Search

In recent works by L. Drewnowski and I. Labuda and J. Mart\'inez et al., non-pathological analytic \( P \)-ideals and non-pathological \( F_\sigma \)-ideals have been characterized and studied in terms of their representations by a sequence \( (x_n)_n \) in a Banach space, as \( \mathcal{C}((x_n)_n) \) and \( \mathcal{B}((x_n)_n) \). The ideal \( \mathcal{C}((x_n)_n) \) consists of sets where the series \( \sum_{n \in A} x_n \) is unconditionally convergent, while \( \mathcal{B}((x_n)_n) \) involves weak unconditional convergence. In this paper, we further study these representations and provide effective descriptions of \( \mathcal{B} \)- and \( \mathcal{C} \)-ideals in the universal spaces \( C([0,1]) \) and \( C(2^{\mathbb{N}}) \), addressing a question posed by Borodulin-Nadzieja et al. A key aspect of our study is the role of the space \( c_0 \) in these representations. We focus particularly on \( \mathcal{B} \)-representations in spaces containing many copies of \( c_0 \), such as \( c_0 \)-saturated spaces of continuous functions. A central tool in our analysis is the concept of \( c \)-coloring ideals, which arise from homogeneous sets of continuous colorings. These ideals, generated by homogeneous sets of 2-colorings, exhibit a rich combinatorial structure. Among our results, we prove that for \( d \geq 3 \), the random \( d \)-homogeneous ideal is pathological, we construct hereditarily non-pathological universal \( c \)-coloring ideals, and we show that every \( \mathcal{B} \)-ideal represented in \( C(K) \), for \( K \) countable, contains a \( c \)-coloring ideal. Furthermore, by leveraging \( c \)-coloring ideals, we provide examples of \( \mathcal{B} \)-ideals that are not \( \mathcal{B} \)-representable in \( c_0 \). These findings highlight the interplay between combinatorial properties of ideals and their representations in Banach spaces.

Text classifiers suffer from small perturbations, that if chosen adversarially, can dramatically change the output of the model. Verification methods can provide robustness certificates against such adversarial perturbations, by computing a sound lower bound on the robust accuracy. Nevertheless, existing verification methods incur in prohibitive costs and cannot practically handle Levenshtein distance constraints. We propose the first method for computing the Lipschitz constant of convolutional classifiers with respect to the Levenshtein distance. We use these Lipschitz constant estimates for training 1-Lipschitz classifiers. This enables computing the certified radius of a classifier in a single forward pass. Our method, LipsLev, is able to obtain $38.80$% and $13.93$% verified accuracy at distance $1$ and $2$ respectively in the AG-News dataset, while being $4$ orders of magnitude faster than existing approaches. We believe our work can open the door to more efficient verification in the text domain., Comment: Accepted in ICLR 2025

This work describes our group's submission to the PROCESS Challenge 2024, with the goal of assessing cognitive decline through spontaneous speech, using three guided clinical tasks. This joint effort followed a holistic approach, encompassing both knowledge-based acoustic and text-based feature sets, as well as LLM-based macrolinguistic descriptors, pause-based acoustic biomarkers, and multiple neural representations (e.g., LongFormer, ECAPA-TDNN, and Trillson embeddings). Combining these feature sets with different classifiers resulted in a large pool of models, from which we selected those that provided the best balance between train, development, and individual class performance. Our results show that our best performing systems correspond to combinations of models that are complementary to each other, relying on acoustic and textual information from all three clinical tasks.

We explore the Mott transition in orthorhombic diamond lattices relevant to (ET)Ag$_4$(CN)$_5$ molecular compounds. The non-interacting phases include nodal line, Dirac and/or Weyl semimetals depending on the strength of spin-orbit coupling and the degree of dimerization of the lattice. Based on an extension of slave-rotor mean-field theory which accounts for magnetic order, we find a transition from a semimetal to a paramagnetic Mott insulator at a critical $U_c$ which becomes N\'eel ordered at a larger Coulomb repulsion, $U_{cm}>U_{c}$. The resulting intermediate Mott phase is a $U(1)$ quantum spin liquid (QSL) consisting on spinon preserving the nodal structure of the nearby semimetallic phases. An analysis of the Green's function in this Mott phase shows how the zeros follow the spinon band dispersions carrying the topology while the poles describe the Hubbard bands. Our results are relevant to recent observations in (ET)Ag$_4$(CN)$_5$ molecular compounds in which the ambient pressure N\'eel ordered Mott insulator is gradually suppressed until semimetallic behavior arises at larger pressures., Comment: 22 pages, 13 figures

It is advantageous for any quantum processor to support different classes of two-qubit quantum logic gates when compiling quantum circuits, a property that is typically not seen with existing platforms. In particular, access to a gate set that includes support for the CZ-type, the iSWAP-type, and the SWAP-type families of gates, renders conversions between these gate families unnecessary during compilation as any two-qubit Clifford gate can be executed using at most one two-qubit gate from this set, plus additional single-qubit gates. We experimentally demonstrate that a SWAP gate can be decomposed into one iSWAP gate followed by one CZ gate, affirming a more efficient compilation strategy over the conventional approach that relies on three iSWAP or three CZ gates to replace a SWAP gate. Our implementation makes use of a superconducting quantum processor design based on fixed-frequency transmon qubits coupled together by a parametrically modulated tunable transmon coupler, extending this platform's native gate set so that any two-qubit Clifford unitary matrix can be realized using no more than two two-qubit gates and single-qubit gates., Comment: 29 pages, 13 figures

The risk of language models reproducing copyrighted material from their training data has led to the development of various protective measures. Among these, inference-time strategies that impose constraints via post-processing have shown promise in addressing the complexities of copyright regulation. However, they often incur prohibitive computational costs or suffer from performance trade-offs. To overcome these limitations, we introduce Copyright-Protecting Model Fusion (CP-Fuse), a novel approach that combines models trained on disjoint sets of copyrighted material during inference. In particular, CP-Fuse adaptively aggregates the model outputs to minimize the reproduction of copyrighted content, adhering to a crucial balancing property that prevents the regurgitation of memorized data. Through extensive experiments, we show that CP-Fuse significantly reduces the reproduction of protected material without compromising the quality of text and code generation. Moreover, its post-hoc nature allows seamless integration with other protective measures, further enhancing copyright safeguards. Lastly, we show that CP-Fuse is robust against common techniques for extracting training data., Comment: 47 pages, 21 Figures. arXiv admin note: substantial text overlap with arXiv:2407.20105

