Search

This research was designed to develop and test a model of the Air Force manpower market. The study indicates that previous manpower supply studies failed to account for simultaneous determination of enlistments and retentions and misinterpreted regressions as supply equations. They are, instead, reduced form equations resulting from joint interaction of Air Force demand and enlistment supply activity. A stochastic process model is used to represent the Air Force manpower market. Variables determining manpower inventory levels, personnel turnover, and new enlistment flows are the mandated force level requirement, Air Force and civilian wages, the unemployment rate, and perceived civilian returns to Air Force-provided training. In the model it is possible to trace the impact of changes in the determining variables upon retention and enlistments. The empirical work supports the simultaneous demand and supply approach developed in the study. The results indicate that (1) changes in force level account for most enlistment rate changes during the all-volunteer force period, (2) higher Air Force wages have improved retention and reduce demand for new enlistments, (3) accessions show lower Air Force Qualifications Test scores, and (4) previous estimates of military manpower supply elasticities are biased downward from true elasticity. (Author/CSS)

Age prediction using brain imaging, such as MRIs, has achieved promising results, with several studies identifying the model's residual as a potential biomarker for chronic disease states. In this study, we developed a brain age predictive model using a dataset of 1,220 U.S. veterans (18--80 years) and convolutional neural networks (CNNs) trained on two-dimensional slices of axial T2-weighted fast spin-echo and T2-weighted fluid attenuated inversion recovery MRI images. The model, incorporating a degree-3 polynomial ensemble, achieved an $R^{2}$ of 0.816 on the testing set. Images were acquired at the level of the anterior commissure and the frontal horns of the lateral ventricles. Residual analysis was performed to assess its potential as a biomarker for five ICD-coded conditions: hypertension (HTN), diabetes mellitus (DM), mild traumatic brain injury (mTBI), illicit substance abuse/dependence (SAD), and alcohol abuse/dependence (AAD). Residuals grouped by the number of ICD-coded conditions demonstrated different trends that were statistically significant ($p = 0.002$), suggesting a relationship between disease states and predicted brain age. This association was particularly pronounced in patients over 49 years, where negative residuals (indicating advanced brain aging) correlated with the presence of multiple ICD codes. These findings support the potential of residuals as biomarkers for detecting latent health conditions.

Grasp User Interfaces (GPUIs) are well-suited for dual-tasking between virtual and physical tasks. These interfaces do not require users to release handheld objects, supporting microinteractions that happen in short bursts and cause minimal interruption on the physical task. Design approaches for these interfaces include user elicitation studies and expert-based strategies, which can be combined with computational techniques for quicker and more cost-effective iterations. Current computational tools for designing GPUIs utilize simulations based on kinematic, geometric, and biomechanical parameters. However, the relationship between these low-level factors and higher-level user preferences remains underexplored. In this study, we gathered user preferences using a two-alternative forced choice paradigm with single-finger reach tasks performed while holding objects representative of real-world activities with different grasp types. We present a quantitative analysis of how various low-level factors influence user preference in grasp interactions, identifying the most significant ones. Leveraging this analysis, we developed a predictive model to estimate user preference and integrated it into an existing simulation tool for GPUI design. In addition to enhancing the understanding of design factors in grasp interactions, our predictive model provides a spatial utility metric based on user preferences, paving the way for adaptive GPUI and mixed-initiative systems for better dual-tasking between virtual and physical environments.

Speculative sampling is a popular technique for accelerating inference in Large Language Models by generating candidate tokens using a fast draft model and accepting or rejecting them based on the target model's distribution. While speculative sampling was previously limited to discrete sequences, we extend it to diffusion models, which generate samples via continuous, vector-valued Markov chains. In this context, the target model is a high-quality but computationally expensive diffusion model. We propose various drafting strategies, including a simple and effective approach that does not require training a draft model and is applicable out of the box to any diffusion model. Our experiments demonstrate significant generation speedup on various diffusion models, halving the number of function evaluations, while generating exact samples from the target model.

