning points but, by controlling the gradient of curvature, we define asymmetric potential wells to realise a robust Bloch point shift-register with non-reciprocal behaviour. These insights into the influence of symmetry and chirality on singularities offers a route to the controlled nucleation and propagation of topological textures, providing opportunities for logic and computing devices.

The functionality of materials is determined by their composition and microstructure, that is, the distribution and orientation of crystalline grains, grain boundaries and the defects within them. The characterisation of the material's microstructure is therefore critical for materials applications such as catalysis, energy storage and buildings. Until now, characterization techniques that map the distribution of grains, their orientation, and the presence of defects have either been limited to surface investigations, to spatial resolutions of a few hundred nanometres, or to systems of thickness around one hundred nanometres, thus requiring destructive sample preparation for measurements and preventing the study of system-representative volumes or the investigation of materials under operational conditions. Here, we present X-ray linear dichroic orientation tomography, a quantitative, non-invasive technique that allows for an intra- and inter-granular characterisation of extended polycrystalline and amorphous materials in three dimensions (3D). We present the detailed characterisation of a polycrystalline sample of vanadium pentoxide (V2O5), a key catalyst in the production of sulfuric acid. In addition to determining the nanoscale composition, we map the crystal orientation throughout the polycrystalline sample with 73 nm spatial resolution. We identify grains, as well as twist, tilt, and twin grain boundaries. We further observe the creation and annihilation of topological defects promoted by the presence of volume crystallographic defects in 3D. Our method's non-destructive and spectroscopic nature opens the door to in-operando combined chemical and microstructural investigations of functional materials, including energy and mechanical materials in existing industries, as well as quantum materials for future technologies.

We present a focused study of radially-resolved varying PAH emission in the low-luminosity AGN-host NGC 4138 using deep Spitzer/IRS spectral maps. Using new model PAH spectra, we investigate whether these variations could be associated with changes to the PAH grain size distribution due to photodestruction by the AGN. Separately, we model the effects of the varying radiation field within NGC 4138, and we use this model to predict the corresponding changes in the PAH emission spectrum. We find that PAH band ratios are strongly variable with radius in this galaxy with short-to-long wavelength band ratios peaking in the starburst ring. The changing mix of starlight appears to have a considerable effect on the trends in these band ratios, and our radiation model predicts the shapes of these trends. However, the amplitude of observed variation is ~2.5 times larger than predicted for some ratios. A cutoff of small grains in the PAH size distribution, as has been suggested for AGN, together with changes in PAH ionization fraction could explain the behavior of the shorter bands, but this model fails to reproduce longer band behaviors. Additionally, we find that short-to-long wavelength PAH band ratios increase slightly within ~270pc of the center, suggesting that the AGN may directly influence PAH emission there., Comment: 19 pages, 13 figures, published in ApJ. Updated values in radiation model to the same as the published version

We use an infrared dataset captured between 1984 and 2017 using several instruments and observatories to report five rare equatorial disturbances that completely altered the appearance of Jupiter's Equatorial Zone (EZ): the clearance of tropospheric clouds revealed a new 5-$\mu$m-bright band encircling the planet at the equator, accompanied by large 5-$\mu$m-bright filaments. Three events were observed in ground-based images in 1973, 1979 and 1992. We report and characterize for the first time the entire evolution of two new episodes of this unusual EZ state that presented their maximum 5-$\mu$m-brightness in December 1999 and February 2007, coinciding with a brown coloration south of the equator and with large bluish filaments and white plumes in the northern EZ at visible wavelengths. We characterize their typical infrared-bright lifetimes of 12-18 months, with possible periodicities of 6-8 or 13-14 years. We predict that a full-scale equatorial disturbance could occur in 2019-21.

Using images at multiple mid-infrared wavelengths, acquired in May 2018 using the VISIR instrument on ESO's Very Large Telescope (VLT), we study Jupiter's pole-to-pole thermal, chemical and aerosol structure in the troposphere and stratosphere. We confirm that the pattern of cool and cloudy anticyclonic zones and warm cloud-free cyclonic belts persists throughout the mid-latitudes, up to the polar boundaries, and evidence a strong correlation with the vertical maximum windshear and the locations of Jupiter's zonal jets. At high latitudes, VISIR images reveal a large region of mid-infrared cooling poleward $\sim$64$^{\circ}$N and $\sim$67$^{\circ}$S extending from the upper troposphere to the stratosphere, co-located with the reflective aerosols observed by JunoCam, and suggesting that aerosols play a key role in the radiative cooling at the poles. Comparison of zonal-mean thermal properties and high-resolution visible imaging from Juno allows us to study the variability of atmospheric properties as a function of altitude and jet boundaries, particularly in the cold southern polar vortex. However, the southern stratospheric polar vortex is partly masked by a warm mid-infrared signature of the aurora. Co-located with the southern main auroral oval, this warming results from the auroral precipitation and/or joule heating which heat the atmosphere and thus cause a significant stratospheric emission. This high emission results from a large enhancement of both ethane and acetylene in the polar region, reinforcing the evidence of enhanced ion-related chemistry in Jupiter's auroral regions.

Considering the growing interest in magnetic materials for unconventional computing, data storage, and sensor applications, there is active research not only on material synthesis but also characterisation of their properties. In addition to structural and integral magnetic characterisations, imaging of magnetization patterns, current distributions and magnetic fields at nano- and microscale is of major importance to understand the material responses and qualify them for specific applications. In this roadmap, we aim to cover a broad portfolio of techniques to perform nano- and microscale magnetic imaging using SQUIDs, spin center and Hall effect magnetometries, scanning probe microscopies, x-ray- and electron-based methods as well as magnetooptics and nanoMRI. The roadmap is aimed as a single access point of information for experts in the field as well as the young generation of students outlining prospects of the development of magnetic imaging technologies for the upcoming decade with a focus on physics, materials science, and chemistry of planar, 3D and geometrically curved objects of different material classes including 2D materials, complex oxides, semi-metals, multiferroics, skyrmions, antiferromagnets, frustrated magnets, magnetic molecules/nanoparticles, ionic conductors, superconductors, spintronic and spinorbitronic materials.