Search

We develop a new framework for establishing approximate factorization of entropy on arbitrary probability spaces, using a geometric notion known as non-negative sectional curvature. The resulting estimates are equivalent to entropy subadditivity and generalized Brascamp-Lieb inequalities, and provide a sharp modified log-Sobolev inequality for the Gibbs sampler of several particle systems in both continuous and discrete settings. The method allows us to obtain simple proofs of known results, as well as some new inequalities. We illustrate this through various applications, including discrete Gaussian free fields on arbitrary networks, the down-up walk on uniform $n$-sets, the uniform measure over permutations, and the uniform measure on the unit sphere in $\R^n$. Our method also yields a simple, coupling-based proof of the celebrated logarithmic Sobolev inequality for Langevin diffusions in a convex potential, which is one of the most emblematic applications of the Bakry-\'Emery criterion., Comment: additional references

We discuss the relaxation time (inverse spectral gap) of the one dimensional $O(N)$ model, for all $N$ and with two types of boundary conditions. We see how its low temperature asymptotic behavior is affected by the topology. The combination of the space dimension, which here is always 1, the boundary condition (free or periodic), and the spin state $S^{N-1}$, determines the existence or absence of non-trivial homotopy classes in some discrete version. Such non-trivial topology reflects in bottlenecks of the dynamics, creating metastable states that the system exits at exponential times; while when only one homotopy class exists the relaxation time depends polynomially on the temperature. We prove in the one dimensional case that, indeed, the relaxation time is a proxy to the model's topological properties via the exponential/polynomial dependence on the temperature., Comment: 26 pages. Comments welcome!

The precise reconstruction of Compton-scatter events is paramount for an imaging medium-energy gamma-ray telescope. The proposed AMEGO-X is enabled by a silicon tracker utilizing AstroPix chips - a pixelated silicon HVCMOS sensor novel for space use. To achieve science goals, each 500 x 500 $\mu$m$^2$ pixel must be sensitive for energy deposits ranging from 25 - 700 keV with an energy resolution of 5 keV at 122 keV (< 10%). This is achieved through depletion of the 500 $\mu$m thick sensor, although complete depletion poses an engineering and design challenge. This work will summarize the current status of depletion measurements highlighting direct measurement with TCT laser scanning and the agreement with simulation. Future plans for further testing will also be identified., Comment: 13 pages, proceedings, SPIE Astronomical Telescopes + Instrumentation 2024, June 16-21 2024

Abstraction is the process of extracting the essential features from raw data while ignoring irrelevant details. This is similar to the process of focusing on large-scale properties, systematically removing irrelevant small-scale details, implemented in the renormalisation group of statistical physics. This analogy is suggestive because the fixed points of the renormalisation group offer an ideal candidate of a truly abstract -- i.e. data independent -- representation. It has been observed that abstraction emerges with depth in neural networks. Deep layers of neural network capture abstract characteristics of data, such as "cat-ness" or "dog-ness" in images, by combining the lower level features encoded in shallow layers (e.g. edges). Yet we argue that depth alone is not enough to develop truly abstract representations. We advocate that the level of abstraction crucially depends on how broad the training set is. We address the issue within a renormalisation group approach where a representation is expanded to encompass a broader set of data. We take the unique fixed point of this transformation -- the Hierarchical Feature Model -- as a candidate for an abstract representation. This theoretical picture is tested in numerical experiments based on Deep Belief Networks trained on data of different breadth. These show that representations in deep layers of neural networks approach the Hierarchical Feature Model as the data gets broader, in agreement with theoretical predictions., Comment: 28 pages, 7 figures

Neutrino emission in cold neutron stars is dominated by the modified urca (murca) process and nucleon-nucleon bremsstrahlung. The standard emission rates were provided by Friman and Maxwell in 1979, effectively based on a chiral Lagrangian framework with pion and rho meson exchange, supplemented by Landau parameters to describe short-range interactions. We reevaluate these rates within the same framework, correcting several errors and removing unnecessary simplifications, notably the triangular approximation - where the Fermi momenta of protons and leptons negligible compared to that of neutrons - in MURCA, and quantify their importance. The impact of rho meson exchange, previously argued to cancel with interference effects, is actually quite relevant. Altogether, the cooling rates are reduced by as much as a factor 2. We provide comprehensive analytical formulas encompassing all contributions, designed for straightforward numerical implementation. Our results are particularly relevant for studies of physics beyond the standard model, where the emission of new particles - such as axions - is typically computed within the same framework we adopt here., Comment: 24+3 pages, 9 Figures, Matching to the version accepted for publication on JCAP

