Search

We provide a mechanism to imprint local temporal correlations in photon streams which have the same character as spatial correlations in liquids. Usual single-photon emitters correspond, in this picture, to a (temporal) gas while uncorrelated light is the ideal gas. We argue that good single-photon sources are those that exhibit such temporal liquid features, i.e., with a plateau for their short-time correlations (as opposed to a linear dependence) and oscillations at later times, which is a direct manifestation of photon time-ordering. We obtain general, closed-form analytical expressions for the second-order coherence function of a broad family of "liquid light" which can be arbitrarily correlated, though never completely crystallized., Comment: Pre-submission version welcoming comments, missing references, etc

Biomedical imaging datasets are often small and biased, meaning that real-world performance of predictive models can be substantially lower than expected from internal testing. This work proposes using generative image editing to simulate dataset shifts and diagnose failure modes of biomedical vision models; this can be used in advance of deployment to assess readiness, potentially reducing cost and patient harm. Existing editing methods can produce undesirable changes, with spurious correlations learned due to the co-occurrence of disease and treatment interventions, limiting practical applicability. To address this, we train a text-to-image diffusion model on multiple chest X-ray datasets and introduce a new editing method RadEdit that uses multiple masks, if present, to constrain changes and ensure consistency in the edited images. We consider three types of dataset shifts: acquisition shift, manifestation shift, and population shift, and demonstrate that our approach can diagnose failures and quantify model robustness without additional data collection, complementing more qualitative tools for explainable AI.

The general form of the linear torsional constitutive relations at finite temperature of the chiral current, energy-momentum tensor, and spin energy potential are computed for a chiral fermion fluid minimally coupled to geometric torsion and with nonzero chiral chemical potential. The corresponding transport coefficients are explicitly calculated in terms of the energy and number densities evaluated at vanishing torsion. A microscopic calculation of these constitutive relations in some particular backgrounds is also presented, confirming the general structure found., Comment: 28 pages, no figures. v2: minor changes, typos corrected, and references added. It matches the version to be published in Journal of High Energy Physics

Photometric time series gathered by space telescopes such as CoRoT and Kepler allow to detect solar-like oscillations in red-giant stars and to measure their global seismic constraints, which can be used to infer global stellar properties (e.g. masses, radii, evolutionary states). Combining such precise constraints with photospheric abundances provides a means of testing mixing processes that occur inside red-giant stars. In this work, we conduct a detailed spectroscopic and seismic analysis of nine nearby (d < 200 pc) red-giant stars observed by Kepler. Both seismic constraints and grid-based modelling approaches are used to determine precise fundamental parameters for those evolved stars. We compare distances and radii derived from Gaia Data Release 3 parallaxes with those inferred by a combination of seismic, spectroscopic and photometric constraints. We find no deviations within errorsbars, however the small sample size and the associated uncertainties are a limiting factor for such comparison. We use the period spacing of mixed modes to distinguish between ascending red-giants and red-clump stars. Based on the evolutionary status, we apply corrections to the values of $\Delta\nu$ for some stars, resulting in a slight improvement to the agreement between seismic and photometric distances. Finally, we couple constraints on detailed chemical abundances with the inferred masses, radii and evolutionary states. Our results corroborate previous studies that show that observed abundances of lithium and carbon isotopic ratio are in contrast with predictions from standard models, giving robust evidence for the occurrence of additional mixing during the red-giant phase., Comment: 18 pages, 19 figures

This work seeks to make explicit the operational connection between the preparation of two-level quantum systems with their corresponding description (as states) in a Hilbert space. This may sound outdated, but we show there is more to this connection than common sense may lead us to believe. To bridge these two separated realms -- the actual laboratory and the space of states -- we rely on a paradigmatic mathematical object: the Hopf fibration. We illustrate how this connection works in practice with a simple optical setup. Remarkably, this optical setup also reflects the necessity of using two charts to cover a sphere. Put another way, our experimental realization reflects the bi-dimensionality of a sphere seen as a smooth manifold., Comment: 10 pages, 2 figures, minor adjustments and a section added according to reviewer's suggestions

