Search

Sparse autoencoders (SAEs) are an unsupervised method for learning a sparse decomposition of a neural network's latent representations into seemingly interpretable features. Despite recent excitement about their potential, research applications outside of industry are limited by the high cost of training a comprehensive suite of SAEs. In this work, we introduce Gemma Scope, an open suite of JumpReLU SAEs trained on all layers and sub-layers of Gemma 2 2B and 9B and select layers of Gemma 2 27B base models. We primarily train SAEs on the Gemma 2 pre-trained models, but additionally release SAEs trained on instruction-tuned Gemma 2 9B for comparison. We evaluate the quality of each SAE on standard metrics and release these results. We hope that by releasing these SAE weights, we can help make more ambitious safety and interpretability research easier for the community. Weights and a tutorial can be found at https://huggingface.co/google/gemma-scope and an interactive demo can be found at https://www.neuronpedia.org/gemma-scope, Comment: 12 main text pages, and 14 pages of acknowledgements, references and appendices

This visual narrative is a first-person reflection of the previous pictorial at the 1st International Workshop on Explainable AI for the Arts (XAIxArts) at ACM Creativity and Cognition 2023. The initial workshop pictorial explored a relationship between researcher and artificial intelligence, navigating creative challenges throughout the 2023 teaching block. It concluded by raising crucial questions regarding attribution transparency, the ethical dimensions of the creative process, and the delicate balance between inspiration and plagiarism. Subsequent discussions at the workshop yielded valuable insights, particularly concerning interpreting the creative journey. This follow-up visual narrative reflects the enduring impact of Makayla Lewis's interaction with AI. A self-portrait that delves into the interplay of creativity and introspection., Comment: 6 Pages, 7 Figures, Explainable AI for the Arts Workshop 2024 (XAIxArts 2024)

Axions are a well-motivated dark matter candidate particle. Haloscopes aim to detect axions in the galactic halo by measuring the photon signal resulting from axions interacting with a strong magnetic field. Existing haloscopes are primarily targeting axion masses which produce microwave-range photons and rely on microwave resonators to enhance the signal power. Only a limited subset of resonator modes are useful for this process, and current cylindrical-style cavities suffer from mode mixing and crowding from other fundamental modes. The majority of these modes can be eliminated by using photonic band gap (PBG) resonators. The band gap behavior of these structures allows for a resonator with mode selectivity based on frequency. We present results from the first tunable PBG resonator, a proof-of-concept design with a footprint compatible with axion haloscopes. We have thoroughly characterized the tuning range of two versions of the structure and report the successful confinement of the operating TM$_{010}$ mode and the elimination of all TE modes within the tuning range., Comment: 7 pages, 8 figures

We present the atmospheric characterization of the substellar companion HD 19467 B as part of the pioneering JWST GTO program to obtain moderate resolution spectra (R$\sim$2,700, 3-5$\mu$m) of a high-contrast companion with the NIRSpec IFU. HD 19467 B is an old, $\sim$9 Gyr, companion to a Solar-type star with multiple measured dynamical masses. The spectra show detections of CO, CO$_2$, CH$_4$, and H$_2$O. We forward model the spectra using Markov Chain Monte Carlo methods and atmospheric model grids to constrain the effective temperature and surface gravity. We then use NEWERA-PHOENIX grids to constrain non-equilibrium chemistry parameterized by $K_{zz}$ and explore molecular abundance ratios of the detected molecules. We find an effective temperature of 1103 K, with a probable range from 1000--1200 K, a surface gravity of 4.50 dex, with a range of 4.14--5.00, and deep vertical mixing, log$_{10}$($K_{zz}$), of 5.03, with a range of 5.00--5.44. All molecular mixing ratios are approximately Solar, leading to a C/O $\sim$0.55, which is expected from a T5.5 brown dwarf. Finally, we calculate an updated dynamical mass of HD 19467 B using newly derived NIRCam astrometry which we find to be $71.6^{+5.3}_{-4.6} M_{\rm{Jup}}$, in agreement with the mass range we derive from evolutionary models, which we find to be 63-75 $M_{\rm{Jup}}$.These observations demonstrate the excellent capabilities of the NIRSpec IFU to achieve detailed spectral characterization of substellar companions at high-contrast close to bright host stars, in this case at a separation of $\sim$1.6\arcsec with a contrast of 10$^{-4}$ in the 3-5 $\mu$m range., Comment: 35 pages, 20 figures

The explosion of artificial intelligence, possible end of Moore's law, dawn of quantum computing and continued exponential growth of data communications traffic have brought new urgency to the need for laser integration on the diversified Si platform. While diode lasers on III-V platforms have long powered internet data communications and other optoelectronic technologies, direct integration with Si remains problematic. A paradigm-shifting solution requires exploring new and unconventional materials and integration approaches. In this work, we show that a sub-10-nm ultra-thin Si$_{1-x}$Ge$_x$ buffer layer fabricated by an oxidative solid-phase epitaxy process can facilitate extraordinarily efficient strain relaxation. The Si$_{1-x}$Ge$_x$ layer is formed by ion implanting Ge into Si(111) and selectively oxidizing Si atoms in the resulting ion-damaged layer, precipitating a fully strain-relaxed Ge-rich layer between the Si substrate and surface oxide. The efficient strain relaxation results from the high oxidation temperature, producing a periodic network of dislocations at the substrate interface, coinciding with modulations of the Ge content in the Si$_{1-x}$Ge$_x$ layer and indicating the presence of defect-mediated diffusion of Si through the layer. The epitaxial growth of high-quality GaAs is demonstrated on this ultra-thin Si$_{1-x}$Ge$_x$ layer, demonstrating a promising new pathway for integrating III-V lasers directly on the Si platform., Comment: 16 pages, 5 figures