Computer science education has been at the heart of Scheme from the beginning. The language was designed in the 1970s concurrently with the MIT course 6.001 and the textbook "Structure and Interpretation of Computer Programs" (SICP). To explain the scope of variables at run time in the presence of higher-order procedures, SICP introduces a mental model called the environment model, along with a pictorial representation of environments and data structures. Recently, the concept of notional machines has emerged in computer science education: a predictive set of abstractions that define the structure and behavior of a computational device. Proponents of notional machines argue that learners benefit when complex dynamic concepts such as the computational structure of Scheme are accompanied with concise notional machines. In this paper, we start with a sublanguage of Scheme sufficient for all programs in SICP that we call SICP Scheme. We extend the environment model to a full notional machine for SICP Scheme that is simple enough to serve as the central mental model in a CS1 course and demonstrate the machine with computer-generated visualizations. Moving beyond SICP Scheme, we show how the notional machine can be further extended to explain Scheme's call/cc and thus make this powerful concept accessible to beginners through a coherent mental model. The presented notional machine serves as the core of a web-based implementation of Scheme that is under development at our university.

Recent works in pathological speech analysis have increasingly relied on powerful self-supervised speech representations, leading to promising results. However, the complex, black-box nature of these embeddings and the limited research on their interpretability significantly restrict their adoption for clinical diagnosis. To address this gap, we propose a novel, interpretable framework specifically designed to support Parkinson's Disease (PD) diagnosis. Through the design of simple yet effective cross-attention mechanisms for both embedding- and temporal-level analysis, the proposed framework offers interpretability from two distinct but complementary perspectives. Experimental findings across five well-established speech benchmarks for PD detection demonstrate the framework's capability to identify meaningful speech patterns within self-supervised representations for a wide range of assessment tasks. Fine-grained temporal analyses further underscore its potential to enhance the interpretability of deep-learning pathological speech models, paving the way for the development of more transparent, trustworthy, and clinically applicable computer-assisted diagnosis systems in this domain. Moreover, in terms of classification accuracy, our method achieves results competitive with state-of-the-art approaches, while also demonstrating robustness in cross-lingual scenarios when applied to spontaneous speech production., Comment: Submitted to the Special Issue on "Modelling and Processing Language and Speech in Neurodegenerative Disorders" published by Journal of Selected Topics in Signal Processing (JSTSP)

The typically large distances, extinction, and crowding of Galactic supermassive star clusters have so far hampered the identification of their very low mass members, required to extend our understanding of star and planet formation, and early stellar evolution, to starburst. This situation has now evolved thanks to the James Webb Space Telescope (JWST), and its unmatched resolution and sensitivity in the infrared. In this paper, the third of the series of the Extended Westerlund 1 and 2 Open Clusters Survey (EWOCS), we present JWST/NIRCam and JWST/MIRI observations of the supermassive star cluster Westerlund 1. These observations are specifically designed to unveil the cluster members down to the BD mass regime, and to allow us to select and study the protoplanetary disks and to study the mutual feedback between the cluster members and the surrounding environment. Westerlund 1 was observed as part of JWST GO-1905 for 23.6 hours. The data have been reduced using the JWST calibration pipeline, together with specific tools necessary to remove artifacts. Source identification and photometry were performed with DOLPHOT. The MIRI images show a plethora of different features. Diffuse nebular emission is observed around the cluster, which is typically composed of myriads of droplet-like features pointing toward the cluster center or the group of massive stars surrounding the WR star W72/A. A long pillar is also observed in the NW. The MIRI images also show resolved shells and outflows surrounding the M-type RSG W20, W26, W75, and W237, the sgB[e] star W9 and the YHG W4. The color-magnitude diagrams built using the NIRCam photometry show a clear cluster sequence, which is marked in its upper part by the 1828 NIRCam stars with X-ray counterparts. NIRCam observations using the F115W filter have reached the 23.8 mag limit with 50\% completeness (roughly corresponding to a 0.06 Msol brown dwarf)., Comment: Accepted for publication in Astronomy & Astrophysics

We present a new class of interacting dark sector models that can address the Hubble tension. Interacting dark radiation (DR) has previously been put forward as a solution to the problem, but this proposal is disfavored by the high-$\ell$ cosmic microwave background (CMB) data. We modify this basic framework by introducing a subcomponent of dark matter (DM) that interacts strongly with the DR, so that together they constitute a tightly coupled fluid at early times. We show that if this subcomponent decouples from the interacting DR during the CMB epoch, the $\ell$ modes of the CMB that entered the horizon before decoupling are impacted differently from those that entered after, allowing a solution to the problem. We present a model that realizes this framework, which we dub "New Atomic Dark Matter", or nuADaM, in which the interacting dark matter (iDM) subcomponent is composed of dark atoms, and dark "neutrinos" with long-range interactions contribute to the DR, hence the name of the model. This iDM subcomponent is acoustic at early times but decouples from the DR following dark recombination. In contrast to conventional atomic dark matter (ADM) models, the dark photon is part of a richer DR sector, which ensures that it continues to be self-interacting even after recombination. We show that this model admits a fit to the available cosmological data that is significantly better than both $\Lambda$CDM and conventional ADM., Comment: 47 pages, 15 figures

Deep Learning (DL) has shown promise for downscaling global climate change projections under different approaches, including Perfect Prognosis (PP) and Regional Climate Model (RCM) emulation. Unlike emulators, PP downscaling models are trained on observational data, so it remains an open question whether they can plausibly extrapolate unseen conditions and changes in future emissions scenarios. Here we focus on this problem as the main drawback for the operationalization of these methods and present the results of 1) a literature review to identify state-of-the-art DL models for PP downscaling and 2) an intercomparison experiment to evaluate the performance of these models and to assess their extrapolation capability using a common experimental framework, taking into account the sensitivity of results to different training replicas. We focus on minimum and maximum temperatures and precipitation over Spain, a region with a range of climatic conditions with different influential regional processes. We conclude with a discussion of the findings, limitations of existing methods, and prospects for future development., Comment: Under review for Earth's Future

While security vulnerabilities in traditional Deep Neural Networks (DNNs) have been extensively studied, the susceptibility of Spiking Neural Networks (SNNs) to adversarial attacks remains mostly underexplored. Until now, the mechanisms to inject backdoors into SNN models have been limited to digital scenarios; thus, we present the first evaluation of backdoor attacks in real-world environments. We begin by assessing the applicability of existing digital backdoor attacks and identifying their limitations for deployment in physical environments. To address each of the found limitations, we present three novel backdoor attack methods on SNNs, i.e., Framed, Strobing, and Flashy Backdoor. We also assess the effectiveness of traditional backdoor procedures and defenses adapted for SNNs, such as pruning, fine-tuning, and fine-pruning. The results show that while these procedures and defenses can mitigate some attacks, they often fail against stronger methods like Flashy Backdoor or sacrifice too much clean accuracy, rendering the models unusable. Overall, all our methods can achieve up to a 100% Attack Success Rate while maintaining high clean accuracy in every tested dataset. Additionally, we evaluate the stealthiness of the triggers with commonly used metrics, finding them highly stealthy. Thus, we propose new alternatives more suited for identifying poisoned samples in these scenarios. Our results show that further research is needed to ensure the security of SNN-based systems against backdoor attacks and their safe application in real-world scenarios. The code, experiments, and results are available in our repository.