The so-called Golden Mass Relation provides a testable correlation between charged-lepton and down-type quark masses, that arises in certain flavor models that do not rely on Grand Unification. Such models typically involve broken family symmetries. In this work, we demonstrate that realistic fermion mass relations can emerge naturally in modular invariant models, without relying on ad hoc flavon alignments. We provide a model-independent derivation of a class of mass relations that are experimentally testable. These relations are determined by both the Clebsch-Gordan coefficients of the specific finite modular group and the expansion coefficients of its modular forms, thus offering potential probes of modular invariant models. As a detailed example, we present a set of viable mass relations based on the $\Gamma_4\cong S_4$ symmetry, which have calculable deviations from the usual Golden Mass Relation., Comment: 23 pages, 5 figures. Comments welcome! v2: Minor comments added, references added, matches published version. v3: Plots updated and improved, Acknowledgments expanded

Bulk V2O3 features concomitant metal-insulator (MIT) and structural (SPT) phase transitions at TC ~ 160 K. In thin films, where the substrate clamping can impose geometrical restrictions on the SPT, the epitaxial relation between the V2O3 film and substrate can have a profound effect on the MIT. Here we present a detailed characterization of domain nucleation and growth across the MIT in (001)-oriented V2O3 films grown on sapphire. By combining scanning electron transmission microscopy (STEM) and photoelectron emission microscopy (PEEM), we imaged the MIT with planar and vertical resolution. We observed that upon cooling, insulating domains nucleate at the top of the film, where strain is lowest, and expand downwards and laterally. This growth is arrested at a critical thickness of 50 nm from the substrate interface, leaving a persistent bottom metallic layer. As a result, the MIT cannot take place in the interior of films below this critical thickness. However, PEEM measurements revealed that insulating domains can still form on a very thin superficial layer at the top interface. Our results demonstrate the intricate spatial complexity of the MIT in clamped V2O3, especially the strain-induced large variations along the c-axis. Engineering the thickness-dependent MIT can provide an unconventional way to build out-of-plane geometry devices by using the persistent bottom metal layer as a native electrode.

Monocular 3D human pose estimation (3D-HPE) is an inherently ambiguous task, as a 2D pose in an image might originate from different possible 3D poses. Yet, most 3D-HPE methods rely on regression models, which assume a one-to-one mapping between inputs and outputs. In this work, we provide theoretical and empirical evidence that, because of this ambiguity, common regression models are bound to predict topologically inconsistent poses, and that traditional evaluation metrics, such as the MPJPE, P-MPJPE and PCK, are insufficient to assess this aspect. As a solution, we propose ManiPose, a novel manifold-constrained multi-hypothesis model capable of proposing multiple candidate 3D poses for each 2D input, together with their corresponding plausibility. Unlike previous multi-hypothesis approaches, our solution is completely supervised and does not rely on complex generative models, thus greatly facilitating its training and usage. Furthermore, by constraining our model to lie within the human pose manifold, we can guarantee the consistency of all hypothetical poses predicted with our approach, which was not possible in previous works. We illustrate the usefulness of ManiPose in a synthetic 1D-to-2D lifting setting and demonstrate on real-world datasets that it outperforms state-of-the-art models in pose consistency by a large margin, while still reaching competitive MPJPE performance.

T Pyx is one of the most enigmatic recurrent novae, and it has been proposed as a potential Galactic type-Ia supernova progenitor. Using spatially-resolved data obtained with MUSE, we characterized the geometrical distribution of the material expelled in previous outbursts surrounding the white dwarf progenitor. We used a 3D model for the ejecta to determine the geometric distribution of the extended remnant. We have also calculated the nebular parallax distance ($d = 3.55 \pm 0.77$ kpc) based on the measured velocity and spatial shift of the 2011 bipolar ejecta. These findings confirm previous results, including data from the GAIA mission. The remnant of T Pyx can be described by a two-component model, consisting of a tilted ring at $i = 63.7$ deg, relative to its normal vector and by fast bipolar ejecta perpendicular to the plane of the equatorial ring. We find an upper limit for the bipolar outflow ejected mass in 2011 of the bipolar outflow of $M_{ej,b} < (3.0 \pm 1.0) \times 10^{-6}$ M$_{\odot}$, which is lower than previous estimates given in the literature. However, only a detailed physical study of the equatorial component could provide an accurate estimate of the total ejecta of the last outburst, a fundamental step to understand if T Pyx will end its life as a type-Ia supernova., Comment: 9 pages, 8 figures

