Search

Traditional implicit generative models are capable of learning highly complex data distributions. However, their training involves distinguishing real data from synthetically generated data using adversarial discriminators, which can lead to unstable training dynamics and mode dropping issues. In this work, we build on the \textit{invariant statistical loss} (ISL) method introduced in \cite{de2024training}, and extend it to handle heavy-tailed and multivariate data distributions. The data generated by many real-world phenomena can only be properly characterised using heavy-tailed probability distributions, and traditional implicit methods struggle to effectively capture their asymptotic behavior. To address this problem, we introduce a generator trained with ISL, that uses input noise from a generalised Pareto distribution (GPD). We refer to this generative scheme as Pareto-ISL for conciseness. Our experiments demonstrate that Pareto-ISL accurately models the tails of the distributions while still effectively capturing their central characteristics. The original ISL function was conceived for 1D data sets. When the actual data is $n$-dimensional, a straightforward extension of the method was obtained by targeting the $n$ marginal distributions of the data. This approach is computationally infeasible and ineffective in high-dimensional spaces. To overcome this, we extend the 1D approach using random projections and define a new loss function suited for multivariate data, keeping problems tractable by adjusting the number of projections. We assess its performance in multidimensional generative modeling and explore its potential as a pretraining technique for generative adversarial networks (GANs) to prevent mode collapse, reporting promising results and highlighting its robustness across various hyperparameter settings.

The design of Systems on Chips (SoCs) is becoming more and more complex due to technological advancements. Missed bugs can cause drastic failures in safety-critical environments leading to the endangerment of lives. To overcome these drastic failures, formal property verification (FPV) has been applied in the industry. However, there exist multiple hardware designs where the results of FPV are not conclusive even for long runtimes of model-checking tools. For this reason, the use of High-level Equivalence Checking (HLEC) tools has been proposed in the last few years. However, the procedure for how to use it inside an industrial toolchain has not been defined. For this reason, we proposed an automated methodology based on metamodeling techniques which consist of two main steps. First, an untimed algorithmic description written in C++ is verified in an early stage using generated assertions; the advantage of this step is that the assertions at the software level run in seconds and we can start our analysis with conclusive results about our algorithm before starting to write the RTL (Register Transfer Level) design. Second, this algorithmic description is verified against its sequential design using HLEC and the respective metamodel parameters. The results show that the presented methodology can find bugs early related to the algorithmic description and prepare the setup for the HLEC verification. This helps to reduce the verification efforts to set up the tool and write the properties manually which is always error-prone. The proposed framework can help teams working on datapaths to verify and make decisions in an early stage of the verification flow., Comment: Published in DVCon Europe 2024

Random feature latent variable models (RFLVMs) represent the state-of-the-art in latent variable models, capable of handling non-Gaussian likelihoods and effectively uncovering patterns in high-dimensional data. However, their heavy reliance on Monte Carlo sampling results in scalability issues which makes it difficult to use these models for datasets with a massive number of observations. To scale up RFLVMs, we turn to the optimization-based variational Bayesian inference (VBI) algorithm which is known for its scalability compared to sampling-based methods. However, implementing VBI for RFLVMs poses challenges, such as the lack of explicit probability distribution functions (PDFs) for the Dirichlet process (DP) in the kernel learning component, and the incompatibility of existing VBI algorithms with RFLVMs. To address these issues, we introduce a stick-breaking construction for DP to obtain an explicit PDF and a novel VBI algorithm called ``block coordinate descent variational inference" (BCD-VI). This enables the development of a scalable version of RFLVMs, or in short, SRFLVM. Our proposed method shows scalability, computational efficiency, superior performance in generating informative latent representations and the ability of imputing missing data across various real-world datasets, outperforming state-of-the-art competitors.

Registers in IP blocks of an SoC perform a variety of functions, most of which are essential to the SoC operation. The complexity of register implementation is relatively low when compared with other design blocks. However, the extensive number of registers, combined with the various potential functions they can perform, necessitates considerable effort during implementation, especially when using a manual approach. Therefore, an in-house register generator was proposed by the design team to reduce the manual effort in the register implementation. This in-house register generator supports not only the generation of register blocks but also bus-related blocks. Meanwhile, to support various requirements, 41 generation options are used for this generator, which is highly-configurable. From the verification perspective, it is infeasible to achieve complete verification results with a manual approach for all options combinations. Besides the complexity caused by configurability, the register verification is still time-consuming due to two widely recognized issues: the unreliability of specifications and the complexity arising from diverse access policies. To deal with the highly-configurable feature and both register verification issues, we propose an automated register verification framework using formal methods following the Model Driven Architecture (MDA). Based on our results, the human effort in the register verification can be reduced significantly, from 20Person-Day (20PD) to 3PD for each configuration, and 100\% code coverage can be achieved. During the project execution, eleven new design bugs were found with the proposed verification framework., Comment: Published in DVCon US 2024