The formation of massive objects via gravitational collapse is relevant both for explaining the origin of the first supermassive black holes and in the context of massive star formation. Here, we analyze simulations of the formation of massive objects pursued by different groups and in various environments, concerning the formation of supermassive black holes, primordial stars, as well as present-day massive stars. We focus particularly on the regime of small virial parameters, i.e., low ratios of the initial kinetic to gravitational energy, low to moderate Mach numbers, and the phase before feedback is very efficient. We compare the outcomes of collapse under different conditions using dimensionless parameters, particularly the star formation efficiency \epsilon_*, the fraction f_* of mass in the most massive object relative to the total stellar mass, and the fraction f_{\rm tot} of mass of the most massive object as a function of the total mass. We find that in all simulations analyzed here, f_{\rm tot} increases as a function of \epsilon_*, although the steepness of the increase depends on the environment. The relation between f_* and \epsilon_* is found to be more complex and also strongly depends on the number of protostars present at the beginning of the simulations. We show that a collision parameter, estimated as the ratio of the system size divided by the typical collision length, allows us to approximately characterize whether collisions will be important. We analyze the statistical correlation between the dimensionless quantities using the Spearman coefficient and confirm via a machine learning analysis that good predictions of f_* can be obtained from \epsilon_* together with a rough estimate of the collision parameter. This suggests that a good estimate of the mass of the most massive object can be obtained once the maximum efficiency for a given environment is known., Comment: 13 pages, 8 figures, 2 tables. Submitted to A&A

Knowing accurate joint positions is crucial for safe and precise control of laparoscopic surgical robots, especially for the automation of surgical sub-tasks. These robots have often been designed with cable-driven arms and tools because cables allow for larger motors to be placed at the base of the robot, further from the operating area where space is at a premium. However, by connecting the joint to its motor with a cable, any stretch in the cable can lead to errors in kinematic estimation from encoders at the motor, which can result in difficulties for accurate control of the surgical tool. In this work, we propose an efficient data-driven calibration of positioning joints of such robots, in this case the RAVEN-II surgical robotics research platform. While the calibration takes only 8-21 minutes, the accuracy of the calibrated joints remains high during a 6-hour heavily loaded operation, suggesting desirable feasibility in real practice. The calibration models take original robot states as input and are trained using zig-zag trajectories within a desired sparsity, requiring no additional sensors after training. Compared to fixed offset compensation, the Deep Neural Network calibration model can further reduce 76 percent of error and achieve accuracy of 0.104 deg, 0.120 deg, and 0.118 mm in joints 1, 2, and 3, respectively. In contrast to end-to-end models, experiments suggest that the DNN model achieves better accuracy and faster convergence when outputting the error to correct original inaccurate joint positions. Furthermore, a linear regression model is shown to have 160 times faster inference speed than DNN models for application within the 1000 Hz servo control loop, with slightly compromised accuracy.

We present the joint astrometric and direct imaging discovery, mass measurement, and orbital analysis of HD 63754 B (HIP 38216 B), a companion near the stellar-substellar boundary orbiting ~20 AU from its Sun-like host. HD 63754 was observed in our ongoing high-contrast imaging survey targeting stars with significant proper-motion accelerations between Hipparcos and Gaia consistent with wide-separation substellar companions. We utilized archival HIRES and HARPS radial velocity (RV) data, together with the host star's astrometric acceleration extracted from the Hipparcos-Gaia Catalog of Accelerations (HGCA), to predict the location of the candidate companion around HD 63754 A. We subsequently imaged HD 63754 B at its predicted location using the Near Infrared Camera 2 (NIRC2) in the $L'$ band at the W. M. Keck Observatory. We then jointly modeled the orbit of HD 63754 B with RVs, Hipparcos-Gaia accelerations, and our new relative astrometry, measuring a dynamical mass of ${81.9}_{-5.8}^{+6.4} M_{jup}$, an eccentricity of ${0.260}_{-0.059}^{+0.065}$, and a nearly face-on inclination of $174.81_{-0.50}^{+0.48}$ degrees. For HD 63754 B, we obtain an L' band absolute magnitude of $L' = 11.39\pm0.06$ mag, from which we infer a bolometric luminosity of $log(L_{bol}/L_{\odot})= -4.55 \pm0.08$ dex using a comparison sample of L and T dwarfs with measured luminosities. Although uncertainties linger in age and dynamical mass estimates, our analysis points toward HD 63754 B's identity as a brown dwarf on the L/T transition rather than a low-mass star, indicated by its inferred bolometric luminosity and model-estimated effective temperature. Future RV, spectroscopic, and astrometric data such as those from JWST and Gaia DR4 will clarify HD 63754 B's mass, and enable spectral typing and atmospheric characterization.

This paper, and its companion [BCV24], are devoted to a negative resolution of the Aldous--Lyons Conjecture [AL07, Ald07]. This conjecture, originated in probability theory, is well known (cf. [Gel18]) to be equivalent to the statement that every invariant random subgroup of the free group is co-sofic. We disprove this last statement. In this part we introduce subgroup tests. These tests are finite distributions over continuous functions from the space of subgroups of the free group to $\{0,1\}$. Subgroup tests provide a general framework in which one can study invariant random subgroups of the free group. Classical notions such as group soficity and group stability arise naturally in this framework. By the correspondence between subgroups of the free group and Schreier graphs, one can view subgroup tests as a property testing model for certain edge-labeled graphs. This correspondence also provides the connection to random networks. Subgroup tests have values, which are their asymptotic optimal expectations when integrated against co-sofic invariant random subgroups. Our first main result is that, if every invariant random subgroup of the free group is co-sofic, then one can approximate the value of a subgroup test up to any positive additive constant. Our second main result is an essentially value preserving correspondence between certain non-local games and subgroup tests. By composing this correspondence with a stronger variant of the reduction in MIP*=RE [JNV+21], proved in the companion paper [BCV24], we deduce that approximating the sofic value of a subgroup test is as hard as the Halting Problem, and in particular, undecidable. The combination of our two main results proves the existence of non co-sofic invariant random subgroups of the free group., Comment: 54 pages