We provide a convergence analysis of deep feature instrumental variable (DFIV) regression (Xu et al., 2021), a nonparametric approach to IV regression using data-adaptive features learned by deep neural networks in two stages. We prove that the DFIV algorithm achieves the minimax optimal learning rate when the target structural function lies in a Besov space. This is shown under standard nonparametric IV assumptions, and an additional smoothness assumption on the regularity of the conditional distribution of the covariate given the instrument, which controls the difficulty of Stage 1. We further demonstrate that DFIV, as a data-adaptive algorithm, is superior to fixed-feature (kernel or sieve) IV methods in two ways. First, when the target function possesses low spatial homogeneity (i.e., it has both smooth and spiky/discontinuous regions), DFIV still achieves the optimal rate, while fixed-feature methods are shown to be strictly suboptimal. Second, comparing with kernel-based two-stage regression estimators, DFIV is provably more data efficient in the Stage 1 samples., Comment: 46 pages, 1 figure, 2 tables

Data Assimilation is the process in which we improve the representation of the state of a physical system by combining information coming from a numerical model, real-world observations, and some prior modelling. It is widely used to model and to improve forecast systems in Earth science fields such as meteorology, oceanography and environmental sciences. One key aspect of Data assimilation is the analysis step, where the output of the numerical model is adjusted in order to account for the observational data. In Variational Data Assimilation and under Gaussian assumptions, the analysis step comes down to solving a high-dimensional non-linear least-square problem. In practice, this minimization involves successive inversions of large, and possibly ill-conditioned matrices constructed using linearizations of the forward model. In order to improve the convergence rate of these methods, and thus reduce the computational burden, preconditioning techniques are often used to get better-conditioned matrices, but require either the sparsity pattern of the matrix to inverse, or some spectral information. We propose to use Deep Neural Networks in order to construct a preconditioner. This surrogate is trained using some properties of the singular value decomposition, and is based on a dataset which can be constructed online to reduce the storage requirements.

Here we present a study of radial chemical mixing in non-rotating massive main-sequence stars driven by internal gravity waves (IGWs), based on multi-dimensional hydrodynamical simulations with the fully compressible code MUSIC. We examine two proposed mechanisms of material mixing in stars by IGWs that are commonly quoted, relating to thermal diffusion and sub-wavelength shearing. Thermal diffusion provides a non-restorative effect to the waves, leaving material displaced from its previous equilibrium, while shearing arising within the waves drives weak localised flows, mixing the fluid there. Using IGW spectra from the simulations, we evaluate theoretical predictions of mixing rates due to these mechanisms. We show, for $20M_\odot$ main-sequence stars, that neither of these mechanisms are likely to create mixing sufficient to correct inaccuracies in current stellar evolution models. Furthermore, we compare these predictions to results obtained from Lagrangian tracer particles, following a method recently used for global simulations of stellar interiors to measure mixing by IGWs in their radiative zones. We demonstrate that tracer particle methods face significant numerical challenges in measuring the small diffusion coefficients predicted by the aforementioned theories, for which they are prone to yielding artificially enhanced coefficients. Diffusion coefficients based on such methods are currently used with stellar evolution codes for asteroseismic studies, but should be viewed with caution. Finally, in a case where tracer particles do not suffer from numerical artefacts, we suggest that a diffusion model is not suitable for timescales typically considered by two-dimensional numerical simulations., Comment: 18 pages, 13 figures; to be published in MNRAS

As introduced by Gutman and Harary, the independence polynomial of a graph serves as the generating polynomial of its independent sets. In 1987, Alavi, Malde, Schwenk and Erd\H{o}s conjectured that the independence polynomials of all trees are unimodal. In this paper we come up with a new way for proving log-concavity of independence polynomials of graphs by means of their chromatic symmetric functions, which is inspired by a result of Stanley connecting properties of polynomials to positivity of symmetric functions. This method turns out to be more suitable for treating trees with irregular structures, and as a simple application we show that all spiders have log-concave independence polynomials, which provides more evidence for the above conjecture. Moreover, we present two symmetric function analogues of a basic recurrence formula for independence polynomials, and show that all pineapple graphs also have log-concave independence polynomials., Comment: 19 pages, 5 figures

