Search

This paper presents a novel approach combining inductive logic programming with reinforcement learning to improve training performance and explainability. We exploit inductive learning of answer set programs from noisy examples to learn a set of logical rules representing an explainable approximation of the agent policy at each batch of experience. We then perform answer set reasoning on the learned rules to guide the exploration of the learning agent at the next batch, without requiring inefficient reward shaping and preserving optimality with soft bias. The entire procedure is conducted during the online execution of the reinforcement learning algorithm. We preliminarily validate the efficacy of our approach by integrating it into the Q-learning algorithm for the Pac-Man scenario in two maps of increasing complexity. Our methodology produces a significant boost in the discounted return achieved by the agent, even in the first batches of training. Moreover, inductive learning does not compromise the computational time required by Q-learning and learned rules quickly converge to an explanation of the agent policy.

AI-based neural decoding reconstructs visual perception by leveraging generative models to map brain activity, measured through functional MRI (fMRI), into latent hierarchical representations. Traditionally, ridge linear models transform fMRI into a latent space, which is then decoded using latent diffusion models (LDM) via a pre-trained variational autoencoder (VAE). Due to the complexity and noisiness of fMRI data, newer approaches split the reconstruction into two sequential steps, the first one providing a rough visual approximation, the second on improving the stimulus prediction via LDM endowed by CLIP embeddings. This work proposes a non-linear deep network to improve fMRI latent space representation, optimizing the dimensionality alike. Experiments on the Natural Scenes Dataset showed that the proposed architecture improved the structural similarity of the reconstructed image by about 2\% with respect to the state-of-the-art model, based on ridge linear transform. The reconstructed image's semantics improved by about 4\%, measured by perceptual similarity, with respect to the state-of-the-art. The noise sensitivity analysis of the LDM showed that the role of the first stage was fundamental to predict the stimulus featuring high structural similarity. Conversely, providing a large noise stimulus affected less the semantics of the predicted stimulus, while the structural similarity between the ground truth and predicted stimulus was very poor. The findings underscore the importance of leveraging non-linear relationships between BOLD signal and the latent representation and two-stage generative AI for optimizing the fidelity of reconstructed visual stimuli from noisy fMRI data., Comment: 14 pages, 5 figures

The observed star formation rates of galaxies in the Local Universe suggests that they are replenishing their gas reservoir across cosmic time. Cosmological simulations predict that this accretion of fresh gas can occur in a hot or a cold mode, yet the existence of low column density ($\sim10^{17}$ cm$^{-2}$) neutral atomic hydrogen (HI) tracing the cold mode has not been unambiguously confirmed by observations. We present the application of unconstrained spectral stacking to attempt to detect the emission from this HI in the Circum-Galactic Medium (CGM) and Inter-Galactic Medium (IGM) of 6 nearby star forming galaxies from the MHONGOOSE sample for which full-depth observations are available. Our stacking procedure consists of a standard spectral stacking algorithm coupled with a one-dimensional spectral line finder designed to extract reliable signal close to the noise level. In agreement with previous studies, we found that the amount of signal detected outside the HI disk is much smaller than implied by simulations. Furthermore, the column density limit that we achieve via stacking ($\sim10^{17}$ cm$^{-2}$) suggests that direct detection of the neutral CGM/IGM component might be challenging in the future, even with the next generation of radio telescopes., Comment: Accepted for publication in Astronomy & Astrophysics

Soft robotic manipulators offer operational advantage due to their compliant and deformable structures. However, their inherently nonlinear dynamics presents substantial challenges. Traditional analytical methods often depend on simplifying assumptions, while learning-based techniques can be computationally demanding and limit the control policies to existing data. This paper introduces a novel approach to soft robotic control, leveraging state-of-the-art policy gradient methods within parallelizable synthetic environments learned from data. We also propose a safety oriented actuation space exploration protocol via cascaded updates and weighted randomness. Specifically, our recurrent forward dynamics model is learned by generating a training dataset from a physically safe \textit{mean reverting} random walk in actuation space to explore the partially-observed state-space. We demonstrate a reinforcement learning approach towards closed-loop control through state-of-the-art actor-critic methods, which efficiently learn high-performance behaviour over long horizons. This approach removes the need for any knowledge regarding the robot's operation or capabilities and sets the stage for a comprehensive benchmarking tool in soft robotics control., Comment: soft manipulator, reinforcement learning, learned controllers