We introduce MoMa, a novel modality-aware mixture-of-experts (MoE) architecture designed for pre-training mixed-modal, early-fusion language models. MoMa processes images and text in arbitrary sequences by dividing expert modules into modality-specific groups. These groups exclusively process designated tokens while employing learned routing within each group to maintain semantically informed adaptivity. Our empirical results reveal substantial pre-training efficiency gains through this modality-specific parameter allocation. Under a 1-trillion-token training budget, the MoMa 1.4B model, featuring 4 text experts and 4 image experts, achieves impressive FLOPs savings: 3.7x overall, with 2.6x for text and 5.2x for image processing compared to a compute-equivalent dense baseline, measured by pre-training loss. This outperforms the standard expert-choice MoE with 8 mixed-modal experts, which achieves 3x overall FLOPs savings (3x for text, 2.8x for image). Combining MoMa with mixture-of-depths (MoD) further improves pre-training FLOPs savings to 4.2x overall (text: 3.4x, image: 5.3x), although this combination hurts performance in causal inference due to increased sensitivity to router accuracy. These results demonstrate MoMa's potential to significantly advance the efficiency of mixed-modal, early-fusion language model pre-training, paving the way for more resource-efficient and capable multimodal AI systems., Comment: v2 -> update related work section v3 -> fix spelling

Plasma in the terrestrial magnetosheath is characterised by very weak particle-particle collisions, so kinetic microinstabilities are thought to be responsible for regulating the thermodynamics of the plasma. By exciting electromagnetic waves, these instabilities redistribute free energy in velocity space, moulding the velocity distribution function (VDF) into a lower energy state. In the high-beta magnetosheath, relatively small perturbations to the VDF can easily excite instabilities compared to in the low-beta inner heliosphere. Since magnetic fields cannot do work on the particles, electric fields mediate energy exchange between the electromagnetic field and the bulk fluid properties of the plasma. We investigate signatures of non-ideal energy conversion associated with turbulent fluctuations in the context of electron and ion temperature anisotropy-beta instabilities, utilising over 24 hours of data spread over 163 distinct intervals of in situ magnetosheath observations from Magnetospheric Multiscale (MMS). We find that average energy conversion into fluid flow is enhanced along instability boundaries, suggesting that turbulence is playing a role in how free energy is redistributed in the plasma. The work enables a quantification of the energetics which are associated with the role of kinetic microinstabilities in regulating collisionless plasma thermodynamics. This work provides insight into the open question of how specific plasma processes couple into the turbulent dynamics and ultimately lead to energy dissipation and particle energisation in collisionless plasmas., Comment: 10 pages, 3 figures, 1 table. Submitted for peer review at Astrophysical Journal Letters (ApJL), July 2024

We give analogues in the finite general linear group of two elementary results concerning long cycles and transpositions in the symmetric group: first, that the long cycles are precisely the elements whose minimum-length factorizations into transpositions yield a generating set, and second, that a long cycle together with an appropriate transposition generates the whole symmetric group., Comment: 12 pages

We describe several omissions and deficiencies in the arguments presented by Ackermann et al. (arXiv:2401.07684), pointing at missing information necessary to back up its conclusions., Comment: Comment on arXiv:2401.07684. Two pages. V2: few additions in footnotes

Real-world physical systems, like composite materials and porous media, exhibit complex heterogeneities and multiscale nature, posing significant computational challenges. Computational homogenization is useful for predicting macroscopic properties from the microscopic material constitution. It involves defining a representative volume element (RVE), solving governing equations, and evaluating its properties such as conductivity and elasticity. Despite its effectiveness, the approach can be computationally expensive. This study proposes a tensor-train (TT)-based asymptotic homogenization method to address these challenges. By deriving boundary value problems at the microscale and expressing them in the TT format, the proposed method estimates material properties efficiently. We demonstrate its validity and effectiveness through numerical experiments applying the proposed method for homogenization of thermal conductivity and elasticity in two- and three-dimensional materials, offering a promising solution for handling the multiscale nature of heterogeneous systems., Comment: 25 pages, 7 figures

We present a measurement of $|V_{ub}|$ from a simultaneous study of the charmless semileptonic decays $B^0\to\pi^- \ell^+ \nu_{\ell}$ and $B^+\to\rho^0 \ell^+\nu_{\ell}$, where $\ell = e, \mu$. This measurement uses a data sample of 387 million $B\overline{B}$ meson pairs recorded by the Belle~II detector at the SuperKEKB electron-positron collider between 2019 and 2022. The two decays are reconstructed without identifying the partner $B$ mesons. We simultaneously measure the differential branching fractions of $B^0\to\pi^- \ell^+ \nu_{\ell}$ and $B^+\to\rho^0 \ell^+\nu_{\ell}$ decays as functions of $q^2$ (momentum transfer squared). From these, we obtain total branching fractions $B(B^0\to\pi^- \ell^+ \nu_{\ell}) = (1.516 \pm 0.042 (\mathrm{stat}) \pm 0.059 (\mathrm{syst})) \times 10^{-4}$ and $B(B^+\to\rho^0 \ell^+\nu_{\ell}) = (1.625 \pm 0.079 (\mathrm{stat}) \pm 0.180 (\mathrm{syst})) \times 10^{-4}$. By fitting the measured $B^0\to\pi^- \ell^+ \nu_{\ell}$ partial branching fractions as functions of $q^2$, together with constraints on the non-perturbative hadronic contribution from lattice QCD calculations, we obtain $|V_{ub}|$ = $(3.93 \pm 0.09 \pm 0.13 \pm 0.19) \times 10^{-3}$. Here, the first uncertainty is statistical, the second is systematic, and the third is theoretical.