After a brief history of the development of quality factor, useful expressions are derived for the robust Qz(w) quality factor that accurately determines the VSWR input-impedance fractional bandwidth of antennas for isolated resonances and a small enough bandwidth power drop. For closely spaced multiple resonances/antiresonances, a definitive formula is given for the increase in fractional bandwidth enabled by Bode-Fano matching. Methods are given for determining the conventional and complex-energy quality factors of antennas from RLC circuit models. New field-based quality factors Q(w) are derived for antennas with known fields produced by an input current I. These Q(w) are remarkably robust because they equal Qz(w) when the input impedance is available. Like Qz(w), the field-based Q(w) is independent of the choice of origin of the antenna fields and is impervious to extra lengths of transmission lines and surplus reactances. These robust field-based quality factors are used to derive new lower bounds on the quality factors (upper bounds on the bandwidths) of spherical-mode antennas that improve upon the previous Chu/Collin-Rothschild lower bounds for spherical modes. A criterion for antenna supergain is found by combining the Harrington maximum gain formula with the recently derived formula for the reactive power boundaries of antennas. Maximum gain versus minimum quality factor for spherical antennas are determined using the improved lower bounds on quality factor for different values of electrical size ka. Lastly, reduced antenna quality factors allowed by dispersive tuning overcome the traditional Chu lower bounds for lower radiation efficiencies and small enough bandwidth power drops., Comment: 15 pages, 8 figures

Recent self-supervised learning (SSL) models trained on human-like egocentric visual inputs substantially underperform on image recognition tasks compared to humans. These models train on raw, uniform visual inputs collected from head-mounted cameras. This is different from humans, as the anatomical structure of the retina and visual cortex relatively amplifies the central visual information, i.e. around humans' gaze location. This selective amplification in humans likely aids in forming object-centered visual representations. Here, we investigate whether focusing on central visual information boosts egocentric visual object learning. We simulate 5-months of egocentric visual experience using the large-scale Ego4D dataset and generate gaze locations with a human gaze prediction model. To account for the importance of central vision in humans, we crop the visual area around the gaze location. Finally, we train a time-based SSL model on these modified inputs. Our experiments demonstrate that focusing on central vision leads to better object-centered representations. Our analysis shows that the SSL model leverages the temporal dynamics of the gaze movements to build stronger visual representations. Overall, our work marks a significant step toward bio-inspired learning of visual representations., Comment: 13 pages

Recent successes in self-supervised learning (SSL) model spatial co-occurrences of visual features either by masking portions of an image or by aggressively cropping it. Here, we propose a new way to model spatial co-occurrences by aligning local representations (before pooling) with a global image representation. We present CO-SSL, a family of instance discrimination methods and show that it outperforms previous methods on several datasets, including ImageNet-1K where it achieves 71.5% of Top-1 accuracy with 100 pre-training epochs. CO-SSL is also more robust to noise corruption, internal corruption, small adversarial attacks, and large training crop sizes. Our analysis further indicates that CO-SSL learns highly redundant local representations, which offers an explanation for its robustness. Overall, our work suggests that aligning local and global representations may be a powerful principle of unsupervised category learning., Comment: 20 pages

A fundamental alcove $\mathcal{A}$ is a tile in a paving of a vector space $V$ by an affine reflection group $W_{\mathrm{aff}}$. Its geometry encodes essential features of $W_{\mathrm{aff}}$, such as its affine Dynkin diagram $\widetilde{D}$ and fundamental group $\Omega$. In this article we investigate its full isometry group $\mathrm{Aut}(\mathcal{A})$. It is well known that the isometry group of a regular polyhedron is generated by hyperplane reflections on its faces. Being a simplex, an alcove $\mathcal{A}$ is the simplest of polyhedra, nevertheless it is seldom a regular one. In our first main result we show that $\mathrm{Aut}(\mathcal{A})$ is isomorphic to $\mathrm{Aut}(\widetilde{D})$. Building on this connection, we establish that $\mathrm{Aut}(\mathcal{A})$ is an abstract Coxeter group, with generators given by affine isometric involutions of the ambient space. Although these involutions are seldom reflections, our second main result leverages them to construct, by slicing the Komrakov--Premet fundamental polytope $\mathcal{K}$ for the action of $\Omega$, a family of fundamental polytopes for the action of $\mathrm{Aut}(\mathcal{A})$ on $\mathcal{A}$, whose vertices are contained in the vertices of $\mathcal{K}$ and whose faces are parametrized by the so-called balanced minuscule roots, which we introduce here. In an appendix, we discuss some related negative results on stratified centralizers and equivariant triangulations., Comment: 6 Figures, 3 Tables