Despite significant advancements in deep probabilistic models, learning low-dimensional discrete latent representations remains a challenging task. In this paper, we introduce a novel method that enhances variational inference in discrete latent variable models by leveraging Error Correcting Codes (ECCs) to introduce redundancy in the latent representations. This redundancy is then exploited by the variational posterior to yield more accurate estimates, thereby narrowing the variational gap. Inspired by ECCs commonly used in digital communications and data storage, we demonstrate proof-of-concept using a Discrete Variational Autoencoder (DVAE) with binary latent variables and block repetition codes. We further extend this idea to a hierarchical structure based on polar codes, where certain latent bits are more robustly protected. Our method improves generation quality, data reconstruction, and uncertainty calibration compared to the uncoded DVAE, even when trained with tighter bounds such as the Importance Weighted Autoencoder (IWAE) objective. In particular, we demonstrate superior performance on MNIST, FMNIST, CIFAR10, and Tiny ImageNet datasets. The general approach of integrating ECCs into variational inference is compatible with existing techniques to boost variational inference, such as importance sampling or Hamiltonian Monte Carlo. We also outline the key properties ECCs must have to effectively enhance discrete variational inference.

Reaction-Diffusion systems arise in diverse areas of science and engineering. Due to the peculiar characteristics of such equations, analytic solutions are usually not available and numerical methods are the main tools for approximating the solutions. In the last decade, artificial neural networks have become an active area of development for solving partial differential equations. However, several challenges remain unresolved with these methods when applied to reaction-diffusion equations. In this work, we focus on two main problems. The implementation of homogeneous Neumann boundary conditions and long-time integrations. For the homogeneous Neumann boundary conditions, we explore four different neural network methods based on the PINN approach. For the long time integration in Reaction-Diffusion systems, we propose a domain splitting method in time and provide detailed comparisons between different implementations of no-flux boundary conditions. We show that the domain splitting method is crucial in the neural network approach, for long time integration in Reaction-Diffusion systems. We demonstrate numerically that domain splitting is essential for avoiding local minima, and the use of different boundary conditions further enhances the splitting technique by improving numerical approximations. To validate the proposed methods, we provide numerical examples for the Diffusion, the Bistable and the Barkley equations and provide a detailed discussion and comparisons of the proposed methods., Comment: 35 pages, 12 figures, research paper

Large Language Models (LLMs) are prone to generating factually incorrect information when responding to queries that involve numerical and statistical data or other timely facts. In this paper, we present an approach for enhancing the accuracy of LLMs by integrating them with Data Commons, a vast, open-source repository of public statistics from trusted organizations like the United Nations (UN), Center for Disease Control and Prevention (CDC) and global census bureaus. We explore two primary methods: Retrieval Interleaved Generation (RIG), where the LLM is trained to produce natural language queries to retrieve data from Data Commons, and Retrieval Augmented Generation (RAG), where relevant data tables are fetched from Data Commons and used to augment the LLM's prompt. We evaluate these methods on a diverse set of queries, demonstrating their effectiveness in improving the factual accuracy of LLM outputs. Our work represents an early step towards building more trustworthy and reliable LLMs that are grounded in verifiable statistical data and capable of complex factual reasoning., Comment: 39 pages - 25 page paper, 14 page Appendix, 7 figures, 9 tables

The transition matrix, frequently abbreviated as T-matrix, contains the complete information in a linear approximation of how a spatially localized object scatters an incident field. The T-matrix is used to study the scattering response of an isolated object and describes the optical response of complex photonic materials made from ensembles of individual objects. T-matrices of certain common structures, potentially, have been repeatedly calculated all over the world again and again. This is not necessary and constitutes a major challenge for various reasons. First, the resources spent on their computation represent an unsustainable financial and ecological burden. Second, with the onset of machine learning, data is the gold of our era, and it should be freely available to everybody to address novel scientific challenges. Finally, the possibility of reproducing simulations could tremendously improve if the considered T-matrices could be shared. To address these challenges, we found it important to agree on a common data format for T-matrices and to enable their collection from different sources and distribution. This document aims to develop the specifications for storing T-matrices and associated metadata. The specifications should allow maximum freedom to accommodate as many use cases as possible without introducing any ambiguity in the stored data. The common format will assist in setting up a public database of T-matrices., Comment: Submitted to the Journal of Quantitative Spectroscopy and Radiative Transfer