Large vision-language models (VLLMs) exhibit promising capabilities for processing multi-modal tasks across various application scenarios. However, their emergence also raises significant data security concerns, given the potential inclusion of sensitive information, such as private photos and medical records, in their training datasets. Detecting inappropriately used data in VLLMs remains a critical and unresolved issue, mainly due to the lack of standardized datasets and suitable methodologies. In this study, we introduce the first membership inference attack (MIA) benchmark tailored for various VLLMs to facilitate training data detection. Then, we propose a novel MIA pipeline specifically designed for token-level image detection. Lastly, we present a new metric called MaxR\'enyi-K%, which is based on the confidence of the model output and applies to both text and image data. We believe that our work can deepen the understanding and methodology of MIAs in the context of VLLMs. Our code and datasets are available at https://github.com/LIONS-EPFL/VL-MIA., Comment: NeurIPS 2024

We propose a relaxation mechanism for the initial misalignment angle of the pre-inflationary QCD axion with a large decay constant. The proposal addresses the challenges posed to the axion dark matter scenario by an overabundance of axions overclosing the Universe, as well as by isocurvature constraints. Many state-of-the-art experiments are searching for QCD axion dark matter with a decay constant as large as $10^{16}\,\mathrm{GeV}$, motivating the need for a theoretical framework such as ours. In our model, hidden sector magnetic monopoles generated in the early Universe give the axion a large mass via the Witten effect, causing early oscillations that reduce the misalignment angle and axion abundance. As the hidden gauge symmetry breaks, its monopoles confine via cosmic strings, dissipating energy into the Standard Model and leading to monopole-antimonopole annihilation. This removes the monopole-induced mass, leaving only the standard QCD term. We consider the symmetry breaking pattern of $\mathrm{SU}(2)' \to \mathrm{U}(1)' \to 1$, leading to monopole and string formation respectively. We calculate the monopole abundance, their interactions with the axion field, and the necessary conditions for monopole-induced axion oscillations, while accounting for UV instanton effects. We present three model variations based on different symmetry breaking scales and show that they can accommodate an axion decay constant of up to $10^{16}\,\mathrm{GeV}$ with an inflationary scale of $10^{15}\,\mathrm{GeV}$. The required alignment between monopole-induced and QCD axion potentials is achieved through a modest Nelson-Barr mechanism, avoiding overclosure without anthropic reasoning., Comment: 25 pages + references; 5 figures

We review some representative results for first-passage problems involving so-called mortal or evanescent walkers, i.e., walkers with a finite lifetime. The mortality constraint plays a key role in the modeling of many real scenarios, as it filters out long Brownian trajectories, thereby drastically modifying space exploration properties. Among such scenarios, we consider here different first-passage problems, including one or many searchers, resetting, anomalous diffusion, evolving domains, and the narrow escape problem. In spite of the different physics, the mathematical treatment draws strongly on the formalism for standard (i.e., immortal) walkers.

Self-supervised learning (SSL) leverages large amounts of unlabelled data to learn rich speech representations, fostering improvements in automatic speech recognition (ASR), even when only a small amount of labelled data is available for fine-tuning. Despite the advances in SSL, a significant challenge remains when the data used for pre-training (source domain) mismatches the fine-tuning data (target domain). To tackle this domain mismatch challenge, we propose a new domain adaptation method for low-resource ASR focused on contrastive mixup for joint-embedding architectures named AC-Mix (agnostic contrastive mixup). In this approach, the SSL model is adapted through additional pre-training using mixed data views created by interpolating samples from the source and the target domains. Our proposed adaptation method consistently outperforms the baseline system, using approximately 11 hours of adaptation data and requiring only 1 hour of adaptation time on a single GPU with WavLM-Large.

Magnetotactic bacteria (MTB) are of significant interest for biophysical applications, particularly in cancer treatment. The biomineralized magnetosomes produced by these bacteria are high-quality magnetic nanoparticles that form chains through a highly reproducible natural process. Specifically, Magnetovibrio blakemorei and Magnetospirillum gryphiswaldense exhibit distinct magnetosome morphologies: truncated hexa-octahedral and truncated octahedral shapes, respectively. Despite having identical compositions (magnetite, Fe3O4) and comparable dimensions, their effective uniaxial anisotropies differ significantly, with M. blakemorei showing ~25 kJ/m^3 and M. gryphiswaldense ~11 kJ/m^3 at 300K. This variation presents a unique opportunity to explore the role of different anisotropy contributions in the magnetic responses of magnetite-based nanoparticles. This study systematically investigates these responses by examining static magnetization as a function of temperature (M vs. T, 5 mT) and magnetic field (M vs. H, up to 1 T). Above the Verwey transition temperature (110 K), the effective anisotropy is dominated by shape anisotropy, notably increasing coercivity for M. blakemorei by up to two-fold compared to M. gryphiswaldense. Below this temperature, the effective uniaxial anisotropy increases non-monotonically, significantly altering magnetic behavior. Our simulations based on dynamic Stoner-Wohlfarth models indicate that below the Verwey temperature, a uniaxial magnetocrystalline contribution emerges, peaking at ~22-24 kJ/m^3 at 5 K, values close to those of bulk magnetite. This demonstrates the profound impact of anisotropic properties on the magnetic behaviors and applications of magnetite-based nanoparticles and highlights the exceptional utility of magnetosomes as ideal model systems for studying the complex interplay of anisotropies in magnetite-based nanoparticles.