The dimensionality of vortical structures has recently been extended beyond two dimensions, providing higher-order topological characteristics and robustness for high-capacity information processing and turbulence control. The generation of high-dimensional vortical structures has mostly been demonstrated in classical systems through the complex interference of fluidic, acoustic, or electromagnetic waves. However, natural materials rarely support three- or higher-dimensional vortical structures and their physical interactions. Here, we present a high-dimensional gradient thickness optical cavity (GTOC) in which the optical coupling of planar metal-dielectric multilayers implements topological interactions across multiple dimensions. Topological interactions in high-dimensional GTOC construct non-trivial topological phases, which induce high-dimensional vortical structures in generalized parameter space in three, four dimensions, and beyond. These emergent high-dimensional vortical structures are observed under electro-optic tomography as optical vortex dynamics in two-dimensional real-space, employing the optical thicknesses of the dielectric layers as synthetic dimensions. We experimentally demonstrate emergent vortical structures, optical vortex lines and vortex rings, in a three-dimensional generalized parameter space and their topological transitions. Furthermore, we explore four-dimensional vortical structures, termed optical vortex sheets, which provide the programmability of real-space optical vortex dynamics. Our findings hold significant promise for emulating high-dimensional physics and developing active topological photonic devices., Comment: 21 pages,5 figures

For any compact Lie group $G$ and any $n$ we construct a smooth $G$-manifold $U_n(G)$ such that any smooth $n$-dimensional $G$-manifold can be embedded in $U_n(G)$ with a trivial normal bundle. Furthermore, we show that such embeddings are unique up to equivariant isotopy It is shown that the (inverse limit) of the cohomology of such spaces gives rise to natural classes which are the analogue for $G$-manifolds of characteristic classes for ordinary manifolds. The cohomotopy groups of $U_n(G)$ are shown to be equal to equivariant bordism groups.

We develop a 6-functor formalism $\mathcal{D}_{[0,\infty)}(-)$ with $\mathbb{Z}_p$-linear coefficients on small v-stacks, and discuss consequences for duality and finiteness for pro-\'etale cohomology of rigid-analytic varieties of general pro-\'etale $\mathbb{Q}_p$-local systems as well as first examples motivated by a potential $p$-adic analog of Fargues-Scholze's geometrization program of the local Langlands correspondence., Comment: 85 pages. Comments welcome!

We consider a recent model of random walk that recursively grows the network on which it evolves, namely the Tree Builder Random Walk (TBRW). We introduce a bias $\rho \in (0,\infty)$ towards the root, and exhibit a phase transition for transience/recurrence at a critical threshold $\rho_c =1+2\overline{\nu}$, where $\overline{\nu}$ is the (possibly infinite) expected number of new leaves attached to the walker's position at each step. This generalizes previously known results, which focused on the unbiased case $\rho=1$. The proofs rely on a recursive analysis of the local times of the walk at each vertex of the tree, after a given number of returns to the root. We moreover characterize the strength of the transience (law of large numbers and central limit theorem with positive speed) via standard arguments, establish recurrence at $\rho_c$, and show a condensation phenomenon in the non-critical recurrent case.

Earth-like planets in the circumstellar habitable zone (HZ) may have dramatically different climate outcomes depending on their spin-orbit parameters, altering their habitability for life as we know it. We present a suite of 93 ROCKE-3D general circulation models (GCMs) for planets with the same surface conditions and average annual insolation as Earth, but with a wide range of rotation periods, obliquities, orbital eccentricities, and longitudes of periastra. Our habitability metric $f_\mathrm{HZ}$ is calculated based on the temperature and precipitation in each model across grid cells over land. Latin Hypercube Sampling (LHS) aids in sampling all 4 of the spin-orbit parameters with a computationally feasible number of GCM runs. Statistical emulation then allows us to model $f_\mathrm{HZ}$ as a smooth function with built-in estimates of statistical uncertainty. We fit our emulator to an initial set of 46 training runs, then test with an additional 46 runs at different spin-orbit values. Our emulator predicts the directly GCM-modeled habitability values for the test runs at the appropriate level of accuracy and precision. For orbital eccentricities up to 0.225, rotation period remains the primary driver of the fraction of land that remains above freezing and with precipitation above a threshold value. For rotation periods greater than $\sim 20$ days, habitability drops significantly (from $\sim 70$% to $\sim 20$%), driven primarily by cooler land temperatures. Obliquity is a significant secondary factor for rotation periods less than $\sim 20$ Earth days, with a factor of two impact on habitability that is maximized at intermediate obliquity., Comment: Accepted for publication in the Astrophysical Journal