We study isometric immersions $f: M^n \rightarrow \mathbb{H}^{n+1}$ into hyperbolic space of dimension $n+1$ of a complete Riemannian manifold of dimension $n$ on which a compact connected group of intrinsic isometries acts with principal orbits of codimension one. We provide a characterization if either $n \geq 3$ and $M^n$ is compact, or $n \geq 5$ and the connected components of the set where the sectional curvature is constant and equal to $-1$ are bounded.

We study the onset of collective spin-self-organization in a thermal ensemble of driven two-level atoms confined in an optical cavity. The atoms spontaneously form a spin-pattern above a critical driving strength that sets a threshold and is determined by the cavity parameters, the initial temperature, and the transition frequency of the atomic spin. Remarkably, we find that inhomogeneous Doppler broadening facilitates the onset of spin-self-organization. In particular, the threshold is non-monotonic when increasing the spin transition frequency and reaches a minimum when the Doppler broadening is of similar magnitude. This feature emerges due to Doppler-induced resonances. Above the threshold, we find cooperative dynamics of spin, spatial, and momentum degrees of freedom leading to density modulations, fast reduction of kinetic energy, and the emergence of non-thermal states. More broadly, our work demonstrates how broadening can facilitate strong light-matter interactions in many-body systems., Comment: 7+7 pages, 3 figures

Quantum spin models with a large number of interaction partners per spin are frequently used to describe modern many-body quantum optical systems like arrays of Rydberg atoms, atom-cavity systems or trapped ion crystals. For theoretical analysis the mean-field (MF) ansatz is routinely applied. However, besides special cases of all-to-all or strong long range interactions, the MF ansatz provides only approximate results. Here we present a systematic correction to MF theory based on diagrammatic perturbation theory for quantum spin correlators in thermal equilibrium. Our analytic results are universally applicable for any lattice geometry and spin-length S. We provide pre-computed and easy-to-use building blocks for Ising, Heisenberg and transverse field Ising models in the symmetry-unbroken regime. We showcase the quality and simplicity of the method by computing magnetic phase boundaries and excitations gaps. We also treat the Dicke-Ising model of ground-state superradiance where we show that corrections to the MF phase boundary vanish., Comment: corrected Eq. (A1) and minor improvements, close to version accepted in Phys. Rev. A

Many reaction-diffusion systems in various applications exhibit traveling wave solutions that evolve on multiple spatio-temporal scales. These traveling wave solutions are crucial for understanding the underlying dynamics of the system. In this work, we present sixth-order weighted essentially non-oscillatory (WENO) methods within the finite difference framework to solve reaction-diffusion systems. The WENO method allows us to use fewer grid points and larger time steps compared to classical finite difference methods. Our focus is on solving the reaction-diffusion system for the traveling wave solution with the sharp front. Although the WENO method is popular for hyperbolic conservation laws, especially for problems with discontinuity, it can be adapted for the equations of parabolic type, such as reaction-diffusion systems, to effectively handle sharp wave fronts. Thus, we employed the WENO methods specifically developed for equations of parabolic type. We considered various reaction-diffusion equations, including Fisher's, Zeldovich, Newell-Whitehead-Segel, bistable equations, and the Lotka-Volterra competition-diffusion system, all of which yield traveling wave solutions with sharp wave fronts. Numerical examples in this work demonstrate that the central WENO method is highly more accurate and efficient than the commonly used finite difference method. We also provide an analysis related to the numerical speed of the sharp propagating front in the Newell-Whitehead-Segel equation. The overall results confirm that the central WENO method is highly efficient and is recommended for solving reaction-diffusion equations with sharp wave fronts.