Artificial Neural Networks (ANNs), commonly mimicking neurons with non-linear functions to output floating-point numbers, consistently receive the same signals of a data point during its forward time. Unlike ANNs, Spiking Neural Networks (SNNs) get various input signals in the forward time of a data point and simulate neurons in a biologically plausible way, i.e., producing a spike (a binary value) if the accumulated membrane potential of a neuron is larger than a threshold. Even though ANNs have achieved remarkable success in multiple tasks, e.g., face recognition and object detection, SNNs have recently obtained attention due to their low power consumption, fast inference, and event-driven properties. While privacy threats against ANNs are widely explored, much less work has been done on SNNs. For instance, it is well-known that ANNs are vulnerable to the Membership Inference Attack (MIA), but whether the same applies to SNNs is not explored. In this paper, we evaluate the membership privacy of SNNs by considering eight MIAs, seven of which are inspired by MIAs against ANNs. Our evaluation results show that SNNs are more vulnerable (maximum 10% higher in terms of balanced attack accuracy) than ANNs when both are trained with neuromorphic datasets (with time dimension). On the other hand, when training ANNs or SNNs with static datasets (without time dimension), the vulnerability depends on the dataset used. If we convert ANNs trained with static datasets to SNNs, the accuracy of MIAs drops (maximum 11.5% with a reduction of 7.6% on the test accuracy of the target model). Next, we explore the impact factors of MIAs on SNNs by conducting a hyperparameter study. Finally, we show that the basic data augmentation method for static data and two recent data augmentation methods for neuromorphic data can considerably (maximum reduction of 25.7%) decrease MIAs' performance on SNNs.

The realization of fault-tolerant quantum computing requires the execution of quantum error-correction (QEC) schemes, to mitigate the fragile nature of qubits. In this context, to ensure the success of QEC, a protocol capable of implementing both qubit reset and leakage reduction is highly desirable. We demonstrate such a protocol in an architecture consisting of fixed-frequency transmon qubits pair-wise coupled via tunable couplers -- an architecture that is compatible with the surface code. We use tunable couplers to transfer any undesired qubit excitation to the readout resonator of the qubit, from which this excitation decays into the feedline. In total, the combination of qubit reset, leakage reduction, and coupler reset takes only 83ns to complete. Our reset scheme is fast, unconditional, and achieves fidelities well above 99%, thus enabling fixed-frequency qubit architectures as future implementations of fault-tolerant quantum computers. Our protocol also provides a means to both reduce QEC cycle runtime and improve algorithmic fidelity on quantum computers.

We consider the possibility of indirect detection of dark sector processes by investigating a novel form of interaction between ambient dark matter (DM) and primordial black holes (PBHs). The basic scenario we envisage is that the ambient DM is ``dormant'', \ie, it has interactions with the SM, but its potential for an associated SM signal is not realized for various reasons. We argue that the presence of PBHs with active Hawking radiation (independent of any DM considerations) can act as a catalyst in this regard by overcoming the aforementioned bottlenecks. The central point is that PBHs radiate all types of particles, whether in the standard model (SM) or beyond (BSM), which have a mass at or below their Hawking temperature. The emission of such radiation is ``democratic" (up to the particle spin), since it is based on a coupling of sorts of gravitational origin. In particular, such shining of (possibly dark sector) particles onto ambient DM can then activate the latter into giving potentially observable SM signals. We illustrate this general mechanism with two specific models. First, we consider asymmetric DM, which is characterized by an absence of ambient anti-DM, and consequently the absence of DM indirect detection signals. In this case, PBHs can ``resurrect'' such a signal by radiating anti-DM, which then annihilates with ambient DM in order to give SM particles such as photons. In our second example, we consider the PBH emission of dark gauge bosons which can excite ambient DM into a heavier state (which is, again, not ambient otherwise), this heavier state later decays back into DM and photons. Finally, we demonstrate that we can obtain observable signals of these BSM models from asteroid-mass PBHs (Hawking radiating currently with $\sim \mathcal{O}(\mathrm{MeV})$ temperatures) at gamma-ray experiments such as AMEGO-X., Comment: 19 pages + 3 appendices, 9 figures

Speech is a rich biomarker that encodes substantial information about the health of a speaker, and thus it has been proposed for the detection of numerous diseases, achieving promising results. However, questions remain about what the models trained for the automatic detection of these diseases are actually learning and the basis for their predictions, which can significantly impact patients' lives. This work advocates for an interpretable health model, suitable for detecting several diseases, motivated by the observation that speech-affecting disorders often have overlapping effects on speech signals. A framework is presented that first defines "reference speech" and then leverages this definition for disease detection. Reference speech is characterized through reference intervals, i.e., the typical values of clinically meaningful acoustic and linguistic features derived from a reference population. This novel approach in the field of speech as a biomarker is inspired by the use of reference intervals in clinical laboratory science. Deviations of new speakers from this reference model are quantified and used as input to detect Alzheimer's and Parkinson's disease. The classification strategy explored is based on Neural Additive Models, a type of glass-box neural network, which enables interpretability. The proposed framework for reference speech characterization and disease detection is designed to support the medical community by providing clinically meaningful explanations that can serve as a valuable second opinion.

Due to the high cost of training, large model (LM) practitioners commonly use pretrained models downloaded from untrusted sources, which could lead to owning compromised models. In-context learning is the ability of LMs to perform multiple tasks depending on the prompt or context. This can enable new attacks, such as backdoor attacks with dynamic behavior depending on how models are prompted. In this paper, we leverage the ability of vision transformers (ViTs) to perform different tasks depending on the prompts. Then, through data poisoning, we investigate two new threats: i) task-specific backdoors where the attacker chooses a target task to attack, and only the selected task is compromised at test time under the presence of the trigger. At the same time, any other task is not affected, even if prompted with the trigger. We succeeded in attacking every tested model, achieving up to 89.90\% degradation on the target task. ii) We generalize the attack, allowing the backdoor to affect \emph{any} task, even tasks unseen during the training phase. Our attack was successful on every tested model, achieving a maximum of $13\times$ degradation. Finally, we investigate the robustness of prompts and fine-tuning as techniques for removing the backdoors from the model. We found that these methods fall short and, in the best case, reduce the degradation from 89.90\% to 73.46\%.