Harnessing the potential computational advantage of quantum computers for machine learning tasks relies on the uploading of classical data onto quantum computers through what are commonly referred to as quantum encodings. The choice of such encodings may vary substantially from one task to another, and there exist only a few cases where structure has provided insight into their design and implementation, such as symmetry in geometric quantum learning. Here, we propose the perspective that category theory offers a natural mathematical framework for analyzing encodings that respect structure inherent in datasets and learning tasks. We illustrate this with pedagogical examples, which include geometric quantum machine learning, quantum metric learning, topological data analysis, and more. Moreover, our perspective provides a language in which to ask meaningful and mathematically precise questions for the design of quantum encodings and circuits for quantum machine learning tasks., Comment: 17 pages body, 10 pages back matter; Comments welcome!

Techniques enabling large language models (LLMs) to "think more" by generating and attending to intermediate reasoning steps have shown promise in solving complex problems. However, the standard approaches generate sequences of discrete tokens immediately before responding, and so they can incur significant latency costs and be challenging to optimize. In this work, we demonstrate that a frozen LLM can be augmented with an offline coprocessor that operates on the model's key-value (kv) cache. This coprocessor augments the cache with a set of latent embeddings designed to improve the fidelity of subsequent decoding. We train this coprocessor using the language modeling loss from the decoder on standard pretraining data, while keeping the decoder itself frozen. This approach enables the model to learn, in an end-to-end differentiable fashion, how to distill additional computation into its kv-cache. Because the decoder remains unchanged, the coprocessor can operate offline and asynchronously, and the language model can function normally if the coprocessor is unavailable or if a given cache is deemed not to require extra computation. We show experimentally that when a cache is augmented, the decoder achieves lower perplexity on numerous subsequent tokens. Furthermore, even without any task-specific training, our experiments demonstrate that cache augmentation consistently reduces perplexity and improves performance across a range of reasoning-intensive tasks.

Neural Radiance Fields (NeRF) revolutionized novel view synthesis in recent years by offering a new volumetric representation, which is compact and provides high-quality image rendering. However, the methods to edit those radiance fields developed slower than the many improvements to other aspects of NeRF. With the recent development of alternative radiance field-based representations inspired by NeRF as well as the worldwide rise in popularity of text-to-image models, many new opportunities and strategies have emerged to provide radiance field editing. In this paper, we deliver a comprehensive survey of the different editing methods present in the literature for NeRF and other similar radiance field representations. We propose a new taxonomy for classifying existing works based on their editing methodologies, review pioneering models, reflect on current and potential new applications of radiance field editing, and compare state-of-the-art approaches in terms of editing options and performance.

Physical skill acquisition, from sports techniques to surgical procedures, requires instruction and feedback. In the absence of a human expert, Physical Task Guidance (PTG) systems can offer a promising alternative. These systems integrate Artificial Intelligence (AI) and Mixed Reality (MR) to provide realtime feedback and guidance as users practice and learn skills using physical tools and objects. However, designing PTG systems presents challenges beyond engineering complexities. The intricate interplay between users, AI, MR interfaces, and the physical environment creates unique interaction design hurdles. To address these challenges, we present an interaction design toolkit derived from our analysis of PTG prototypes developed by eight student teams during a 10-week-long graduate course. The toolkit comprises Design Considerations, Design Patterns, and an Interaction Canvas. Our evaluation suggests that the toolkit can serve as a valuable resource for practitioners designing PTG systems and researchers developing new tools for human-AI interaction design.

In this note, we show there exist infinitely many trajectories which are bi-normal (i.e. normal at initial and final times) to the xz-plane, in the Spatial Circular Restricted Three-Body Problem, for energies below or slightly above the first critical value and near the primaries, under the assumption of the twist condition as defined by Moreno-van-Koert in arXiv:2011.06562. This is an application of the relative Poincar\'e-Birkhoff theorem for Lagrangians in Liouville domains, as proven by the authors in arXiv:2408.06919.