The increasing availability of biomedical data is helping to design more robust deep learning (DL) algorithms to analyze biomedical samples. Currently, one of the main limitations to train DL algorithms to perform a specific task is the need for medical experts to label data. Automatic methods to label data exist, however automatic labels can be noisy and it is not completely clear when automatic labels can be adopted to train DL models. This paper aims to investigate under which circumstances automatic labels can be adopted to train a DL model on the classification of Whole Slide Images (WSI). The analysis involves multiple architectures, such as Convolutional Neural Networks (CNN) and Vision Transformer (ViT), and over 10000 WSIs, collected from three use cases: celiac disease, lung cancer and colon cancer, which one including respectively binary, multiclass and multilabel data. The results allow identifying 10% as the percentage of noisy labels that lead to train competitive models for the classification of WSIs. Therefore, an algorithm generating automatic labels needs to fit this criterion to be adopted. The application of the Semantic Knowledge Extractor Tool (SKET) algorithm to generate automatic labels leads to performance comparable to the one obtained with manual labels, since it generates a percentage of noisy labels between 2-5%. Automatic labels are as effective as manual ones, reaching solid performance comparable to the one obtained training models with manual labels., Comment: pre-print of the journal paper

Mesh-based scene representation offers a promising direction for simplifying large-scale hierarchical visual localization pipelines, combining a visual place recognition step based on global features (retrieval) and a visual localization step based on local features. While existing work demonstrates the viability of meshes for visual localization, the impact of using synthetic databases rendered from them in visual place recognition remains largely unexplored. In this work we investigate using dense 3D textured meshes for large-scale Visual Place Recognition (VPR). We identify a significant performance drop when using synthetic mesh-based image databases compared to real-world images for retrieval. To address this, we propose MeshVPR, a novel VPR pipeline that utilizes a lightweight features alignment framework to bridge the gap between real-world and synthetic domains. MeshVPR leverages pre-trained VPR models and is efficient and scalable for city-wide deployments. We introduce novel datasets with freely available 3D meshes and manually collected queries from Berlin, Paris, and Melbourne. Extensive evaluations demonstrate that MeshVPR achieves competitive performance with standard VPR pipelines, paving the way for mesh-based localization systems. Data, code, and interactive visualizations are available at https://meshvpr.github.io/, Comment: Website: https://mesh-vpr.github.io/

The autofluorescence of calabrian Opuntia ficus-indica bioactive extract upon UV illumination is explored by fluorescence spectroscopy enabling to investigate the typology and distribution of responsible molecules within green cladodes. The spectroscopic analysis of the extract shows a significative red emission, suggesting an abundance of chlorophylls in the sample. Such molecules show pronounced fluorescence in the visible range (400-800 nm) with a very large Stokes shift, when excited with UV light source. The fluorescence profiling is performed also in the case of polymers, such as poly(methyl methacrylate) (PMMA), poly(vinylpyrrolidone) (PVP) and polyvinyl alcohol (PVA), enriched with OFI extract with a signal improvement up to 40 times greater than the extract alone. This by-product fluorescent molecule can replace commercial dyes into several applications spanning from nano-optics to anti-counterfeiting and bioimaging.

Federated Learning (FL) methods often struggle in highly statistically heterogeneous settings. Indeed, non-IID data distributions cause client drift and biased local solutions, particularly pronounced in the final classification layer, negatively impacting convergence speed and accuracy. To address this issue, we introduce Federated Recursive Ridge Regression (Fed3R). Our method fits a Ridge Regression classifier computed in closed form leveraging pre-trained features. Fed3R is immune to statistical heterogeneity and is invariant to the sampling order of the clients. Therefore, it proves particularly effective in cross-device scenarios. Furthermore, it is fast and efficient in terms of communication and computation costs, requiring up to two orders of magnitude fewer resources than the competitors. Finally, we propose to leverage the Fed3R parameters as an initialization for a softmax classifier and subsequently fine-tune the model using any FL algorithm (Fed3R with Fine-Tuning, Fed3R+FT). Our findings also indicate that maintaining a fixed classifier aids in stabilizing the training and learning more discriminative features in cross-device settings. Official website: https://fed-3r.github.io/., Comment: Accepted at ICML 2024 - https://fed-3r.github.io/