We examine simple models with an inert scalar and vector-like leptons that can explain the recent $g_{\mu}-2$ measurement reported at FNAL. Prompted by this exciting result, we explore the viability of a simple interpretation and determine the required parameters. We also embed these models within a 3-3-1 gauge extension of the Standard Model (SM), showing that the $g_{\mu}-2$ anomaly can be accommodated in agreement with current data. We also show how our theory can be tested at high-energy colliders such as HL-LHC and FCC-hh., Comment: 9 pages, 6 figures

DUNE is an international experiment dedicated to addressing some of the questions at the forefront of particle physics and astrophysics, including the mystifying preponderance of matter over antimatter in the early universe. The dual-site experiment will employ an intense neutrino beam focused on a near and a far detector as it aims to determine the neutrino mass hierarchy and to make high-precision measurements of the PMNS matrix parameters, including the CP-violating phase. It will also stand ready to observe supernova neutrino bursts, and seeks to observe nucleon decay as a signature of a grand unified theory underlying the standard model. The DUNE far detector implements liquid argon time-projection chamber (LArTPC) technology, and combines the many tens-of-kiloton fiducial mass necessary for rare event searches with the sub-centimeter spatial resolution required to image those events with high precision. The addition of a photon detection system enhances physics capabilities for all DUNE physics drivers and opens prospects for further physics explorations. Given its size, the far detector will be implemented as a set of modules, with LArTPC designs that differ from one another as newer technologies arise. In the vertical drift LArTPC design, a horizontal cathode bisects the detector, creating two stacked drift volumes in which ionization charges drift towards anodes at either the top or bottom. The anodes are composed of perforated PCB layers with conductive strips, enabling reconstruction in 3D. Light-trap-style photon detection modules are placed both on the cryostat's side walls and on the central cathode where they are optically powered. This Technical Design Report describes in detail the technical implementations of each subsystem of this LArTPC that, together with the other far detector modules and the near detector, will enable DUNE to achieve its physics goals., Comment: 425 pages; 281 figures Central editing team: A. Heavey, S. Kettell, A. Marchionni, S. Palestini, S. Rajogopalan, R. J. Wilson

Alzheimer's disease has an increasing prevalence in the population world-wide, yet current diagnostic methods based on recommended biomarkers are only available in specialized clinics. Due to these circumstances, Alzheimer's disease is usually diagnosed late, which contrasts with the currently available treatment options that are only effective for patients at an early stage. Blood-based biomarkers could fill in the gap of easily accessible and low-cost methods for early diagnosis of the disease. In particular, immune-based blood-biomarkers might be a promising option, given the recently discovered cross-talk of immune cells of the central nervous system with those in the peripheral immune system. With the help of machine learning algorithms and mechanistic modeling approaches, such as agent-based modeling, an in-depth analysis of the simulation of cell dynamics is possible as well as of high-dimensional omics resources indicative of pathway signaling changes. Here, we give a background on advances in research on brain-immune system cross-talk in Alzheimer's disease and review recent machine learning and mechanistic modeling approaches which leverage modern omics technologies for blood-based immune system-related biomarker discovery.

In recent years, the potential applications of machine learning to Minimally Invasive Surgery (MIS) have spurred interest in data sets that can be used to develop data-driven tools. This paper introduces a novel dataset recorded during ex vivo pseudo-cholecystectomy procedures on pig livers, utilizing the da Vinci Research Kit (dVRK). Unlike current datasets, ours bridges a critical gap by offering not only full kinematic data but also capturing all pedal inputs used during the procedure and providing a time-stamped record of the endoscope's movements. Contributed by seven surgeons, this data set introduces a new dimension to surgical robotics research, allowing the creation of advanced models for automating console functionalities. Our work addresses the existing limitation of incomplete recordings and imprecise kinematic data, common in other datasets. By introducing two models, dedicated to predicting clutch usage and camera activation, we highlight the dataset's potential for advancing automation in surgical robotics. The comparison of methodologies and time windows provides insights into the models' boundaries and limitations., Comment: 6 pages, 8 figures, 5 tables. Accepted for presentation at the 2024 International Symposium on Medical Robotics