Magnetic fields are ubiquitous across different physical systems of current interest; from the early Universe, compact astrophysical objects and heavy-ion collisions to condensed matter systems. A proper treatment of the effects produced by magnetic fields during the dynamical evolution of these systems, can help to understand observables that otherwise show a puzzling behavior. Furthermore, when these fields are comparable to or stronger than \Lambda_QCD, they serve as excellent probes to help elucidate the physics of strongly interacting matter under extreme conditions of temperature and density. In this work we provide a comprehensive review of recent developments on the description of QED and QCD systems where magnetic field driven effects are important. These include the modification of meson static properties such as masses and form factors, the chiral magnetic effect, the description of anomalous transport coefficients, superconductivity in extreme magnetic fields, the properties of neutron stars, the evolution of heavy-ion collisions, as well as effects on the QCD phase diagram. We describe recent theory and phenomenological developments using effective models as well as LQCD methods. The work represents a state-of-the-art review of the field, motivated by presentations and discussions during the "Workshop on Strongly Interacting Matter in Strong Electromagnetic Fields" that took place in the European Centre for Theoretical Studies in Nuclear Physics and Related Areas (ECT*) in the city of Trento, Italy, September 25-29, 2023., Comment: 325 pages-long review of recent topics of interest in the field of magnetic field effects on QED and QCD matter. To be susbmitted to PNPP

Plebanski's second heavenly equation reduces the problem of finding a self-dual Einstein metric to solving a non-linear second-order PDE for a single function. Plebanski's original equation is for self-dual metrics obtained as perturbations of the flat metric. Recently, a version of this equation was discovered for self-dual metrics arising as perturbations around a constant curvature background. We provide a new simple derivation of both versions of the Plebanski second heavenly equation. Our derivation relies on the `pure connection' description of self-dual gravity. Our results also suggest a new interpretation to the kinematic algebra of self-dual Yang-Mills theory, as the Lie algebra of (0,1) vector fields on a R4 endowed with a complex structure., Comment: 12 pages, no figures

For each positive integer $n$, let $\mathbb F_{q^n}$ be the unique $n$-degree extension of the finite field $\mathbb F_q$ with $q$ elements, where $q$ is a prime power. It is known that for arbitrary $q$ and $n$, there exists an element $\beta\in \mathbb F_{q^n}$ such that its Galois conjugates $\beta, \beta^q, \ldots, \beta^{q^{n-1}}$ form a basis for $\mathbb F_{q^n}$ as an $\mathbb F_q$-vector space. These elements are called normal and they work as additive generators of finite fields. On the other hand, the multiplicative group $\mathbb F_{q^n}^*$ is cyclic and any generator of this group is a primitive element. Many past works have dealt with the existence of primitive and normal elements with specified properties, including the existence of primitive elements whose traces over intermediate extensions are prescribed. Inspired by the latter, in this paper we explore the existence of normal elements whose norms over intermediate extensions are prescribed. We combine combinatorial and number-theoretic ideas and obtain both asymptotic and concrete results. In particular, we completely solve the problem in the case where only one intermediate extension is considered.

Industrial control systems (ICSs) increasingly rely on digital technologies vulnerable to cyber attacks. Cyber attackers can infiltrate ICSs and execute malicious actions. Individually, each action seems innocuous. But taken together, they cause the system to enter an unsafe state. These attacks have resulted in dramatic consequences such as physical damage, economic loss, and environmental catastrophes. This paper introduces a methodology that restricts actions using protocols. These protocols only allow safe actions to execute. Protocols are written in a domain specific language we have embedded in an interactive theorem prover (ITP). The ITP enables formal, machine-checked proofs to ensure protocols maintain safety properties. We use dynamic attestation to ensure ICSs conform to their protocol even if an adversary compromises a component. Since protocol conformance prevents unsafe actions, the previously mentioned cyber attacks become impossible. We demonstrate the effectiveness of our methodology using an example from the Fischertechnik Industry 4.0 platform. We measure dynamic attestation's impact on latency and throughput. Our approach is a starting point for studying how to combine formal methods and protocol design to thwart attacks intended to cripple ICSs., Comment: This paper was accepted into the ICSS'24 workshop