Reinforcement learning is a subfield of machine learning that is having a huge impact in the different conventional disciplines, including physical sciences. Here, we show how reinforcement learning methods can be applied to solve optimization problems in the context of radiative heat transfer. We illustrate their use with the optimization of the near-field radiative heat transfer between multilayer hyperbolic metamaterials. Specifically, we show how this problem can be formulated in the language of reinforcement learning and tackled with a variety of algorithms. We show that these algorithms allow us to find solutions that outperform those obtained using physical intuition. Overall, our work shows the power and potential of reinforcement learning methods for the investigation of a wide variety of problems in the context of radiative heat transfer and related topics., Comment: 21 pages, 9 figures

The risk of language models unintentionally reproducing copyrighted material from their training data has led to the development of various protective measures. In this paper, we propose model fusion as an effective solution to safeguard against copyright infringement. In particular, we introduce Copyright-Protecting Fusion (CP-Fuse), an algorithm that adaptively combines language models to minimize the reproduction of protected materials. CP-Fuse is inspired by the recently proposed Near-Access Free (NAF) framework and additionally incorporates a desirable balancing property that we demonstrate prevents the reproduction of memorized training data. Our results show that CP-Fuse significantly reduces the memorization of copyrighted content while maintaining high-quality text and code generation. Furthermore, we demonstrate how CP-Fuse can be integrated with other techniques for enhanced protection.

We provide a comprehensive study of the capabilities of modulated electron wavefunctions for the preparation and readout of the quantum state of the quantum emitters (QEs) they interact with. First, we consider perfectly periodic electron combs, which do not produce QE-electron entanglement, preserving the purity of the QE while inducing Rabi-like dynamics in it. We extend our findings to realistic, non-ideally modulated electron wavepackets, showing that the phenomenology persists, and exploring their use to prepare the emitter in a desired quantum state. Thus, we establish the balance that electron comb size, emitter radiative decay, and electron-emitter coupling strength must fulfil in order to implement our ideas in experimentally feasible platforms. Finally, moving into the limit of small electron combs, we reveal that these wavefunctions allow for quantum state tomography of their target, providing access not only to the populations, but also the coherences of the QE density matrix. We believe that our theoretical results showcase modulated free-electrons as very promising tools for quantum technologies based on light-matter coupling., Comment: 15 pages, 5 figures

We have developed a convolutional neural network (CNN) to reconstruct the shape of irregular rough particles from their interferometric images. The CNN is based on a UNET architecture with residual block modules. The database has been constructed using the experimental patterns generated by perfectly known pseudo-particles programmed on a Digital Micromirror Device (DMD) and under laser illumination. The CNN has been trained on a basis of 18000 experimental interferometric images using the AUSTRAL super computer (at CRIANN in Normandy). The CNN is tested in the case of centrosymmetric (stick, cross, dendrite) and non-centrosymmetric (like T, Y or L) particles. The size and the 3D orientation of the programmed particles are random. The different shapes are reconstructed by the CNN with good accuracy. Using three angles of view, the 3D reconstruction of particles from three reconstructed faces can be further done.

Based on the PubMed database, there are around 260 manuscripts describing nanothermometers. These research articles detail the synthesis, performance, and application of intracellular nanothermometers. This intracellular prevalence is due to the significant importance, complexity, and utility of the intracellular compartments for understanding cell metabolism and disease treatment. However, in recent years, the extracellular environment of the cell has emerged as a crucial factor in medicine, particularly in hyperthermia and immunotherapy. Despite this, we have not seen evidence in the literature describing the utilization or performance of a nanothermometer designed for extracellular temperature measurements. This oversight not only neglects the potential for measuring extracellular temperature but also fails to address the extracellular environment of the cell. Here, we introduce a nanothermometer designed specifically for measuring extracellular temperature by directly converting serum proteins into nanothermometers (either unmodified or labeled with the clinically approved dye Fluorescein). Additionally, leveraging the extracellular localization of these nanothermometers, we demonstrate (1) the enhancement of their temperature sensitivity by combining them with gold nanorods, and (2) their capability to generate damage and disrupt the plasma membrane, thus opening the door to their use as photodynamic therapy agents. We firmly believe that these advancements represent not only a broadening of the applications of nanothermometry but also a pioneering step in showcasing the ability of nanothermometers to induce cell death., Comment: 18 pages, 5 figures

Deep learning has emerged as a promising tool for precipitation downscaling. However, current models rely on likelihood-based loss functions to properly model the precipitation distribution, leading to spatially inconsistent projections when sampling. This work explores a novel approach by fusing the strengths of likelihood-based and adversarial losses used in generative models. As a result, we propose a likelihood-based generative approach for precipitation downscaling, leveraging the benefits of both methods., Comment: Accepted at ICML 2024 Machine Learning for Earth System Modeling workshop

Long-range interactions are a key resource in many quantum phenomena and technologies. Free-space photons mediate power-law interactions but lack tunability and suffer from decoherence processes due to their omnidirectional emission. Engineered dielectrics can yield tunable and coherent interactions, but typically at the expense of making them both shorter-ranged and sensitive to material disorder and photon loss. Here, we propose a platform that can circumvent all these limitations based on three-dimensional subwavelength atomic arrays subjected to magnetic fields. Our key result is to show how to design the polaritonic bands of these atomic metamaterials to feature a pair of frequency-isolated Weyl points. These Weyl excitations can thus mediate interactions that are simultaneously long-range, due to their gapless nature; robust, due to the topological protection of Weyl points; and decoherence-free, due to their subradiant character. We demonstrate the robustness of these isolated Weyl points for a large regime of interatomic distances and magnetic field values and characterize the emergence of their corresponding Fermi arcs surface states. The latter can as well lead to two-dimensional, non-reciprocal atomic interactions with no analogue in other chiral quantum optical setups., Comment: 15 pages, 6 figures

Since 2019, GRAVITY has provided direct observations of giant planets and brown dwarfs at separations of down to 95 mas from the host star. Some of these observations have provided the first direct confirmation of companions previously detected by indirect techniques (astrometry and radial velocities). We want to improve the observing strategy and data reduction in order to lower the inner working angle of GRAVITY in dual-field on-axis mode. We also want to determine the current limitations of the instrument when observing faint companions with separations in the 30-150 mas range. To improve the inner working angle, we propose a fiber off-pointing strategy during the observations to maximize the ratio of companion-light-to-star-light coupling in the science fiber. We also tested a lower-order model for speckles to decouple the companion light from the star light. We then evaluated the detection limits of GRAVITY using planet injection and retrieval in representative archival data. We compare our results to theoretical expectations. We validate our observing and data-reduction strategy with on-sky observations; first in the context of brown dwarf follow-up on the auxiliary telescopes with HD 984 B, and second with the first confirmation of a substellar candidate around the star Gaia DR3 2728129004119806464. With synthetic companion injection, we demonstrate that the instrument can detect companions down to a contrast of $8\times 10^{-4}$ ($\Delta \mathrm{K}= 7.7$ mag) at a separation of 35 mas, and a contrast of $3\times 10^{-5}$ ($\Delta \mathrm{K}= 11$ mag) at 100 mas from a bright primary (K<6.5), for 30 min exposure time. With its inner working angle and astrometric precision, GRAVITY has a unique reach in direct observation parameter space. This study demonstrates the promising synergies between GRAVITY and Gaia for the confirmation and characterization of substellar companions., Comment: 16 pages, 14 figures. Submitted to A&A