Semi-Dirac materials in 2D present an anisotropic dispersion relation, linear along one direction and quadratic along the perpendicular one. This study explores the topological properties and the influence of disorder in a 2D semi-Dirac Hamiltonian. Anisotropic edge states appear only in one direction. Their topological protection can be rigorously founded on the Zak phase of the one-dimensional reduction of the semi-Dirac Hamiltonian, parametrically depending on one of the momenta. In general, only a single value of the momentum is topologically protected so these systems can be considered as high order momentum topological insulators. We explore the dependence on the disorder of the edge states and the robustness of the topological protection in these materials. We also explore the consequences of the high order topological protection in momentum space for the transport properties in a two-terminal configuration., Comment: 10 pages, 10 figures

Mitigating biases in generative AI and, particularly in text-to-image models, is of high importance given their growing implications in society. The biased datasets used for training pose challenges in ensuring the responsible development of these models, and mitigation through hard prompting or embedding alteration, are the most common present solutions. Our work introduces a novel approach to achieve diverse and inclusive synthetic images by learning a direction in the latent space and solely modifying the initial Gaussian noise provided for the diffusion process. Maintaining a neutral prompt and untouched embeddings, this approach successfully adapts to diverse debiasing scenarios, such as geographical biases. Moreover, our work proves it is possible to linearly combine these learned latent directions to introduce new mitigations, and if desired, integrate it with text embedding adjustments. Furthermore, text-to-image models lack transparency for assessing bias in outputs, unless visually inspected. Thus, we provide a tool to empower developers to select their desired concepts to mitigate. The project page with code is available online., Comment: Accepted at CVPR workshop 2024, proceedings of ReGenAI: First Workshop on Responsible Generative AI

Partial differential equations (PDEs) with multiple scales or those defined over sufficiently large domains arise in various areas of science and engineering and often present problems when approximating the solutions numerically. Machine learning techniques are a relatively recent method for solving PDEs. Despite the increasing number of machine learning strategies developed to approximate PDEs, many remain focused on relatively small domains. When scaling the equations, a large domain is naturally obtained, especially when the solution exhibits multiscale characteristics. This study examines two-scale equations whose solution structures exhibit distinct characteristics: highly localized in some regions and significantly flat in others. These two regions must be adequately addressed over a large domain to approximate the solution more accurately. We focus on the vanishing gradient problem given by the diminishing gradient zone of the activation function over large domains and propose a stratified sampling algorithm to address this problem. We compare the uniform random classical sampling method over the entire domain and the proposed stratified sampling method. The numerical results confirm that the proposed method yields more accurate and consistent solutions than classical methods., Comment: 35 pages, 14 figures

Dissipative light-matter systems can display emergent collective behavior. Here, we report a $\mathbb{Z}_2$-symmetry-breaking phase transition in a system of multi-level $^{87}$Rb atoms strongly coupled to a weakly driven two-mode optical cavity. In the symmetry-broken phase, non-ergodic dynamics manifests in the emergence of multiple stationary states with disjoint basins of attraction. This feature enables the amplification of a small atomic population imbalance into a characteristic macroscopic cavity transmission signal. Our experiment does not only showcase strongly dissipative atom-cavity systems as platforms for probing non-trivial collective many-body phenomena, but also highlights their potential for hosting technological applications in the context of sensing, density classification, and pattern retrieval dynamics within associative memories.

Maxwell's equations are solved when the amplitude and phase of the electromagnetic field are determined at all points in space. Generally, the Stokes parameters can only capture the amplitude and polarization state of the electromagnetic field in the radiation (far) zone. Therefore, the measurement of the Stokes parameters is, in general, insufficient to solve Maxwell's equations. In this Letter, we solve Maxwell's equations for a set of objects widely used in Nanophotonics using the Stokes parameters alone. Our method for solving Maxwell's equations endows the Stokes parameters an even more fundamental role in the electromagnetic scattering theory.

Deep Neural Networks (DNN) have shown great promise in many classification applications, yet are widely known to have poorly calibrated predictions when they are over-parametrized. Improving DNN calibration without comprising on model accuracy is of extreme importance and interest in safety critical applications such as in the health-care sector. In this work, we show that decoupling the training of feature extraction layers and classification layers in over-parametrized DNN architectures such as Wide Residual Networks (WRN) and Visual Transformers (ViT) significantly improves model calibration whilst retaining accuracy, and at a low training cost. In addition, we show that placing a Gaussian prior on the last hidden layer outputs of a DNN, and training the model variationally in the classification training stage, even further improves calibration. We illustrate these methods improve calibration across ViT and WRN architectures for several image classification benchmark datasets., Comment: Proceedings of the 41 st International Conference on Machine Learning (ICML) 2024