The widespread use of machine learning in credit scoring has brought significant advancements in risk assessment and decision-making. However, it has also raised concerns about potential biases, discrimination, and lack of transparency in these automated systems. This tutorial paper performed a non-systematic literature review to guide best practices for developing responsible machine learning models in credit scoring, focusing on fairness, reject inference, and explainability. We discuss definitions, metrics, and techniques for mitigating biases and ensuring equitable outcomes across different groups. Additionally, we address the issue of limited data representativeness by exploring reject inference methods that incorporate information from rejected loan applications. Finally, we emphasize the importance of transparency and explainability in credit models, discussing techniques that provide insights into the decision-making process and enable individuals to understand and potentially improve their creditworthiness. By adopting these best practices, financial institutions can harness the power of machine learning while upholding ethical and responsible lending practices.

We present the discovery of a low-mass gas-rich low-surface brightness galaxy in the Dorado Group, at a distance of 17.7 Mpc. Combining deep MeerKAT 21-cm observations from the MeerKAT HI Observations of Nearby Galactic Objects: Observing Southern Emitters (MHONGOOSE) survey with deep photometric images from the VST Early-type Galaxy Survey (VEGAS) we find a stellar and neutral atomic hydrogen (HI) gas mass of $M_\star = 2.23\times10^6$ M$_\odot$ and $M_{\rm HI}=1.68\times10^6$ M$_\odot$, respectively. This low-surface brightness galaxy is the lowest mass HI detection found in a group beyond the Local Universe ($D\gtrsim 10$ Mpc). The dwarf galaxy has the typical overall properties of gas-rich low surface brightness galaxies in the Local group, but with some striking differences. Namely, the MHONGOOSE observations reveal a very low column density ($\sim 10^{18-19}$ cm$^{-2}$) HI disk with asymmetrical morphology possibly supported by rotation and higher velocity dispersion in the centre. There, deep optical photometry and UV-observations suggest a recent enhancement of the star formation. Found at galactocentric distances where in the Local Group dwarf galaxies are depleted of cold gas (at $390$ projected-kpc distance from the group centre), this galaxy is likely on its first orbit within the Dorado group. We discuss the possible environmental effects that may have caused the formation of the HI disk and the enhancement of star formation, highlighting the short-lived phase (a few hundreds of Myr) of the gaseous disk, before either SF or hydrodynamical forces will deplete the gas of the galaxy., Comment: 18 pages, 14 figures, 5 tables; accepted for publication in A&A

The MHONGOOSE (MeerKAT HI Observations of Nearby Galactic Objects: Observing Southern Emitters) survey maps the distribution and kinematics of the neutral atomic hydrogen (HI) gas in and around 30 nearby star-forming spiral and dwarf galaxies to extremely low HI column densities. The HI column density sensitivity (3 sigma over 16 km/s) ranges from ~ 5 x 10^{17} cm^{-2} at 90'' resolution to ~4 x 10^{19} cm^{-2} at the highest resolution of 7''. The HI mass sensitivity (3 sigma over 50 km/s) is ~5.5 X 10^5 M_sun at a distance of 10 Mpc (the median distance of the sample galaxies). The velocity resolution of the data is 1.4 km/s. One of the main science goals of the survey is the detection of cold, accreting gas in the outskirts of the sample galaxies. The sample was selected to cover a range in HI masses, from 10^7 M_sun to almost 10^{11} M_sun, to optimally sample possible accretion scenarios and environments. The distance to the sample galaxies ranges from 3 to 23 Mpc. In this paper, we present the sample selection, survey design, and observation and reduction procedures. We compare the integrated HI fluxes based on the MeerKAT data with those derived from single-dish measurement and find good agreement, indicating that our MeerKAT observations are recovering all flux. We present HI moment maps of the entire sample based on the first ten percent of the survey data, and find that a comparison of the zeroth- and second-moment values shows a clear separation between the physical properties of the HI in areas with star formation and areas without, related to the formation of a cold neutral medium. Finally, we give an overview of the HI-detected companion and satellite galaxies in the 30 fields, five of which have not previously been catalogued. We find a clear relation between the number of companion galaxies and the mass of the main target galaxy., Comment: Accepted for publication in Astronomy & Astrophysics