Adversarial attacks in Natural Language Processing apply perturbations in the character or token levels. Token-level attacks, gaining prominence for their use of gradient-based methods, are susceptible to altering sentence semantics, leading to invalid adversarial examples. While character-level attacks easily maintain semantics, they have received less attention as they cannot easily adopt popular gradient-based methods, and are thought to be easy to defend. Challenging these beliefs, we introduce Charmer, an efficient query-based adversarial attack capable of achieving high attack success rate (ASR) while generating highly similar adversarial examples. Our method successfully targets both small (BERT) and large (Llama 2) models. Specifically, on BERT with SST-2, Charmer improves the ASR in 4.84% points and the USE similarity in 8% points with respect to the previous art. Our implementation is available in https://github.com/LIONS-EPFL/Charmer., Comment: Accepted in ICML 2024

Membership Inference (MI) poses a substantial privacy threat to the training data of Automatic Speech Recognition (ASR) systems, while also offering an opportunity to audit these models with regard to user data. This paper explores the effectiveness of loss-based features in combination with Gaussian and adversarial perturbations to perform MI in ASR models. To the best of our knowledge, this approach has not yet been investigated. We compare our proposed features with commonly used error-based features and find that the proposed features greatly enhance performance for sample-level MI. For speaker-level MI, these features improve results, though by a smaller margin, as error-based features already obtained a high performance for this task. Our findings emphasise the importance of considering different feature sets and levels of access to target models for effective MI in ASR systems, providing valuable insights for auditing such models., Comment: Trustworthy Speech Processing, Satellite Workshop at ICASSP 2024

Randomized trials are considered the gold standard for making informed decisions in medicine, yet they often lack generalizability to the patient populations in clinical practice. Observational studies, on the other hand, cover a broader patient population but are prone to various biases. Thus, before using an observational study for decision-making, it is crucial to benchmark its treatment effect estimates against those derived from a randomized trial. We propose a novel strategy to benchmark observational studies beyond the average treatment effect. First, we design a statistical test for the null hypothesis that the treatment effects estimated from the two studies, conditioned on a set of relevant features, differ up to some tolerance. We then estimate an asymptotically valid lower bound on the maximum bias strength for any subgroup in the observational study. Finally, we validate our benchmarking strategy in a real-world setting and show that it leads to conclusions that align with established medical knowledge., Comment: Accepted for presentation at the Conference on Uncertainty in Artificial Intelligence (UAI) 2024

We aim to provide a detailed characterization of near-infrared spectra for young LT brown dwarfs, including robust spectral typing, calibrating spectral indices, identifying possible binaries, and selecting suitable spectral standards. We processed and analyzed archival spectra from VLT/X-shooter for a sample of 56 dwarfs with ages between 10 and 600 Myr and spectral types between late-M and mid-T. We re-determine spectral types by comparing them with a set of literature templates and assess a large range of spectral indices, calibrated using a specifically designed literature sample. We are able to identify 15 spectral indices that are useful for spectral typing for specific spectral ranges discussed here and provide the scaling relations with spectral types. Moreover, we also identify 6 spectral indices which can be used to separate young L dwarfs from the field ones. The equivalent-widths of the alkali lines show a correlation with age, increasing towards the objects with higher surface gravity. From our sample, we confirm 3 that are likely to be binaries by their anomalous spectra that appear to be better fitted by a combination of spectral types. Finally, we identify 12 objects as preliminary near-infrared spectral standards for young LT dwarfs. This paper presents a significant step toward understanding the spectral sequence and properties of young L and T dwarfs. The relations and standards provided here will be useful for future spectroscopic work on young brown dwarfs and giant planets., Comment: 25 pages, 17 figures. Accepted by A&A, X-shooter spectroscopic data will be made public on Vizier upon publication

The time evolution of the dependence of the mass accretion rate with the stellar mass and the disk mass represents a fundamental way to understand the evolution of protoplanetary disks and the formation of planets. In this work, we present observations with X-Shooter of 26 Class II very low-mass stars and brown dwarfs in the Ophiuchus, Cha-I, and Upper Scorpius star-forming regions (SFRs). These new observations extend down to SpT M9 ($\sim$0.02 $M_\odot$) the measurement of the mass accretion rate in Ophiuchus and Cha-I and add 11 very-low-mass stars to the sample of objects studied with broadband spectroscopy in Upper Scorpius. We obtained their SpT, extinction and physical parameters, and we used the intensity of various emission lines to derive their accretion luminosity and mass accretion rates. Combining these new observations with data from the literature, we compare relations between accretion and stellar and disk properties of four different SFRs with different ages: Ophiuchus (1 Myr), Lupus (2 Myr), Cha-I (3 Myr), and Upper Scorpius (5-12 Myr). We find the slopes of the $L_*-L\mathrm{_{acc}}$ and $M_*-\dot{M}\mathrm{_{acc}}$ relationships to steepen between Ophiuchus, Lupus, and Cha-I and that both relationships may be better described with a single power law. We also find the relationship between the disk mass and the mass accretion rate of the stellar population to steepen with time down to the age of Upper Scorpius. Overall, we observe hints of a faster evolution into low accretion rates of low-mass stars and brown dwarfs. We also find that brown dwarfs present higher $M\mathrm{_{disk}}/\dot{M}\mathrm{_{acc}}$ ratios (i.e., longer accretion depletion timescales) than stars in Ophiuchus, Lupus, and Cha-I. This apparently contradictory result may imply that the evolution of protoplanetary disks around brown dwarfs is different from what is seen in the stellar regime., Comment: Accepted for publication at A&A, X-Shooter spectra will be made public on Vizier upon publication

This paper investigates the vulnerability of spiking neural networks (SNNs) and federated learning (FL) to backdoor attacks using neuromorphic data. Despite the efficiency of SNNs and the privacy advantages of FL, particularly in low-powered devices, we demonstrate that these systems are susceptible to such attacks. We first assess the viability of using FL with SNNs using neuromorphic data, showing its potential usage. Then, we evaluate the transferability of known FL attack methods to SNNs, finding that these lead to suboptimal attack performance. Therefore, we explore backdoor attacks involving single and multiple attackers to improve the attack performance. Our primary contribution is developing a novel attack strategy tailored to SNNs and FL, which distributes the backdoor trigger temporally and across malicious devices, enhancing the attack's effectiveness and stealthiness. In the best case, we achieve a 100 attack success rate, 0.13 MSE, and 98.9 SSIM. Moreover, we adapt and evaluate an existing defense against backdoor attacks, revealing its inadequacy in protecting SNNs. This study underscores the need for robust security measures in deploying SNNs and FL, particularly in the context of backdoor attacks.