We establish the concept of topological pumping in one-dimensional systems with long-range couplings and apply it to the transport of a photon in quantum optical systems. In our theoretical investigation, we introduce an extended version of the Rice-Mele model with all-to-all couplings. By analyzing its properties, we identify the general conditions for topological pumping and theoretically and numerically demonstrate topologically protected and dispersionless transport of a photon on a one-dimensional emitter chain. As concrete examples, we investigate three different popular quantum optics platforms, namely Ryd\-berg atom lattices, dense lattices of atoms excited to low-lying electronic states, and atoms coupled to waveguides, using experimentally relevant parameters. We observe that despite the long-ranged character of the dipole-dipole interactions, topological pumping facilitates the transport of a photon with a fidelity per cycle which can reach 99.9\%. Moreover, we find that the photon pumping process remains topologically protected against local disorder in the coupling parameters., Comment: 16 pages, 9 figures

Alzheimer's disease (AD) and sleep disorders exhibit a close association, where disruptions in sleep patterns often precede the onset of Mild Cognitive Impairment (MCI) and early-stage AD. This study delves into the potential of utilizing sleep-related electroencephalography (EEG) signals acquired through polysomnography (PSG) for the early detection of AD. Our primary focus is on exploring semi-supervised Deep Learning techniques for the classification of EEG signals due to the clinical scenario characterized by the limited data availability. The methodology entails testing and comparing the performance of semi-supervised SMATE and TapNet models, benchmarked against the supervised XCM model, and unsupervised Hidden Markov Models (HMMs). The study highlights the significance of spatial and temporal analysis capabilities, conducting independent analyses of each sleep stage. Results demonstrate the effectiveness of SMATE in leveraging limited labeled data, achieving stable metrics across all sleep stages, and reaching 90% accuracy in its supervised form. Comparative analyses reveal SMATE's superior performance over TapNet and HMM, while XCM excels in supervised scenarios with an accuracy range of 92 - 94%. These findings underscore the potential of semi-supervised models in early AD detection, particularly in overcoming the challenges associated with the scarcity of labeled data. Ablation tests affirm the critical role of spatio-temporal feature extraction in semi-supervised predictive performance, and t-SNE visualizations validate the model's proficiency in distinguishing AD patterns. Overall, this research contributes to the advancement of AD detection through innovative Deep Learning approaches, highlighting the crucial role of semi-supervised learning in addressing data limitations., Comment: 12 pages, 14 figures. Submitted to IEEE Biomedical and Health Informatics for publication

General Effect Modelling (GEM) is an umbrella over different methods that utilise effects in the analyses of data with multiple design variables and multivariate responses. To demonstrate the methodology, we here use GEM in gene expression data where we use GEM to combine data from different cohorts and apply multivariate analysis of the effects of the targeted disease across the cohorts. Omics data are by nature multivariate, yet univariate analysis is the dominating approach used for such data. A major challenge in omics data is that the number of features such as genes, proteins and metabolites are often very large, whereas the number of samples is limited. Furthermore, omics research aims to obtain results that are generically valid across different backgrounds. The present publication applies GEM to address these aspects. First, we emphasise the benefit of multivariate analysis for multivariate data. Then we illustrate the use of GEM to combine data from two different cohorts for multivariate analysis across the cohorts, and we highlight that multivariate analysis can detect information that is lost by univariate validation., Comment: 12 pages

Implicit generative models have the capability to learn arbitrary complex data distributions. On the downside, training requires telling apart real data from artificially-generated ones using adversarial discriminators, leading to unstable training and mode-dropping issues. As reported by Zahee et al. (2017), even in the one-dimensional (1D) case, training a generative adversarial network (GAN) is challenging and often suboptimal. In this work, we develop a discriminator-free method for training one-dimensional (1D) generative implicit models and subsequently expand this method to accommodate multivariate cases. Our loss function is a discrepancy measure between a suitably chosen transformation of the model samples and a uniform distribution; hence, it is invariant with respect to the true distribution of the data. We first formulate our method for 1D random variables, providing an effective solution for approximate reparameterization of arbitrary complex distributions. Then, we consider the temporal setting (both univariate and multivariate), in which we model the conditional distribution of each sample given the history of the process. We demonstrate through numerical simulations that this new method yields promising results, successfully learning true distributions in a variety of scenarios and mitigating some of the well-known problems that state-of-the-art implicit methods present., Comment: Proceedings of the 27th International Conference on Artificial Intelligence and Statistics (AISTATS) 2024