Monocular facial performance capture in-the-wild is challenging due to varied capture conditions, face shapes, and expressions. Most current methods rely on linear 3D Morphable Models, which represent facial expressions independently of identity at the vertex displacement level. We propose SEREP (Semantic Expression Representation), a model that disentangles expression from identity at the semantic level. It first learns an expression representation from unpaired 3D facial expressions using a cycle consistency loss. Then we train a model to predict expression from monocular images using a novel semi-supervised scheme that relies on domain adaptation. In addition, we introduce MultiREX, a benchmark addressing the lack of evaluation resources for the expression capture task. Our experiments show that SEREP outperforms state-of-the-art methods, capturing challenging expressions and transferring them to novel identities., Comment: For our project page, see https://ubisoft-laforge.github.io/character/serep/

The reasoning abilities of Large Language Models (LLMs) are attracting increasing attention. In this work, we focus on causal reasoning and address the task of establishing causal relationships based on correlation information, a highly challenging problem on which several LLMs have shown poor performance. We introduce a prompting strategy for this problem that breaks the original task into fixed subquestions, with each subquestion corresponding to one step of a formal causal discovery algorithm, the PC algorithm. The proposed prompting strategy, PC-SubQ, guides the LLM to follow these algorithmic steps, by sequentially prompting it with one subquestion at a time, augmenting the next subquestion's prompt with the answer to the previous one(s). We evaluate our approach on an existing causal benchmark, Corr2Cause: our experiments indicate a performance improvement across five LLMs when comparing PC-SubQ to baseline prompting strategies. Results are robust to causal query perturbations, when modifying the variable names or paraphrasing the expressions.

We use numerical simulations to determine whether the saturation length of the self-modulation (SM) instability of a long proton bunch in plasma could be determined by measuring the radius of the bunch halo SM produces. Results show that defocused protons acquire their maximum transverse momentum and exit the wakefields at a distance approximately equal to the saturation length of the wakefields. This suggests that measuring the radius of the halo as a function of plasma length in the AWAKE experiment would yield a very good estimate for the saturation length of SM., Comment: 13 pages, 2 tables, 4 figures, AAC2024 Proceedings

The automatic extraction of character networks from literary texts is generally carried out using natural language processing (NLP) cascading pipelines. While this approach is widespread, no study exists on the impact of low-level NLP tasks on their performance. In this article, we conduct such a study on a literary dataset, focusing on the role of named entity recognition (NER) and coreference resolution when extracting co-occurrence networks. To highlight the impact of these tasks' performance, we start with gold-standard annotations, progressively add uniformly distributed errors, and observe their impact in terms of character network quality. We demonstrate that NER performance depends on the tested novel and strongly affects character detection. We also show that NER-detected mentions alone miss a lot of character co-occurrences, and that coreference resolution is needed to prevent this. Finally, we present comparison points with 2 methods based on large language models (LLMs), including a fully end-to-end one, and show that these models are outperformed by traditional NLP pipelines in terms of recall.

Objective: This study sought to compare the drop in predictive performance over time according to the modeling approach (regression versus machine learning) used to build a kidney transplant failure prediction model with a time-to-event outcome. Study Design and Setting: The Kidney Transplant Failure Score (KTFS) was used as a benchmark. We reused the data from which it was developed (DIVAT cohort, n=2,169) to build another prediction algorithm using a survival super learner combining (semi-)parametric and non-parametric methods. Performance in DIVAT was estimated for the two prediction models using internal validation. Then, the drop in predictive performance was evaluated in the same geographical population approximately ten years later (EKiTE cohort, n=2,329). Results: In DIVAT, the super learner achieved better discrimination than the KTFS, with a tAUROC of 0.83 (0.79-0.87) compared to 0.76 (0.70-0.82). While the discrimination remained stable for the KTFS, it was not the case for the super learner, with a drop to 0.80 (0.76-0.83). Regarding calibration, the survival SL overestimated graft survival at development, while the KTFS underestimated graft survival ten years later. Brier score values were similar regardless of the approach and the timing. Conclusion: The more flexible SL provided superior discrimination on the population used to fit it compared to a Cox model and similar discrimination when applied to a future dataset of the same population. Both methods are subject to calibration drift over time. However, weak calibration on the population used to develop the prediction model was correct only for the Cox model, and recalibration should be considered in the future to correct the calibration drift., Comment: 10 pages, 2 Tables, 3 Figures, Submitted