ComPair is a prototype gamma-ray telescope for the development of key technologies for next-generation gamma-ray detectors consisting of four subsystems: a 10-layer double-sided silicon strip detector tracker, a cadmium zinc telluride calorimeter, a cesium iodide calorimeter, and a plastic anti-coincidence detector (ACD). The ACD acts as an active shield to veto charged particle events and consists of 5 plastic scintillating panels. ComPair was launched as a balloon payload from Ft. Sumner, New Mexico and completed a 6-hour flight on August 27, 2023. Here we detail the design adn calibration of the ComPair ACD, and report on the ACD's veto efficiency and other performance metrics during the ComPair fight., Comment: Proceedings for SPIE Astronomical Telescopes + Instrumentation 2024; June 16-21 2024; Yokohama, Japan

We consider resonant wave-like dark matter conversion into low-frequency radio waves in the Earth's ionosphere. Resonant conversion occurs when the dark matter mass and the plasma frequency coincide, defining a range $m_{ \text{DM} } \sim 10^{-9} - 10^{-8}$ eV where this approach is best suited. Owing to the non-relativistic nature of dark matter and the typical variational scale of the Earth's ionosphere, the standard linearized approach to computing dark matter conversion is not suitable. We therefore solve a second-order boundary-value problem, effectively framing the ionosphere as a driven cavity filled with a positionally-varying plasma. An electrically-small dipole antenna targeting the generated radio waves can be orders of magnitude more sensitive to dark photon and axion-like particle dark matter in the relevant mass range. The present study opens up a promising way of testing hitherto unexplored parameter space which could be further improved with a dedicated instrument., Comment: 11 pages, 4 figures

The ComPair gamma-ray telescope is a technology demonstrator for a future gamma-ray telescope called the All-sky Medium Energy Gamma-ray Observatory (AMEGO). The instrument is composed of four subsystems, a double-sided silicon strip detector, a virtual Frisch grid CdZnTe calorimeter, a CsI:Tl based calorimeter, and an anti-coincidence detector (ACD). The CsI calorimeter's goal is to measure the position and energy deposited from high-energy events. To demonstrate the technological readiness, the calorimeter has flown onboard a NASA scientific balloon as part of the GRAPE-ComPair mission and accumulated around 3 hours of float time at an altitude of 40 km. During the flight, the CsI calorimeter observed background radiation, Regener-Pfotzer Maximum, and several gamma-ray activation lines originating from aluminum.

Precise, pixel-wise geolocalization of astronaut photography is critical to unlocking the potential of this unique type of remotely sensed Earth data, particularly for its use in disaster management and climate change research. Recent works have established the Astronaut Photography Localization task, but have either proved too costly for mass deployment or generated too coarse a localization. Thus, we present EarthMatch, an iterative homography estimation method that produces fine-grained localization of astronaut photographs while maintaining an emphasis on speed. We refocus the astronaut photography benchmark, AIMS, on the geolocalization task itself, and prove our method's efficacy on this dataset. In addition, we offer a new, fair method for image matcher comparison, and an extensive evaluation of different matching models within our localization pipeline. Our method will enable fast and accurate localization of the 4.5 million and growing collection of astronaut photography of Earth. Webpage with code and data at https://earthloc-and-earthmatch.github.io, Comment: CVPR 2024 IMW - webpage: https://earthloc-and-earthmatch.github.io

We introduce the DREAMS project, an innovative approach to understanding the astrophysical implications of alternative dark matter models and their effects on galaxy formation and evolution. The DREAMS project will ultimately comprise thousands of cosmological hydrodynamic simulations that simultaneously vary over dark matter physics, astrophysics, and cosmology in modeling a range of systems -- from galaxy clusters to ultra-faint satellites. Such extensive simulation suites can provide adequate training sets for machine-learning-based analyses. This paper introduces two new cosmological hydrodynamical suites of Warm Dark Matter, each comprised of 1024 simulations generated using the Arepo code. One suite consists of uniform-box simulations covering a $(25~h^{-1}~{\rm M}_\odot)^3$ volume, while the other consists of Milky Way zoom-ins with sufficient resolution to capture the properties of classical satellites. For each simulation, the Warm Dark Matter particle mass is varied along with the initial density field and several parameters controlling the strength of baryonic feedback within the IllustrisTNG model. We provide two examples, separately utilizing emulators and Convolutional Neural Networks, to demonstrate how such simulation suites can be used to disentangle the effects of dark matter and baryonic physics on galactic properties. The DREAMS project can be extended further to include different dark matter models, galaxy formation physics, and astrophysical targets. In this way, it will provide an unparalleled opportunity to characterize uncertainties on predictions for small-scale observables, leading to robust predictions for testing the particle physics nature of dark matter on these scales., Comment: 28 pages, 8 figures, DREAMS website: https://www.dreams-project.org