We show that links of invertible polynomials coming from the Johnson and Koll\'ar list of K\"ahler-Einstein 3-folds that are rational homology 7-spheres remain rational homology 7-spheres under the so-called Berglund-H\"ubsch transpose rule coming from classical mirror symmetry constructions. Actually, this rule produces twins, that is, links with same degree, Milnor number and homology H_3, with the exception of iterated Thom-Sebastiani sums of singularities of chain and cycle type, where the torsion and the Milnor number may vary. The Berglund-H\"ubsch transpose rule not only gives a framework to better understand the existence of SasakiEinstein twins but also gives a mechanism for producing new examples of Sasaki-Einstein twins in the rational homology 7 -sphere setting. We also give reasonable conditions for a Sasaki-Einstein rational homology 7-sphere to remain Sasaki-Einstein under the BH-transpose rule. In particular, we found 75 new examples of Sasaki-Einstein rational homology 7-spheres arising as links of not well-formed hypersurface singularities., Comment: We have changed the title, corrected minor typos and added Remark 4.2

We present a radiology-specific multimodal model for the task for generating radiological reports from chest X-rays (CXRs). Our work builds on the idea that large language model(s) can be equipped with multimodal capabilities through alignment with pre-trained vision encoders. On natural images, this has been shown to allow multimodal models to gain image understanding and description capabilities. Our proposed model (MAIRA-1) leverages a CXR-specific image encoder in conjunction with a fine-tuned large language model based on Vicuna-7B, and text-based data augmentation, to produce reports with state-of-the-art quality. In particular, MAIRA-1 significantly improves on the radiologist-aligned RadCliQ metric and across all lexical metrics considered. Manual review of model outputs demonstrates promising fluency and accuracy of generated reports while uncovering failure modes not captured by existing evaluation practices. More information and resources can be found on the project website: https://aka.ms/maira., Comment: 18 pages, 9 tables, 5 figures. v2 adds test IDs and image encoder citation. v3 fixes error in NPV/specificity

We study the behavior of the solution of a stochastic Allen-Cahn equation $\frac{\partial u_\eps }{\partial t}=\frac 12 \frac{\partial^2 u_\eps }{\partial x^2}+ u_\eps -u_\eps^3+\sqrt\eps\, \dot W$, with Dirichlet boundary conditions on a suitably large space interval $[-L_\eps , L_\eps]$, starting from the identically zero function, and where $\dot W$ is a space-time white noise. Our main goal is the description, in the small noise limit, of the onset of the phase separation, with the emergence of spatial regions where $u_\eps$ becomes close $1$ or $-1$. The time scale and the spatial structure are determined by a suitable Gaussian process that appears as the solution of the corresponding linearized equation. This issue has been initially examined by De Masi et al. [Ann. Probab. {\bf 22}, (1994), 334-371] in the related context of a class of reaction-diffusion models obtained as a superposition of a speeded up stirring process and a spin flip dynamics on $\{-1,1\}^{\mathbb{Z}_\eps}$, where $\mathbb{Z}_\eps=\mathbb{Z}$ modulo $\lfloor\eps^{-1}L_\eps\rfloor$.

Optical communication can be revolutionized by encoding data into the orbital angular momentum of light beams. However, state-of-the-art approaches for dynamic control of complex optical wavefronts are mainly based on liquid crystal spatial light modulators or miniaturized mirrors, which suffer from intrinsically slow response times. Here, we experimentally realize a hybrid meta-optical system that enables complex control of the wavefront of light with pulse-duration limited dynamics. Specifically, by combining ultrafast polarization switching in a WSe2 monolayer with a dielectric metasurface, we demonstrate second harmonic beam deflection and structuring of orbital angular momentum on the femtosecond timescale. Our results pave the way to robust encoding of information for free space optical links, while reaching response times compatible with real-world telecom applications.