Brown dwarfs (BD) are model degenerate in age and mass. High-contrast imaging and spectroscopy of BD companions to host stars where the mass and age can be independently constrained by dynamics and stellar age indicators respectively provide valuable tests of BD evolution models. In this paper, we present a new epoch of Subaru/CHARIS H- and K-band observations of one such previously discovered system, HD 33632 Ab. We reanalyze the mass and orbit using our new epoch of extracted relative astrometry, and fit extracted spectra to the newest generation of equilibrium, disequilibrium, and cloudy spectral and evolution models for BDs. No spectral model perfectly agrees with evolutionary tracks and the derived mass and age, instead favoring a somewhat younger BD than the host star's inferred age. This tension can potentially be resolved using atmosphere and evolution models that consider both clouds and disequilibrium chemistry simultaneously, or by additional future spectra at higher resolution or in other band passes. Photometric measurements alone remain consistent with the luminosity predicted by evolutionary tracks. Our work highlights the importance of considering complexities like clouds, disequilibrium chemistry, and composition when comparing spectral models to evolutionary tracks., Comment: 15 pages, 7 figures, accepted for publication in The Astrophysical Journal

The rapidly growing field of integrated photonics is enabling a large number of novel devices for optical data processing, neuromorphic computing and circuits for quantum photonics. While many photonic devices are based on linear optics, nonlinear responses at low threshold power are of high interest for optical switching and computing. In the case of counterpropagating light, nonlinear interactions can be utilized for chip-based isolators and logic gates. In our work we find a symmetry breaking of the phases of counterpropagating light waves in high-Q ring resonators. This abrupt change in the phases can be used for optical switches and logic gates. In addition to our experimental results, we provide theoretical models that describe the phase symmetry breaking of counterpropagating light in ring resonators., Comment: 11 pages, 9 figures

Physical infrastructure systems supply crucial resources to residential, commercial, and industrial activities. These infrastructure systems generally consist of multiple types of infrastructure assets that are interdependent. In the event of a disaster, some of the infrastructure assets can be damaged and disabled, creating failures that propagate to other assets that depend on the disabled assets and cause a cascade of failures that may lead to a potential system collapse. We present a bilevel interdiction model in this paper to study this problem of cascading failures in a system of interdependent infrastructure systems with a nondeterministic dependency graph. We also propose a computationally tractable reformulation of the proposed bilevel model and utilize a Benders-type decomposition algorithm to solve the resulting formulation. Computational experiments are performed using infrastructure networks generated from anonymized real-world data to validate the performance of this algorithm.

In the last decade, remarkable advances in integrated photonic technologies have enabled table-top experiments and instrumentation to be scaled down to compact chips with significant reduction in size, weight, power consumption, and cost. Here, we demonstrate an integrated continuously tunable laser in a heterogeneous gallium arsenide-on-silicon nitride (GaAs-on-SiN) platform that emits in the far-red radiation spectrum near 780 nm, with 20 nm tuning range, <6 kHz intrinsic linewidth, and a >40 dB side-mode suppression ratio. The GaAs optical gain regions are heterogeneously integrated with low-loss SiN waveguides. The narrow linewidth lasing is achieved with an extended cavity consisting of a resonator-based Vernier mirror and a phase shifter. Utilizing synchronous tuning of the integrated heaters, we show mode-hop-free wavelength tuning over a range larger than 100 GHz (200 pm). To demonstrate the potential of the device, we investigate two illustrative applications: (i) the linear characterization of a silicon nitride microresonator designed for entangled-photon pair generation, and (ii) the absorption spectroscopy and locking to the D1 and D2 transition lines of 87-Rb. The performance of the proposed integrated laser holds promise for a broader spectrum of both classical and quantum applications in the visible range, encompassing communication, control, sensing, and computing.

The 16 Myr-old A0V star HD 156623 in the Scorpius--Centaurus association hosts a high-fractional-luminosity debris disk, recently resolved in scattered light for the first time by the Gemini Planet Imager (GPI) in polarized intensity. We present new analysis of the GPI H-band polarimetric detection of the HD 156623 debris disk, with particular interest in its unique morphology. This debris disk lacks a visible inner clearing, unlike the majority of low-inclination disks in the GPI sample and in Sco-Cen, and it is known to contain CO gas, positioning it as a candidate ``hybrid'' or ``shielded'' disk. We use radiative transfer models to constrain the geometric parameters of the disk based on scattered light data and thermal models to constrain the unresolved inner radius based on the system's spectral energy distribution (SED). We also compute a measurement of the polarized scattering phase function, adding to the existing sample of empirical phase function measurements. We find that HD 156623's debris disk inner radius is constrained to less than 26.6 AU from scattered light imagery and less than 13.4 AU from SED modeling at a 99.7% confidence interval, and suggest that gas drag may play a role in retaining sub-blowout size dust grains so close to the star., Comment: 18 pages, 10 figures. Accepted for publication in AJ

AI progress is creating a growing range of risks and opportunities, but it is often unclear how they should be navigated. In many cases, the barriers and uncertainties faced are at least partly technical. Technical AI governance, referring to technical analysis and tools for supporting the effective governance of AI, seeks to address such challenges. It can help to (a) identify areas where intervention is needed, (b) identify and assess the efficacy of potential governance actions, and (c) enhance governance options by designing mechanisms for enforcement, incentivization, or compliance. In this paper, we explain what technical AI governance is, why it is important, and present a taxonomy and incomplete catalog of its open problems. This paper is intended as a resource for technical researchers or research funders looking to contribute to AI governance., Comment: Ben Bucknall and Anka Reuel contributed equally and share the first author position

Responsible AI (RAI) is the science and the practice of making the design, development, and use of AI socially sustainable: of reaping the benefits of innovation while controlling the risks. Naturally, industry practitioners play a decisive role in our collective ability to achieve the goals of RAI. Unfortunately, we do not yet have consolidated educational materials and effective methodologies for teaching RAI to practitioners. In this paper, we propose a novel stakeholder-first educational approach that uses interactive case studies to achieve organizational and practitioner -level engagement and advance learning of RAI. We discuss a partnership with Meta, an international technology company, to co-develop and deliver RAI workshops to a diverse audience within the company. Our assessment results indicate that participants found the workshops engaging and reported a positive shift in understanding and motivation to apply RAI to their work.