We research into the clinical, biochemical and neuroimaging factors associated with the outcome of stroke patients to generate a predictive model using machine learning techniques for prediction of mortality and morbidity 3 months after admission. The dataset consisted of patients with ischemic stroke (IS) and non-traumatic intracerebral hemorrhage (ICH) admitted to Stroke Unit of a European Tertiary Hospital prospectively registered. We identified the main variables for machine learning Random Forest (RF), generating a predictive model that can estimate patient mortality/morbidity. In conclusion, machine learning algorithms RF can be effectively used in stroke patients for long-term outcome prediction of mortality and morbidity., Comment: 12 pages, 5 figures

We present a novel approach for differentially private data synthesis of protected tabular datasets, a relevant task in highly sensitive domains such as healthcare and government. Current state-of-the-art methods predominantly use marginal-based approaches, where a dataset is generated from private estimates of the marginals. In this paper, we introduce PrivPGD, a new generation method for marginal-based private data synthesis, leveraging tools from optimal transport and particle gradient descent. Our algorithm outperforms existing methods on a large range of datasets while being highly scalable and offering the flexibility to incorporate additional domain-specific constraints., Comment: Published at the Forty-first International Conference on Machine Learning

Background and purpose: The unanticipated detection by magnetic resonance imaging (MRI) in the brain of asymptomatic subjects of white matter lesions suggestive of multiple sclerosis (MS) has been named radiologically isolated syndrome (RIS). As the difference between early MS [i.e. clinically isolated syndrome (CIS)] and RIS is the occurrence of a clinical event, it is logical to improve detection of the subclinical form without interfering with MRI as there are radiological diagnostic criteria for that. Our objective was to use machine-learning classification methods to identify morphometric measures that help to discriminate patients with RIS from those with CIS. Methods: We used a multimodal 3-T MRI approach by combining MRI biomarkers (cortical thickness, cortical and subcortical grey matter volume, and white matter integrity) of a cohort of 17 patients with RIS and 17 patients with CIS for single-subject level classification. Results: The best proposed models to predict the diagnosis of CIS and RIS were based on the Naive Bayes, Bagging and Multilayer Perceptron classifiers using only three features: the left rostral middle frontal gyrus volume and the fractional anisotropy values in the right amygdala and right lingual gyrus. The Naive Bayes obtained the highest accuracy [overall classification, 0.765; area under the receiver operating characteristic (AUROC), 0.782]. Conclusions: A machine-learning approach applied to multimodal MRI data may differentiate between the earliest clinical expressions of MS (CIS and RIS) with an accuracy of 78%. Keywords: Bagging; Multilayer Perceptron; Naive Bayes classifier; clinically isolated syndrome; diffusion tensor imaging; machine-learning; magnetic resonance imaging; multiple sclerosis; radiologically isolated syndrome., Comment: 24 pages, 2 tables

The method of Arterial Spin Labeling (ASL) has experienced a significant rise in its application to functional imaging, since it is the only technique capable of measuring blood perfusion in a truly non-invasive manner. Currently, there are no commercial packages for processing ASL data and there is no recognised standard for normalising ASL data to a common frame of reference. This work describes a new Automated Software for ASL Processing (ASAP) that can automatically process several ASL datasets. ASAP includes functions for all stages of image pre-processing: quantification, skull-stripping, co-registration, partial volume correction and normalization. To assess the applicability and validity of the toolbox, this work shows its application in the study of hypoperfusion in a sample of healthy subjects at risk of progressing to Alzheimer's Disease. ASAP requires limited user intervention, minimising the possibility of random and systematic errors, and produces cerebral blood flow maps that are ready for statistical group analysis. The software is easy to operate and results in excellent quality of spatial normalisation. The results found in this evaluation study are consistent with previous studies that find decreased perfusion, Comment: 10 pages, 8 figures, 3 tables

We argue that the striking similarity between the cosmic abundances of baryons and dark matter, despite their very different astrophysical behavior, strongly motivates the scenario in which dark matter resides within a rich dark sector parallel in structure to that of the standard model. The near cosmic coincidence is then explained by an approximate $\mathbb{Z}_2$ exchange symmetry between the two sectors, where dark matter consists of stable dark neutrons, with matter and dark matter asymmetries arising via parallel WIMP baryogenesis mechanisms. Taking a top-down perspective, we point out that an adequate $\mathbb{Z}_2$ symmetry necessitates solving the electroweak hierarchy problem in each sector, without our committing to a specific implementation. A higher-dimensional realization in the far UV is presented, in which the hierarchical couplings of the two sectors and the requisite $\mathbb{Z}_2$-breaking structure arise naturally from extra-dimensional localization and gauge symmetries. We trace the cosmic history, paying attention to potential pitfalls not fully considered in previous literature. Residual $\mathbb{Z}_2$-breaking can very plausibly give rise to the asymmetric reheating of the two sectors, needed to keep the cosmological abundance of relativistic dark particles below tight bounds. We show that, despite the need to keep inter-sector couplings highly suppressed after asymmetric reheating, there can naturally be order-one couplings mediated by TeV scale particles which can allow experimental probes of the dark sector at high energy colliders. Massive mediators can also induce dark matter direct detection signals, but likely at or below the neutrino floor., Comment: Published version. 29 pages + references, 6 figures, 2 sectors