We introduce Spiral, a third-order integration algorithm for the rotational motion of extended bodies. It requires only one force calculation per time step, does not require quaternion normalization at each time step, and can be formulated for both leapfrog and synchronous integration schemes, making it compatible with many particle simulation codes. The stability and precision of Spiral exceed those of state-of-the-art algorithms currently used in popular DEM codes such as Yade, MercuryDPM, LIGGGHTS, PFC, and more, at only slightly higher computational cost. Also, beyond DEM, we see potential applications in all numerical simulations that involve the 3D rotation of extended bodies., Comment: 25 pages, 9 figures

We report a detailed study of high-order harmonic generation (HHG) in helium. When comparing predictions from a single-active-electron model with those from all-electron simulations, such as ATTOMESA and R-matrix with time-dependence, which can include different numbers of states in the close-coupling expansion, it seems imperative to generate absolute numbers for the HHG spectrum in a well-defined framework. While qualitative agreement in the overall frequency dependence of the spectrum, including the cut-off frequency predicted by a semi-classical model, can be achieved by many models in arbitrary units, only absolute numbers can be used for benchmark comparisons between different approaches.

We investigate phonon lifetimes in VO2 single crystals. We do so in the metallic state above the metal-insulator transition (MIT), where strong structural fluctuations are known to take place. By combining inelastic X-ray scattering and Raman spectroscopy, we track the temperature dependence of several acoustic and optical phonon modes up to 1000 K. Contrary to what is commonly observed, we find that phonon lifetimes decrease with decreasing temperature. Our results show that pre-transitional fluctuations in the metallic state give rise to strong electron-phonon scattering that onsets hundreds of degrees above the transition and increases as the MIT is approached. Notably, this effect is not limited to specific points of reciprocal space that could be associated with the structural transition.

Interoperability has been a focus of attention over at least four decades, with the emergence of several interoperability types (or levels), diverse models, frameworks, and solutions, also as a result of a continuous effort from different domains. The current heterogeneity in technologies such as blockchain, IoT and new application domains such as Industry 4.0 brings not only new interaction possibilities but also challenges for interoperability. Moreover, confusion and ambiguity in the current understanding of interoperability types exist, hampering stakeholders' communication and decision making. This work presents an updated panorama of software-intensive systems interoperability with particular attention to its types. For this, we conducted a tertiary study that scrutinized 37 secondary studies published from 2012 to 2023, from which we found 36 interoperability types associated with 117 different definitions, besides 13 interoperability models and six frameworks in various domains. This panorama reveals that the concern with interoperability has migrated from technical to social-technical issues going beyond the software systems' boundary and still requiring solving many open issues. We also address the urgent actions and also potential research opportunities to leverage interoperability as a multidisciplinary research field to achieve low-coupled, cost-effective, and interoperable systems., Comment: 33 pages

The recent success of general-domain large language models (LLMs) has significantly changed the natural language processing paradigm towards a unified foundation model across domains and applications. In this paper, we focus on assessing the performance of GPT-4, the most capable LLM so far, on the text-based applications for radiology reports, comparing against state-of-the-art (SOTA) radiology-specific models. Exploring various prompting strategies, we evaluated GPT-4 on a diverse range of common radiology tasks and we found GPT-4 either outperforms or is on par with current SOTA radiology models. With zero-shot prompting, GPT-4 already obtains substantial gains ($\approx$ 10% absolute improvement) over radiology models in temporal sentence similarity classification (accuracy) and natural language inference ($F_1$). For tasks that require learning dataset-specific style or schema (e.g. findings summarisation), GPT-4 improves with example-based prompting and matches supervised SOTA. Our extensive error analysis with a board-certified radiologist shows GPT-4 has a sufficient level of radiology knowledge with only occasional errors in complex context that require nuanced domain knowledge. For findings summarisation, GPT-4 outputs are found to be overall comparable with existing manually-written impressions., Comment: EMNLP 2023 main