The majority of Ethical Hacking (EH) tools utilised in penetration testing are developed by practitioners within the industry or underground communities. Similarly, academic researchers have also contributed to developing security tools. However, there appears to be limited awareness among practitioners of academic contributions in this domain, creating a significant gap between industry and academia's contributions to EH tools. This research paper aims to survey the current state of EH academic research, primarily focusing on research-informed security tools. We categorise these tools into process-based frameworks (such as PTES and Mitre ATT\&CK) and knowledge-based frameworks (such as CyBOK and ACM CCS). This classification provides a comprehensive overview of novel, research-informed tools, considering their functionality and application areas. The analysis covers licensing, release dates, source code availability, development activity, and peer review status, providing valuable insights into the current state of research in this field., Comment: This is the extended version of the paper published in the Journal of Cybersecurity and Privacy, 4, no. 3: pp 410-448, 2024

In the drug discovery process, where experiments can be costly and time-consuming, computational models that predict drug-target interactions are valuable tools to accelerate the development of new therapeutic agents. Estimating the uncertainty inherent in these neural network predictions provides valuable information that facilitates optimal decision-making when risk assessment is crucial. However, such models can be poorly calibrated, which results in unreliable uncertainty estimates that do not reflect the true predictive uncertainty. In this study, we compare different metrics, including accuracy and calibration scores, used for model hyperparameter tuning to investigate which model selection strategy achieves well-calibrated models. Furthermore, we propose to use a computationally efficient Bayesian uncertainty estimation method named Bayesian Linear Probing (BLP), which generates Hamiltonian Monte Carlo (HMC) trajectories to obtain samples for the parameters of a Bayesian Logistic Regression fitted to the hidden layer of the baseline neural network. We report that BLP improves model calibration and achieves the performance of common uncertainty quantification methods by combining the benefits of uncertainty estimation and probability calibration methods. Finally, we show that combining post hoc calibration method with well-performing uncertainty quantification approaches can boost model accuracy and calibration.

Overfitting describes a machine learning phenomenon where the model fits too closely to the training data, resulting in poor generalization. While this occurrence is thoroughly documented for many forms of supervised learning, it is not well examined in the context of unsupervised learning. In this work we examine the nature of overfitting in unsupervised contrastive learning. We show that overfitting can indeed occur and the mechanism behind overfitting.

We report high-precision multi-wavelength linear-polarization observations of the bright B9 (or A0) star $\epsilon$ Sagittarii. The polarization shows the distinctive wavelength dependence expected for a rapidly rotating star. Analysis of the polarization data reveals an angular rotation rate $\omega$ (= $\Omega/\Omega_{crit})$ of 0.995 or greater, the highest yet measured for a star in our galaxy. An additional wavelength-independent polarization component is attributed to electron scattering in a low-density edge-on gas disk that also produces the narrow absorption components seen in the spectrum. Several properties of the star (polarization due to a disk, occasional weak H$\alpha$ emission, and multiple periodicities seen in space photometry) resemble those of Be stars, but the level of activity in all cases is much lower than that of typical Be stars. The stellar properties are inconsistent with single rotating-star evolutionary tracks, indicating that it is most likely a product of binary interaction. The star is an excellent candidate for observation by interferometry, optical spectropolarimetry to detect the \"{O}hman effect, and UV polarimetry; any of which would allow its extreme rotation to be tested and its stellar properties to be refined., Comment: Accepted by Astrophysical Journal, 18 pages, 13 figures

We show that the loop space of a moment-angle complex associated to a $2$-dimensional simplicial complex decomposes as a finite type product of spheres, loops on spheres, and certain indecomposable spaces which appear in the loop space decomposition of Moore spaces. We also give conditions on certain subcomplexes under which, localised away from sufficiently many primes, the loop space of a moment-angle complex decomposes as a finite type product of spheres and loops on spheres., Comment: 22 pages, comments welcome

To operate in open-ended environments where humans interact in complex, diverse ways, autonomous robots must learn to predict their behaviour, especially when that behavior is potentially dangerous to other agents or to the robot. However, reducing the risk of accidents requires prior knowledge of where potential collisions may occur and how. Therefore, we propose to gain this information by analyzing locations and speeds that commonly correspond to high-risk interactions within the dataset, and use it within training to generate better predictions in high risk situations. Through these location-based and speed-based re-weighting techniques, we achieve improved overall performance, as measured by most-likely FDE and KDE, as well as improved performance on high-speed vehicles, and vehicles within high-risk locations. 2023 IEEE International Conference on Robotics and Automation (ICRA)

Transmission spectroscopy has been a workhorse technique over the past two decades to constrain the physical and chemical properties of exoplanet atmospheres. One of its classical key assumptions is that the portion of the atmosphere it probes -- the terminator region -- is homogeneous. Several works in the past decade, however, have put this into question for highly irradiated, hot ($T_{eq}\gtrsim 1000$ K) gas giant exoplanets both empirically and via 3-dimensional modelling. While models predict clear differences between the evening (day-to-night) and morning (night-to-day) terminators, direct morning/evening transmission spectra in a wide wavelength range has not been reported for an exoplanet to date. Under the assumption of precise and accurate orbital parameters on WASP-39 b, here we report the detection of inhomogeneous terminators on the exoplanet WASP-39 b, which allows us to retrieve its morning and evening transmission spectra in the near-infrared ($2-5\ \mu$m) using JWST. We observe larger transit depths in the evening which are, on average, $405 \pm 88$ ppm larger than the morning ones, also having qualitatively larger features than the morning spectrum. The spectra are best explained by models in which the evening terminator is hotter than the morning terminator by $177^{+65}_{-57}$ K with both terminators having C/O ratios consistent with solar. General circulation models (GCMs) predict temperature differences broadly consistent with the above value and point towards a cloudy morning terminator and a clearer evening terminator., Comment: Published in Nature at https://www.nature.com/articles/s41586-024-07768-4. All code to produce plots (with data) can be found at https://github.com/nespinoza/wasp39-terminators

Recent work has shown that certain integrable and conformal field theories in two dimensions can be given a higher-dimensional origin from holomorphic Chern-Simons in six dimensions. Along with anti-self-dual Yang-Mills and four-dimensional Chern-Simons, this gives rise to a diamond correspondence of theories. In this work we extend this framework to incorporate models realised through gaugings. As well as describing a higher-dimensional origin of coset CFTs, by choosing the details of the reduction from higher dimensions, we obtain rich classes of two-dimensional integrable models including homogeneous sine-Gordon models and generalisations that are new to the literature., Comment: 44 pages