Pose refinement is an interesting and practically relevant research direction. Pose refinement can be used to (1) obtain a more accurate pose estimate from an initial prior (e.g., from retrieval), (2) as pre-processing, i.e., to provide a better starting point to a more expensive pose estimator, (3) as post-processing of a more accurate localizer. Existing approaches focus on learning features / scene representations for the pose refinement task. This involves training an implicit scene representation or learning features while optimizing a camera pose-based loss. A natural question is whether training specific features / representations is truly necessary or whether similar results can be already achieved with more generic features. In this work, we present a simple approach that combines pre-trained features with a particle filter and a renderable representation of the scene. Despite its simplicity, it achieves state-of-the-art results, demonstrating that one can easily build a pose refiner without the need for specific training. The code is at https://github.com/ga1i13o/mcloc_poseref, Comment: Accepted to CVPR2024 (Highlight)

Graphene has been widely studied for various applications due to its outstanding electrical and mechanical properties. However, its potential in thermoelectric applications has been limited by a low Seebeck coefficient and high thermal conductivity. Efforts to enhance its thermoelectric properties have involved the usage of carbon-based nanoribbons, strain engineering, and heteroatom co-doping, particularly with Nitrogen and/or Boron atoms. In this work, multiscale simulation approaches combining DFT calculations and semi-empirical models are used to explore the potential improvement of the thermoelectric properties via borazine (B$_3$N$_3$)-ring doping. As bandgap engineering can be obtained with this doping, the thermoelectric properties of graphene are significantly enlarged, albeit at the cost of reduced conductance. The effects observed are not only dependent on the concentration of BN within the graphene lattice but are also notably influenced by the relative rotational alignment of the BN rings. Furthermore, the effect of the distribution and rotational disorder are considered, showing reduced electronic conductance compared to the periodic case highlighting the importance of precise control over the doping parameters of BN-ring. Lastly, the thermal lattice conductance is estimated revealing a substantial reduction of up to 40$\%$ compared to pristine graphene opening up the possibility of enhancing the thermoelectric efficiency of BNC materials. The present theoretical approach highlights how BN-ring doping can refine the thermoelectric properties of 2D graphene, offering a pathway for enhancing its suitability in practical thermoelectric applications.

Visual Place Recognition aims at recognizing previously visited places by relying on visual clues, and it is used in robotics applications for SLAM and localization. Since typically a mobile robot has access to a continuous stream of frames, this task is naturally cast as a sequence-to-sequence localization problem. Nevertheless, obtaining sequences of labelled data is much more expensive than collecting isolated images, which can be done in an automated way with little supervision. As a mitigation to this problem, we propose a novel Joint Image and Sequence Training protocol (JIST) that leverages large uncurated sets of images through a multi-task learning framework. With JIST we also introduce SeqGeM, an aggregation layer that revisits the popular GeM pooling to produce a single robust and compact embedding from a sequence of single-frame embeddings. We show that our model is able to outperform previous state of the art while being faster, using 8 times smaller descriptors, having a lighter architecture and allowing to process sequences of various lengths. Code is available at https://github.com/ga1i13o/JIST

We report a theoretical study of dielectric properties of models of amorphous Boron Nitride, using interatomic potentials generated by machine learning. We first perform first-principles simulations on small (about $100$ atoms in the periodic cell) sample sizes to explore the emergence of mid-gap states and its correlation with structural features. Next, by using a simplified tight-binding electronic model, we analyse the dielectric functions for complex three dimensional models (containing about $10.000$ atoms) embedding varying concentrations of ${\rm sp^{1}, sp^{2}}$ and ${\rm sp^3}$ bonds between B and N atoms. Within the limits of these methodologies, the resulting value of the zero-frequency dielectric constant is shown to be influenced by the population density of such mid-gap states and their localization characteristics. We observe nontrivial correlations between the structure-induced electronic fluctuations and the resulting dielectric constant values. Our findings are however just a first step in the quest of accessing fully accurate dielectric properties of as-grown amorphous BN of relevance for interconnect technologies and beyond., Comment: 27 pages, 10 figures