The microbunching instability has been a long-standing issue for high-brightness free-electron lasers (FELs), and is a significant show-stopper to achieving full longitudinal coherence in the x-ray regime. This paper reports the first experimental demonstration of microbunching instability mitigation through transverse Landau damping, based on linear optics control in a dispersive region. Analytical predictions for the microbunching content are supported by numerical calculations of the instability gain and confirmed through the experimental characterization of the spectral brightness of the FERMI FEL under different transverse optics configurations of the transfer line between the linear accelerator and the FEL.

The existing reservoirs of neutral atomic hydrogen gas (H$\,$I) in galaxies are insufficient to have maintained the observed levels of star formation without some kind of replenishment. {This refuelling of the H$\,$I reservoirs} is likely to occur at column densities an order of magnitude lower than previous observational limits (N$_{\rm{H\,I}\, limit} \sim 10^{19}\,$cm$^{-2}$ at 30$''$ resolution over a linewidth of $20\,$km/s). In this paper, we present recent deep H$\,$I observations of NGC 5068, a nearby isolated star-forming galaxy observed by MeerKAT as part of the MHONGOOSE survey. With these new data, we are able to detect low column density H$\,$I around NGC 5068 with a $3\sigma$ detection limit of N$_{\rm{H\,I}} = 6.4 \times 10^{17}\,$cm$^{-2}$ at 90$''$ resolution over a $20\,$km/s linewidth. The high sensitivity and resolution of the MeerKAT data reveal a complex morphology of the H$\,$I in this galaxy -- a regularly rotating inner disk coincident with the main star-forming disk of the galaxy, a warped outer disk of low column density gas (N$_{\rm{H\,I}} < 9 \times 10^{19}\,$cm$^{-2}$), in addition to clumps of gas on the north west side of the galaxy. We employ a simple two disk model that describe the inner and outer disks, and are able to identify anomalous gas that deviates from the rotation of the main galaxy. The morphology and the kinematics of the anomalous gas suggest a possible extra-galactic origin. We explore a number of possible origin scenarios that may explain the anomalous gas, and conclude that fresh accretion is the most likely scenario., Comment: 17 pages, 13 figures, 5 tables. Accepted for publication in A&A

We present a thorough study of the Changing-Look Active Galactic Nucleus (CL-AGN) Mrk 1018, utilizing an extensive dataset spanning optical, UV, and X-ray spectro-photometric data from 2005 to 2019. We analysed X-ray spectra and broad-band photometry, and performed optical-to-X-ray spectral energy distribution (SED) fitting to comprehend the observed changing-look behaviour. We found that over the 14 years in analysis, significant changes in X-ray spectra occurred, as the hardness ratio increases by a factor of ~2. We validated also the broad-band dimming, with optical, UV, and X-ray luminosities decreasing by factors of >7, >24 and ~9, respectively. These dims are attributed to the declining UV emission. We described the X-ray spectra with a two-Comptonization model, revealing a consistent hot comptonizing medium but a cooling warm component. This cooling, linked to the weakening of the magnetic fields in the accretion disk, explains the UV dimming. We propose that the weakening is caused by the formation of a jet, in turn originated from the change of state of the inner accretion flow. Our optical-to-X-ray SED fitting supports this conclusion, as the normalised accretion rate is super-critical ($\mu=$0.06>0.02) in the bright state and sub-critical ($\mu=$0.01<0.02) in the faint state. Instabilities arising at the interface of the state-transition are able to reduce the viscous timescale to the observed ~10 years of Mrk 1018 variability. We explored a possible triggering mechanism for this state transition, involving gaseous clouds pushed onto the AGN sub-pc regions by a recent merging event or by cold chaotic accretion. This scenario, if validated by future simulations, could enhance our understanding of CL-AGN and raises questions about an accretion rate of ~0.02, coupled with minor disturbances in the accretion disk, being the primary factor in the changing-look phenomenon., Comment: 18 pages, 8 figures