Various reversible deep neural networks (DNN) models have been proposed to reduce memory consumption in the training process. However, almost all existing reversible DNNs either require special non-standard architectures or are constructed by modifying existing DNN architectures considerably to enable reversibility. In this work we present the BDIA-transformer, which is an exact bit-level reversible transformer that uses an unchanged standard architecture for inference. The basic idea is to first treat each transformer block as the Euler integration approximation for solving an ordinary differential equation (ODE) and then incorporate the technique of bidirectional integration approximation (BDIA) into the neural architecture, together with activation quantization to make it exactly bit-level reversible. In the training process, we let a hyper-parameter $\gamma$ in BDIA-transformer randomly take one of the two values $\{0.5, -0.5\}$ per training sample per transformer block for averaging every two consecutive integration approximations. As a result, BDIA-transformer can be viewed as training an ensemble of ODE solvers parameterized by a set of binary random variables, which regularizes the model and results in improved validation accuracy. Lightweight side information per transformer block is required to be stored in the forward process to account for binary quantization loss to enable exact bit-level reversibility. In the inference procedure, the expectation $\mathbb{E}(\gamma)=0$ is taken to make the resulting architectures of BDIA-transformer identical to transformers up to activation quantization. Our experiments in both image classification and language translation show that BDIA-transformers outperform their conventional counterparts significantly in terms of validation performance while also requiring considerably less training memory.

We present measurements of $B^{+}\rightarrow\rho^{+}\gamma$ and $B^{0}\rightarrow\rho^{0}\gamma$ decays using a combined data sample of $772 \times 10^6$ $B\overline{B}$ pairs collected by the Belle experiment and $387\times 10^6$ $B\overline{B}$ pairs collected by the Belle II experiment in $e^{+}e^{-}$ collisions at the $\Upsilon (4S)$ resonance. After an optimized selection, a simultaneous fit to the Belle and Belle II data sets yields $114\pm 12$ $B^{+}\rightarrow\rho^{+}\gamma$ and $99\pm 12$ $B^{0}\rightarrow\rho^{0}\gamma$ decays. The measured branching fractions are $(13.1^{+2.0 +1.3}_{-1.9 -1.2})\times 10^{-7}$ and $(7.5\pm 1.3^{+1.0}_{-0.8})\times 10^{-7}$ for $B^{+}\rightarrow\rho^{+}\gamma$ and $B^{0}\rightarrow\rho^{0}\gamma$ decays, respectively, where the first uncertainty is statistical and the second is systematic. We also measure the isospin asymmetry $A_{\rm I}(B\rightarrow\rho\gamma)=(10.9^{+11.2 +7.8}_{-11.7 -7.3})\%$ and the direct CP asymmetry $A_{CP}(B^{+}\rightarrow\rho^{+}\gamma)=(-8.2\pm 15.2^{+1.6}_{-1.2})\%$., Comment: 12 pages, 4 figures

We update earlier lattice results for the binding energies of the flavor antitriplet of strong-interaction-stable doubly bottom, $J^P=1^+$ tetraquarks, employing an extended sink construction which produces significantly improved ground-state effective-mass plateaus, as well as new, larger-volume ensembles which reduce possible finite-volume effects at lighter pion masses. The updated bindings are $115(17)$ MeV for the $I=0$ member of the antitriplet and $47(8)$ MeV for its $I=1/2$ partner. We also provide an update of our earlier study of the variable heavy mass dependence of binding in the $1^+$ channel and new results on this dependence for binding in the $0^+$ channel, accessible when the two heavy quarks have unequal masses. Implications of these results of potential relevance to experimental searches for signals of the production of doubly bottom tetraquarks and/or a possible bottom-charm partner of the $T_{cc}$ are also discussed., Comment: 27 pages, 7 figures. This version accepted by PRD

This paper presents a new perspective on integrability in theories of gravity. We show how the stationary, axisymmetric sector of General Relativity can be described by the boundary dynamics of a four-dimensional Chern-Simons theory. This approach shows promise for simplifying solution generating methods in both General Relativity and higher-dimensional supergravity theories. The four-dimensional Chern-Simons theory presented generalises those for flat space integrable models by introducing a space-time dependent branch cut in the spectral plane. We also make contact with twistor space approaches to integrability, showing how the branch cut defects of four-dimensional Chern-Simons theory arise from a discrete reduction of six-dimensional Chern-Simons theory., Comment: 41 pages

The PHENIX experiment measured the centrality dependence of two-pion Bose-Einstein correlation functions in $\sqrt{s_{_{NN}}}=200$~GeV Au$+$Au collisions at the Relativistic Heavy Ion Collider at Brookhaven National Laboratory. The data are well represented by L\'evy-stable source distributions. The extracted source parameters are the correlation-strength parameter $\lambda$, the L\'evy index of stability $\alpha$, and the L\'evy-scale parameter $R$ as a function of transverse mass $m_T$ and centrality. The $\lambda(m_T)$ parameter is constant at larger values of $m_T$, but decreases as $m_T$ decreases. The L\'evy scale parameter $R(m_T)$ decreases with $m_T$ and exhibits proportionality to the length scale of the nuclear overlap region. The L\'evy exponent $\alpha(m_T)$ is independent of $m_T$ within uncertainties in each investigated centrality bin, but shows a clear centrality dependence. At all centralities, the L\'evy exponent $\alpha$ is significantly different from that of Gaussian ($\alpha=2$) or Cauchy ($\alpha=1$) source distributions. Comparisons to the predictions of Monte-Carlo simulations of resonance-decay chains show that in all but the most peripheral centrality class (50%-60%), the obtained results are inconsistent with the measurements, unless a significant reduction of the in-medium mass of the $\eta'$ meson is included. In each centrality class, the best value of the in-medium $\eta'$ mass is compared to the mass of the $\eta$ meson, as well as to several theoretical predictions that consider restoration of $U_A(1)$ symmetry in hot hadronic matter., Comment: 401 authors from 75 institutions, 20 pages, 15 figures, 2 tables. v1 is version submitted to Physical Review C. HEPdata tables for the points plotted in figures for this and previous PHENIX publications are (or will be) publicly available at http://www.phenix.bnl.gov/papers.html