Thoracic aortic aneurysm (TAA) is a fatal disease which potentially leads to dissection or rupture through progressive enlargement of the aorta. It is usually asymptomatic and screening recommendation are limited. The gold-standard evaluation is performed by computed tomography angiography (CTA) and radiologists time-consuming assessment. Scans for other indications could help on this screening, however if acquired without contrast enhancement or with low dose protocol, it can make the clinical evaluation difficult, besides increasing the scans quantity for the radiologists. In this study, it was selected 587 unique CT scans including control and TAA patients, acquired with low and standard dose protocols, with or without contrast enhancement. A novel segmentation model, DeepVox, exhibited dice score coefficients of 0.932 and 0.897 for development and test sets, respectively, with faster training speed in comparison to models reported in the literature. The novel TAA classification model, SAVE-CT, presented accuracies of 0.930 and 0.922 for development and test sets, respectively, using only the binary segmentation mask from DeepVox as input, without hand-engineered features. These two models together are a potential approach for TAA screening, as they can handle variable number of slices as input, handling thoracic and thoracoabdominal sequences, in a fully automated contrast- and dose-independent evaluation. This may assist to decrease TAA mortality and prioritize the evaluation queue of patients for radiologists., Comment: 23 pages, 4 figures, 7 tables

A forward modelling approach provides simple, fast and realistic simulations of galaxy surveys, without a complex underlying model. For this purpose, galaxy clustering needs to be simulated accurately, both for the usage of clustering as its own probe and to control systematics. We present a forward model to simulate galaxy surveys, where we extend the Ultra-Fast Image Generator to include galaxy clustering. We use the distribution functions of the galaxy properties, derived from a forward model adjusted to observations. This population model jointly describes the luminosity functions, sizes, ellipticities, SEDs and apparent magnitudes. To simulate the positions of galaxies, we then use a two-parameter relation between galaxies and halos with Subhalo Abundance Matching (SHAM). We simulate the halos and subhalos using the fast PINOCCHIO code, and a method to extract the surviving subhalos from the merger history. Our simulations contain a red and a blue galaxy population, for which we build a SHAM model based on star formation quenching. For central galaxies, mass quenching is controlled with the parameter M$_{\mathrm{limit}}$, with blue galaxies residing in smaller halos. For satellite galaxies, environmental quenching is implemented with the parameter t$_{\mathrm{quench}}$, where blue galaxies occupy only recently merged subhalos. We build and test our model by comparing to imaging data from the Dark Energy Survey Year 1. To ensure completeness in our simulations, we consider the brightest galaxies with $i<20$. We find statistical agreement between our simulations and the data for two-point correlation functions on medium to large scales. Our model provides constraints on the two SHAM parameters M$_{\mathrm{limit}}$ and t$_{\mathrm{quench}}$ and offers great prospects for the quick generation of galaxy mock catalogues, optimized to agree with observations., Comment: Accepted for publication in JCAP. 33 pages, 17 figures

Breast cancer molecular subtypes classification plays an import role to sort patients with divergent prognosis. The biomarkers used are Estrogen Receptor (ER), Progesterone Receptor (PR), HER2, and Ki67. Based on these biomarkers expression levels, subtypes are classified as Luminal A (LA), Luminal B (LB), HER2 subtype, and Triple-Negative Breast Cancer (TNBC). Immunohistochemistry is used to classify subtypes, although interlaboratory and interobserver variations can affect its accuracy, besides being a time-consuming technique. The Fourier transform infrared micro-spectroscopy may be coupled with deep learning for cancer evaluation, where there is still a lack of studies for subtypes and biomarker levels prediction. This study presents a novel 2D deep learning approach to achieve these predictions. Sixty micro-FTIR images of 320x320 pixels were collected from a human breast biopsies microarray. Data were clustered by K-means, preprocessed and 32x32 patches were generated using a fully automated approach. CaReNet-V2, a novel convolutional neural network, was developed to classify breast cancer (CA) vs adjacent tissue (AT) and molecular subtypes, and to predict biomarkers level. The clustering method enabled to remove non-tissue pixels. Test accuracies for CA vs AT and subtype were above 0.84. The model enabled the prediction of ER, PR, and HER2 levels, where borderline values showed lower performance (minimum accuracy of 0.54). Ki67 percentage regression demonstrated a mean error of 3.6%. Thus, CaReNet-V2 is a potential technique for breast cancer biopsies evaluation, standing out as a screening analysis technique and helping to prioritize patients., Comment: 32 pages, 3 figures, 6 tables