Gaussian Boson Sampling (GBS) is a recently developed paradigm of quantum computing consisting of sending a Gaussian state through a linear interferometer and then counting the number of photons in each output mode. When the system encodes a symmetric matrix, GBS can be viewed as a tool to sample subgraphs: the most sampled are those with a large number of perfect matchings, and thus are the densest ones. This property has been the foundation of the novel clustering approach we propose in this work, called GBS-based clustering, which relies solely on GBS, without the need of classical algorithms. The GBS-based clustering has been tested on several datasets and benchmarked with two well-known classical clustering algorithms. Results obtained by using a GBS simulator show that on average our approach outperforms the two classical algorithms in two out of the three chosen metrics, proposing itself as a viable full-quantum clustering option.

The occurrence of a noncollinear magnetic structure at a Mn monolayer grown epitaxially on Fe(100) is predicted theoretically, using spinor density-functional theory, and observed experimentally, using x-ray magnetic circular dichroism (XMCD) and linear dichroism (XMLD) spectroscopies. The combined use of XMCD and XMLD at the Mn-absorption edge allows us to assess the existence of ferromagnetic and antiferromagnetic order at the interface, and also to determine the moment orientations with element specificity. The experimental results thus obtained are in excellent agreement with the magnetic structure determined theoretically.

The Beam Driven Plasma Neutralizer (BDPN) has been proposed as a more efficient alternative to the gas neutralizer for negative-ion based Neutral Beam Injection (NNBI). In this paper we model the performance of an entire NNBI beamline with a BDPN. We simultaneously consider all the relevant physics and engineering aspects, the most important being the plasma density and degree of ionization inside the BDPN as a function of its geometry and feed gas flow, the geometrical transmission of the beamline, the dependence of the neutral gas distribution in the beamline on the geometry of the beamline components and gas flows, and the species evolution of the extracted D$^-$ beam through this neutral and charged particle distribution. Furthermore, we calculate the heat loads expected on the BDPN parts and on the NBI components located downstream of it and study the effect of the magnetic cusp field across the BDPN entrance on beamline transmission. While our results constitute an optimization only under the applied boundary conditions, we find that the beamline with a BDPN increases the system's wall plug efficiency by about 13% to 0.34 from the 0.30 estimated for a gas neutralizer., Comment: 16 pages, 14 figures. Published in Fusion Engineering and Design in occasion of the SOFT 2022 conference

ITER envisages the use of two heating neutral beam injectors plus an optional one as part of the auxiliary heating and current drive system. The 16.5 MW expected neutral beam power per injector is several notches higher than worldwide existing facilities. A Neutral Beam Test Facility (NBTF) was established at Consorzio RFX, exploiting the synergy of two test beds, SPIDER and MITICA. SPIDER is dedicated to developing and characterizing large efficient negative ion sources at relevant parameters in ITER-like conditions: source and accelerator located in the same vacuum where the beam propagates, immunity to electromagnetic interferences of multiple radio-frequency (RF) antennas, avoidance of RF-induced discharges on the outside of the source. Three years of experiments on SPIDER have addressed to the necessary design modifications to enable full performances. The source is presently under a long shut-down phase to incorporate learnings from the experimental campaign. Parallelly, developments on MITICA, the full-scale prototype of the ITER NBI featuring a 1 MV accelerator and ion neutralization, are underway including manufacturing of in-vessel components, while power supplies and auxiliary plants are already under final testing and commissioning. Integration, commissioning and tests of the 1MV power supplies are essential for this first-of-kind system, unparalleled both in research and industry field. The integrated test to confirm 1MV output by combining invertor systems, DC generators and transmission lines extracted errors/accidents in some components. To realize a concrete system for ITER, solutions for the repair and the improvement of the system were developed. Hence, NBTF is emerging as a necessary facility, due to the large gap with existing injectors, effectively dedicated to identify issues and find solutions to enable successful ITER NBI operations in a time bound fashion.