An elusive goal in navigation research is to build an intelligent agent that can understand multimodal instructions including natural language and image, and perform useful navigation. To achieve this, we study a widely useful category of navigation tasks we call Multimodal Instruction Navigation with demonstration Tours (MINT), in which the environment prior is provided through a previously recorded demonstration video. Recent advances in Vision Language Models (VLMs) have shown a promising path in achieving this goal as it demonstrates capabilities in perceiving and reasoning about multimodal inputs. However, VLMs are typically trained to predict textual output and it is an open research question about how to best utilize them in navigation. To solve MINT, we present Mobility VLA, a hierarchical Vision-Language-Action (VLA) navigation policy that combines the environment understanding and common sense reasoning power of long-context VLMs and a robust low-level navigation policy based on topological graphs. The high-level policy consists of a long-context VLM that takes the demonstration tour video and the multimodal user instruction as input to find the goal frame in the tour video. Next, a low-level policy uses the goal frame and an offline constructed topological graph to generate robot actions at every timestep. We evaluated Mobility VLA in a 836m^2 real world environment and show that Mobility VLA has a high end-to-end success rates on previously unsolved multimodal instructions such as "Where should I return this?" while holding a plastic bin. A video demonstrating Mobility VLA can be found here: https://youtu.be/-Tof__Q8_5s

This paper proves that the 2-Calabi-Yau triangulated category associated with the preprojective algebra of an affine or hyperbolic graph admits many real variations of stability conditions, in the sense of Anno, Bezrukavnikov, and Mirkovi\'{c}. We do this by connecting the Coxeter arrangements of the graph with the homological algebra by introducing the concept of real flows. This categorifies the notion of above and below for alcoves in a hyperplane arrangement and allows us to generalise much of the machinery of real variations for affine arrangements to the hyperbolic setting. In the process, we show that the derived category with nilpotent cohomology is equivalent to the derived category of nilpotent modules over the preprojective algebra., Comment: 27 pages

Periodic activation functions, often referred to as learned Fourier features have been widely demonstrated to improve sample efficiency and stability in a variety of deep RL algorithms. Potentially incompatible hypotheses have been made about the source of these improvements. One is that periodic activations learn low frequency representations and as a result avoid overfitting to bootstrapped targets. Another is that periodic activations learn high frequency representations that are more expressive, allowing networks to quickly fit complex value functions. We analyse these claims empirically, finding that periodic representations consistently converge to high frequencies regardless of their initialisation frequency. We also find that while periodic activation functions improve sample efficiency, they exhibit worse generalization on states with added observation noise -- especially when compared to otherwise equivalent networks with ReLU activation functions. Finally, we show that weight decay regularization is able to partially offset the overfitting of periodic activation functions, delivering value functions that learn quickly while also generalizing.

We attempt to compute expressions in terms of the Dedekind eta function for all weight 2 new cusp forms with level up to 100, using methods of Allen et. al. In cases where no expression exists, we raise the level instead of the weight, meaning our eta quotients are always holomorphic. Of the forms we examine, we find expressions for all but 4. We also present methods to find expressions with relatively few terms, and how these expressions can be used to demonstrate zeroes of modular forms.

We present a search for the baryon number $B$ and lepton number $L$ violating decays $\tau^- \rightarrow \Lambda \pi^-$ and $\tau^- \rightarrow \bar{\Lambda} \pi^-$ produced from the $e^+e^-\to \tau^+\tau^-$ process, using a 364 fb$^{-1}$ data sample collected by the Belle~II experiment at the SuperKEKB collider. No evidence of signal is found in either decay mode, which have $|\Delta(B-L)|$ equal to $2$ and $0$, respectively. Upper limits at 90\% credibility level on the branching fractions of $\tau^- \rightarrow \Lambda\pi^-$ and $\tau^- \rightarrow \bar{\Lambda}\pi^-$ are determined to be $4.7 \times 10^{-8}$ and $4.3 \times 10^{-8}$, respectively., Comment: 8 pages, 4 figures

The increasing integration of Artificial Intelligence (AI) into everyday life makes it essential to explain AI-based decision-making in a way that is understandable to all users, including those with disabilities. Accessible explanations are crucial as accessibility in technology promotes digital inclusion and allows everyone, regardless of their physical, sensory, or cognitive abilities, to use these technologies effectively. This paper presents a systematic literature review of the research on the accessibility of Explainable Artificial Intelligence (XAI), specifically considering persons with sight loss. Our methodology includes searching several academic databases with search terms to capture intersections between XAI and accessibility. The results of this survey highlight the lack of research on Accessible XAI (AXAI) and stress the importance of including the disability community in XAI development to promote digital inclusion and accessibility and remove barriers. Most XAI techniques rely on visual explanations, such as heatmaps or graphs, which are not accessible to persons who are blind or have low vision. Therefore, it is necessary to develop explanation methods through non-visual modalities, such as auditory and tactile feedback, visual modalities accessible to persons with low vision, and personalized solutions that meet the needs of individuals, including those with multiple disabilities. We further emphasize the importance of integrating universal design principles into AI development practices to ensure that AI technologies are usable by everyone., Comment: This work has been submitted to the IEEE for possible publication

Artificial Intelligence (AI) is increasingly employed to enhance assistive technologies, yet it can fail in various ways. We conducted a systematic literature review of research into AI-based assistive technology for persons with visual impairments. Our study shows that most proposed technologies with a testable prototype have not been evaluated in a human study with members of the sight-loss community. Furthermore, many studies did not consider or report failure cases or possible risks. These findings highlight the importance of inclusive system evaluations and the necessity of standardizing methods for presenting and analyzing failure cases and threats when developing AI-based assistive technologies.