In this work we explore the propagation of massive scalar fields on some wormhole backgrounds. On one side, we consider the Bronnikov-Ellis wormhole solution and wormhole geometries with a non-constant redshift function by introducing a gravitational mass $M$, which goes over into the Bronnikov-Ellis wormhole when the gravitational mass parameter vanishes. We employ the continued fraction method to calculate accurately the quasinormal frequencies of massive scalar fields, particularly focusing on low values of the angular number, and we show an anomalous behaviour of the decay rate of the quasinormal frequencies, for $n \geq \ell$. Also, we show that for a massive scalar field and $M \neq 0$ the effective potential allows potential wells for some values of the parameters which support bound states, which are obtained using the continued fraction method and they are characterized by having only a frequency of oscillation and they do not decay; however, for the Bronnikov-Ellis wormhole the effective potential do not support bound states. On the other side, we consider a wormhole geometry which is an exact solution of $f(R)$ modified gravity. For this geometry the quasinormal frequencies can be obtained analytically, being the longest-lived modes the ones with lowest angular number $\ell$. So, in this wormhole background the anomalous behaviour is avoided., Comment: Version accepted for publication in PRD. arXiv admin note: text overlap with arXiv:2205.06079

A bimetallic Au-Pd catalyst supported on ceriazirconia with Au:Pd molar ratio 0.8 has been synthesized using a simultaneous deposition-precipitation method and oxidized at 250, 450, and 700 $^\circ$C in order to modify its particle size, nanostructure, and composition. Combined Xray energy dispersive spectroscopy and Xray photoelectron spectroscopy analysis clearly evidence that the bimetallic Au-Pd catalyst oxidized at 250 $^\circ$C is made up of a mixture of monometallic Au and Pd and bimetallic Au-Pd nanoparticles with Au:Pd ratios varying over a wide range. Increasing oxidation temperature leads to a stronger interaction between Au and Pd. Meanwhile, a slight increase of particle size and a narrowing of the Au:Pd ratio in the bimetallic nanoparticles take place. Compared with titania and activated carbon supports, the resistance against sintering at high temperatures of Au-Pd metal particles supported on ceriazirconia is proven to be higher. A synergistic effect has been observed for selective oxidation of benzyl alcohol on these catalysts. The catalytic activity decreases only slightly after oxidation at 450 $^\circ$C. However, oxidation at 700 $^\circ$C results in much lower catalytic activity. Migration of Pd onto Au particles during oxidation of benzyl alcohol enhances the catalytic activity of a physical mixture of monometallic Au and Pd supported on ceriazirconia catalysts. This fact, jointly with an analysis of the intrinsic activity, reveals the influence of the actual nature of Au-Pd interactions in the bimetallic particles, which points to higher activity of Au@Pd or Au@Pd@Pd nanostructures on ceria-zirconia support., Comment: 12 pages, 8 figures

Catastrophic overfitting (CO) in single-step adversarial training (AT) results in abrupt drops in the adversarial test accuracy (even down to 0%). For models trained with multi-step AT, it has been observed that the loss function behaves locally linearly with respect to the input, this is however lost in single-step AT. To address CO in single-step AT, several methods have been proposed to enforce local linearity of the loss via regularization. However, these regularization terms considerably slow down training due to Double Backpropagation. Instead, in this work, we introduce a regularization term, called ELLE, to mitigate CO effectively and efficiently in classical AT evaluations, as well as some more difficult regimes, e.g., large adversarial perturbations and long training schedules. Our regularization term can be theoretically linked to curvature of the loss function and is computationally cheaper than previous methods by avoiding Double Backpropagation. Our thorough experimental validation demonstrates that our work does not suffer from CO, even in challenging settings where previous works suffer from it. We also notice that adapting our regularization parameter during training (ELLE-A) greatly improves the performance, specially in large $\epsilon$ setups. Our implementation is available in https://github.com/LIONS-EPFL/ELLE ., Comment: Accepted in ICLR 2024

"Papel y lápiz" (Paper and pencil) is the result of a qualitative research project carried out using the Participatory Action Research (PAR) as a model. "Papel y lapiz" seeks to teach young people and children, who are identified as being at high social risk, in 35 different public educational institutions (PEIs) from Cartagena de Indias, Colombia. The principle aim of the project will be educating them about the importance of knowing and understanding the often-harsh realities of their social situations with particular focus on the social risks each of them might encounter. "Papel y lapiz" also aims to teach students about the social situation of their city using media and specifically the press. Working alongside Educommunication, the aim is to start educating the young people in school classrooms, in other words the most formative years of their youth. Between 2019 and 2022 this research project has reached 712 students from various public Educational Institutions (EIs) in Cartagena. The project was materialized in collaboration with teachers and directors by creating 6 educational cards that incorporate the use of the press to analyze some of the most critical issues the city is facing. [For the full proceedings, see ED654100.]