Breast cancer treatment still remains a challenge, where molecular subtypes classification plays a crucial role in selecting appropriate and specific therapy. The four subtypes are Luminal A (LA), Luminal B (LB), HER2 subtype, and Triple-Negative Breast Cancer (TNBC). Immunohistochemistry is the gold-standard evaluation, although interobserver variations are reported and molecular signatures identification is time-consuming. Fourier transform infrared micro-spectroscopy with machine learning approaches have been used to evaluate cancer samples, presenting biochemical-related explainability. However, this explainability is harder when using deep learning. This study created a 1D deep learning tool for breast cancer subtype evaluation and biochemical contribution. Sixty hyperspectral images were acquired from a human breast cancer microarray. K-Means clustering was applied to select tissue and paraffin spectra. CaReNet-V1, a novel 1D convolutional neural network, was developed to classify breast cancer (CA) and adjacent tissue (AT), and molecular subtypes. A 1D adaptation of Grad-CAM was applied to assess the biochemical impact to the classifications. CaReNet-V1 effectively classified CA and AT (test accuracy of 0.89), as well as HER2 and TNBC subtypes (0.83 and 0.86), with greater difficulty for LA and LB (0.74 and 0.68). The model enabled the evaluation of the most contributing wavenumbers to the predictions, providing a direct relationship with the biochemical content. Therefore, CaReNet-V1 and hyperspectral images is a potential approach for breast cancer biopsies assessment, providing additional information to the pathology report. Biochemical content impact feature may be used for other studies, such as treatment efficacy evaluation and development new diagnostics and therapeutic methods., Comment: 23 pages, 5 figures, 2 tables

We propose SelfVC, a training strategy to iteratively improve a voice conversion model with self-synthesized examples. Previous efforts on voice conversion focus on factorizing speech into explicitly disentangled representations that separately encode speaker characteristics and linguistic content. However, disentangling speech representations to capture such attributes using task-specific loss terms can lead to information loss. In this work, instead of explicitly disentangling attributes with loss terms, we present a framework to train a controllable voice conversion model on entangled speech representations derived from self-supervised learning (SSL) and speaker verification models. First, we develop techniques to derive prosodic information from the audio signal and SSL representations to train predictive submodules in the synthesis model. Next, we propose a training strategy to iteratively improve the synthesis model for voice conversion, by creating a challenging training objective using self-synthesized examples. We demonstrate that incorporating such self-synthesized examples during training improves the speaker similarity of generated speech as compared to a baseline voice conversion model trained solely on heuristically perturbed inputs. Our framework is trained without any text and achieves state-of-the-art results in zero-shot voice conversion on metrics evaluating naturalness, speaker similarity, and intelligibility of synthesized audio., Comment: Accepted at ICML 2024

The Quantum Approximate Optimization Algorithm (QAOA) is an algorithm originally proposed to find approximate solutions to Combinatorial Optimization problems on quantum computers. However, the algorithm has also attracted interest for sampling purposes since it was theoretically demonstrated under reasonable complexity assumptions that one layer of the algorithm already engineers a probability distribution beyond what can be simulated by classical computers. In this regard, a recent study has shown as well that, in universal Ising models, this global probability distribution resembles pure but thermal-like distributions at a temperature that depends on internal correlations of the spin model. In this work, through an interferometric interpretation of the algorithm, we extend the theoretical derivation of the amplitudes of the eigenstates, and the Boltzmann distributions generated by single-layer QAOA. We also review the implications that this behavior has from both a practical and fundamental perspective., Comment: 20 pages, 6 figures

Objects' rigid motions in 3D space are described by rotations and translations of a highly-correlated set of points, each with associated $x,y,z$ coordinates that real-valued networks consider as separate entities, losing information. Previous works exploit quaternion algebra and their ability to model rotations in 3D space. However, these algebras do not properly encode translations, leading to sub-optimal performance in 3D learning tasks. To overcome these limitations, we employ a dual quaternion representation of rigid motions in the 3D space that jointly describes rotations and translations of point sets, processing each of the points as a single entity. Our approach is translation and rotation equivariant, so it does not suffer from shifts in the data and better learns object trajectories, as we validate in the experimental evaluations. Models endowed with this formulation outperform previous approaches in a human pose forecasting application, attesting to the effectiveness of the proposed dual quaternion formulation for rigid motions in 3D space., Comment: Accepted at IEEE MLSP 2023 (Honorable Mention Top 10% Outstanding Paper)