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

Context. With a mass exceeding several 10^4 solar masses and a rich and dense population of massive stars, supermassive young star clusters represent the most massive star-forming environment that is dominated by the feedback from massive stars and gravitational interactions among stars. Aims. In this paper we present the "Extended Westerlund 1 and 2 Open Clusters Survey" (EWOCS) project, which aims to investigate the influence of the starburst environment on the formation of stars and planets, and on the evolution of both low and high mass stars. The primary targets of this project are Westerlund 1 and 2, the closest supermassive star clusters to the Sun. Methods. The project is based primarily on recent observations conducted with the Chandra and JWST observatories. Specifically, the Chandra survey of Westerlund 1 consists of 36 new ACIS-I observations, nearly co-pointed, for a total exposure time of 1 Msec. Additionally, we included 8 archival Chandra/ACIS-S observations. This paper presents the resulting catalog of X-ray sources within and around Westerlund 1. Sources were detected by combining various existing methods, and photon extraction and source validation were carried out using the ACIS-Extract software. Results. The EWOCS X-ray catalog comprises 5963 validated sources out of the 9420 initially provided to ACIS-Extract, reaching a photon flux threshold of approximately 2x10^-8 photons/cm^2/s. The X-ray sources exhibit a highly concentrated spatial distribution, with 1075 sources located within the central 1 arcminute. We have successfully detected X-ray emissions from 126 out of the 166 known massive stars of the cluster, and we have collected over 71000 photons from the magnetar CXO J164710.20-455217, Comment: The paper has been accepted for publication by Astronomy and Astrophysics

Thanks to their exceptional spatial, spectral and temporal resolution, highly-coherent free-electron beams have emerged as powerful probes for material excitations, enabling their characterization even in the quantum regime. Here, we investigate strong light-matter coupling through monochromatic and modulated electron wavepackets. In particular, we consider an archetypal target, comprising a nanophotonic cavity next to a single two-level emitter. We propose a model Hamiltonian describing the coherent interaction between the passing electron beam and the hybrid photonic-excitonic target, which is constructed using macroscopic quantum electrodynamics and fully parameterized in terms of the electromagnetic Dyadic Green's function. Using this framework, we first describe electron-energy-loss and cathodoluminescence spectroscopies, and photon-induced near-field electron emission microscopy. Finally, we show the power of modulated electrons beams as quantum tools for the manipulation of polaritonic targets presenting a complex energy landscape of excitations., Comment: 44 pages, 7 figures. Pages 1 to 12 are the main manuscript (with 6 figures). Pages 13-44 Suplementay Material

In the era of fast-paced precision medicine, observational studies play a major role in properly evaluating new treatments in clinical practice. Yet, unobserved confounding can significantly compromise causal conclusions drawn from non-randomized data. We propose a novel strategy that leverages randomized trials to quantify unobserved confounding. First, we design a statistical test to detect unobserved confounding with strength above a given threshold. Then, we use the test to estimate an asymptotically valid lower bound on the unobserved confounding strength. We evaluate the power and validity of our statistical test on several synthetic and semi-synthetic datasets. Further, we show how our lower bound can correctly identify the absence and presence of unobserved confounding in a real-world setting., Comment: Accepted for presentation at the International Conference on Artificial Intelligence and Statistics (AISTATS) 2024

Deep Neural Networks (DNNs) have shown great promise in various domains. Alongside these developments, vulnerabilities associated with DNN training, such as backdoor attacks, are a significant concern. These attacks involve the subtle insertion of triggers during model training, allowing for manipulated predictions. More recently, DNNs for tabular data have gained increasing attention due to the rise of transformer models. Our research presents a comprehensive analysis of backdoor attacks on tabular data using DNNs, mainly focusing on transformers. We also propose a novel approach for trigger construction: an in-bounds attack, which provides excellent attack performance while maintaining stealthiness. Through systematic experimentation across benchmark datasets, we uncover that transformer-based DNNs for tabular data are highly susceptible to backdoor attacks, even with minimal feature value alterations. We also verify that our attack can be generalized to other models, like XGBoost and DeepFM. Our results demonstrate up to 100% attack success rate with negligible clean accuracy drop. Furthermore, we evaluate several defenses against these attacks, identifying Spectral Signatures as the most effective. Nevertheless, our findings highlight the need to develop tabular data-specific countermeasures to defend against backdoor attacks.

The connection between symmetries and conservation laws is a cornerstone of physics. It underlies Bloch's theorem which explains wave phenomena in all linear periodic systems. Here we demonstrate that, in a nonlinear grating with memory, diffracted waves can spontaneously acquire momentum parallel to the lattice vector in quantities unconstrained by the grating period. In this breakdown of Bloch's theorem, which we also evidence in solutions to nonlinear Maxwell's equations, wave amplitudes no longer respect the discrete translation symmetry of the grating. Our findings reveal a rich phenomenology for waves in nonlinear periodic systems, and point to numerous opportunities for nonlinear lattices with broken symmetry in the context of imaging, sensing, and information processing in general., Comment: Main text: 12 pages plus 5 figures. Supplemental Information: pages 18-36

Speaker embeddings are ubiquitous, with applications ranging from speaker recognition and diarization to speech synthesis and voice anonymisation. The amount of information held by these embeddings lends them versatility, but also raises privacy concerns. Speaker embeddings have been shown to contain information on age, sex, health and more, which speakers may want to keep private, especially when this information is not required for the target task. In this work, we propose a method for removing and manipulating private attributes from speaker embeddings that leverages a Vector-Quantized Variational Autoencoder architecture, combined with an adversarial classifier and a novel mutual information loss. We validate our model on two attributes, sex and age, and perform experiments with ignorant and fully-informed attackers, and with in-domain and out-of-domain data., Comment: Article published in IEEE Access

This paper is a continuation of our paper [Med. J. Math 19, Article number: 31 (2022)] in which we extended the notion of generalized Drazin-Riesz invertible operators to closed operators. We establish here, results relating the notion of closed generalized Drazin-Riesz invertibility with the theory of $C_{0}$-semigroups. Firstly, we generalize results obtained in the bounded case [1] to the context of closed operators. Secondly, we investigate when an infinitesimal generator $A$ of a given $C_{0}$-semigroup is closed generalized Drazin-Riesz invertible. An application to $C_{0}$-groups and abstract second order differential equations is proposed, and an example of a $C_{0}$-group with closed generalized Drazin-Riesz invertible infinitesimal generator is given.

Conformers have recently been proposed as a promising modelling approach for automatic speech recognition (ASR), outperforming recurrent neural network-based approaches and transformers. Nevertheless, in general, the performance of these end-to-end models, especially attention-based models, is particularly degraded in the case of long utterances. To address this limitation, we propose adding a fully-differentiable memory-augmented neural network between the encoder and decoder of a conformer. This external memory can enrich the generalization for longer utterances since it allows the system to store and retrieve more information recurrently. Notably, we explore the neural Turing machine (NTM) that results in our proposed Conformer-NTM model architecture for ASR. Experimental results using Librispeech train-clean-100 and train-960 sets show that the proposed system outperforms the baseline conformer without memory for long utterances.