Astronaut photography, spanning six decades of human spaceflight, presents a unique Earth observations dataset with immense value for both scientific research and disaster response. Despite its significance, accurately localizing the geographical extent of these images, crucial for effective utilization, poses substantial challenges. Current manual localization efforts are time-consuming, motivating the need for automated solutions. We propose a novel approach - leveraging image retrieval - to address this challenge efficiently. We introduce innovative training techniques, including Year-Wise Data Augmentation and a Neutral-Aware Multi-Similarity Loss, which contribute to the development of a high-performance model, EarthLoc. We develop six evaluation datasets and perform a comprehensive benchmark comparing EarthLoc to existing methods, showcasing its superior efficiency and accuracy. Our approach marks a significant advancement in automating the localization of astronaut photography, which will help bridge a critical gap in Earth observations data. Code and datasets are available at https://github.com/gmberton/EarthLoc, Comment: CVPR 2024

Quantum processors require a signal-delivery architecture with high addressability (low crosstalk) to ensure high performance already at the scale of dozens of qubits. Signal crosstalk causes inadvertent driving of quantum gates, which will adversely affect quantum-gate fidelities in scaled-up devices. Here, we demonstrate packaged flip-chip superconducting quantum processors with signal-crosstalk performance competitive with those reported in other platforms. For capacitively coupled qubit-drive lines, we find on-resonant crosstalk better than -27 dB (average -37 dB). For inductively coupled magnetic-flux-drive lines, we find less than 0.13 % direct-current flux crosstalk (average 0.05 %). These observed crosstalk levels are adequately small and indicate a decreasing trend with increasing distance, which is promising for further scaling up to larger numbers of qubits. We discuss the implication of our results for the design of a low-crosstalk, on-chip signal delivery architecture, including the influence of a shielding tunnel structure, potential sources of crosstalk, and estimation of crosstalk-induced qubit-gate error in scaled-up quantum processors., Comment: 21 pages, 16 figures, includes appendices, updated discussion on source of xy crosstalk, flux stability, etc

The goal of this paper is to continue the study of the relation between the Poincar\'e inequality and the lower bounds of Minkowski content of separating sets, initiated in our previous work [Caputo, Cavallucci: Poincar\'e inequality and energy of separating sets, arXiv 2401.02762]. A new shorter proof is provided. An intermediate tool is the study of the lower bound of another geometric quantity, called separating ratio. The main novelty is the description of the relation between the infima of the separating ratio and the Minkowski content of separating sets. We prove a quantitative comparison between the two infima in the local quasigeodesic case and equality in the local geodesic one. No Poincar\'e assumption is needed to prove it. The main tool employed in the proof is a new function, called the position function, which allows in a certain sense to fibrate a set in boundaries of separating sets. We also extend the proof to measure graphs, where due to the combinatorial nature of the problem, the approach is more intuitive. In the appendix, we revise some classical characterizations of the p-Poincar\'e inequality, by proving along the way equivalence with a notion of p-pencil that extends naturally the definition for p = 1., Comment: 38 pages, 4 figures

Many practical applications require training of semantic segmentation models on unlabelled datasets and their execution on low-resource hardware. Distillation from a trained source model may represent a solution for the first but does not account for the different distribution of the training data. Unsupervised domain adaptation (UDA) techniques claim to solve the domain shift, but in most cases assume the availability of the source data or an accessible white-box source model, which in practical applications are often unavailable for commercial and/or safety reasons. In this paper, we investigate a more challenging setting in which a lightweight model has to be trained on a target unlabelled dataset for semantic segmentation, under the assumption that we have access only to black-box source model predictions. Our method, named CoRTe, consists of (i) a pseudo-labelling function that extracts reliable knowledge from the black-box source model using its relative confidence, (ii) a pseudo label refinement method to retain and enhance the novel information learned by the student model on the target data, and (iii) a consistent training of the model using the extracted pseudo labels. We benchmark CoRTe on two synthetic-to-real settings, demonstrating remarkable results when using black-box models to transfer knowledge on lightweight models for a target data distribution., Comment: 11 pages, 6 figures, ICCV2023 workshop