A series of data samples was collected with the Belle~II detector at the SuperKEKB collider from March 2019 to June 2022. We determine the integrated luminosities of these data samples using three distinct methodologies involving Bhabha ($e^+e^- \to e^+e^-(n\gamma)$), digamma ($e^+e^- \to \gamma\gamma(n\gamma)$), and dimuon ($e^+e^- \to \mu^+ \mu^- (n\gamma)$) events. The total integrated luminosity obtained with Bhabha, digamma, and dimuon events is ({426.88} $\pm$ 0.03 $\pm$ {2.61})~fb$^{-1}$, ({429.28} $\pm$ 0.03 $\pm$ {2.62})~fb$^{-1}$, and ({423.99} $\pm$ 0.04 $\pm$ {3.83})~fb$^{-1}$, where the first uncertainties are statistical and the second are systematic. The resulting total integrated luminosity obtained from the combination of the three methods is ({427.87 $\pm$ 2.01})~fb$^{-1}$., Comment: 12 pages, 3 figures; accepted for publication in Chinese Physics C

The sustainable foraging problem is a dynamic environment testbed for exploring the forms of agent cognition in dealing with social dilemmas in a multi-agent setting. The agents need to resist the temptation of individual rewards through foraging and choose the collective long-term goal of sustainability. We investigate methods of online learning in Neuro-Evolution and Deep Recurrent Q-Networks to enable agents to attempt the problem one-shot as is often required by wicked social problems. We further explore if learning temporal dependencies with Long Short-Term Memory may be able to aid the agents in developing sustainable foraging strategies in the long term. It was found that the integration of Long Short-Term Memory assisted agents in developing sustainable strategies for a single agent, however failed to assist agents in managing the social dilemma that arises in the multi-agent scenario., Comment: 6 pages, 13 figures, accepted for publication by the Second International Workshop on Sustainability and Scalability of Self-Organisation (SaSSO 2024), DOI to be provided once published

This study modifies the Covariance Matrix Adaptation Evolution Strategy (CMA-ES) algorithm for multi-modal optimization problems. The enhancements focus on addressing the challenges of multiple global minima, improving the algorithm's ability to maintain diversity and explore complex fitness landscapes. We incorporate niching strategies and dynamic adaptation mechanisms to refine the algorithm's performance in identifying and optimizing multiple global optima. The algorithm generates a population of candidate solutions by sampling from a multivariate normal distribution centered around the current mean vector, with the spread determined by the step size and covariance matrix. Each solution's fitness is evaluated as a weighted sum of its contributions to all global minima, maintaining population diversity and preventing premature convergence. We implemented the algorithm on 8 tunable composite functions for the GECCO 2024 Competition on Benchmarking Niching Methods for Multi-Modal Optimization (MMO), adhering to the competition's benchmarking framework. The results are presenting in many ways such as Peak Ratio, F1 score on various dimensions. They demonstrate the algorithm's robustness and effectiveness in handling both global optimization and MMO- specific challenges, providing a comprehensive solution for complex multi-modal optimization problems., Comment: 15 pages, 1 figure, 16 tables. Submitted for GECCO 2024 competition on Benchmarking Niching Methods for Multimodal Optimization

We report a study of the hadronic transitions $\chi_{bJ}(2P)\to\omega\Upsilon(1S)$, with $\omega\to\pi^{+}\pi^{-}\pi^{0}$, using $28.2\times10^6~\Upsilon(3S)$ mesons recorded by the Belle detector. We present the first evidence for the near--threshold transition $\chi_{b0}(2P)\to\omega\Upsilon(1S)$, the analog of the charm sector decay $\chi_{c1}(3872)\to\omega J/\psi$, with a branching fraction of $B\big(\chi_{b0}(2P)\to\omega\Upsilon(1S)\big) = \big(0.55\pm0.19\pm0.07\big)\%$. We also obtain branching fractions of $B\big(\chi_{b1}(2P)\to\omega\Upsilon(1S)\big) = \big(2.39{}^{+0.20}_{-0.19}\pm0.24\big)\%$ and $B\big(\chi_{b2}(2P)\to\omega\Upsilon(1S)\big) = \big(0.47{}^{+0.13}_{-0.12}\pm0.06\big)\%$, confirming the measurement of the $\omega$ transitions of the $J=1,2~P$--wave states. The ratio for the $J=2$ to $J=1$ transitions is also measured and found to differ by 3.3 standard deviations from the expected value in the QCD multipole expansion., Comment: 6 pages, 2 figures

The ability to efficiently infer system parameters is essential in any signal-processing task that requires fast operation. Dealing with quantum systems, a serious challenge arises due to substantial growth of the underlying Hilbert space with the system size. As the statistics of the measurement data observed, i.e. the likelihood, can no longer be easily computed, common approaches such as maximum-likelihood estimators or particle filters become impractical. To address this issue, we propose the use of the Approximate Bayesian Computation (ABC) algorithm, which evades likelihood computation by sampling from a library of measurement data -- a priori prepared for a given quantum device. We apply ABC to interpret photodetection click-patterns arising when probing in real time a two-level atom and an optomechanical system. For the latter, we consider both linear and non-linear regimes, in order to show how to tailor the ABC algorithm by understanding the quantum measurement statistics. Our work demonstrates that fast parameter inference may be possible no matter the complexity of a quantum device and the measurement scheme involved., Comment: 20 pages, 10 figures, submitted version

Neutrinoless double beta decay is one of the most sensitive probes for new physics beyond the Standard Model of particle physics. One of the isotopes under investigation is $^{136}$Xe, which would double beta decay into $^{136}$Ba. Detecting the single $^{136}$Ba daughter provides a sort of ultimate tool in the discrimination against backgrounds. Previous work demonstrated the ability to perform single atom imaging of Ba atoms in a single-vacancy site of a solid xenon matrix. In this paper, the effort to identify signal from individual barium atoms is extended to Ba atoms in a hexa-vacancy site in the matrix and is achieved despite increased photobleaching in this site. Abrupt fluorescence turn-off of a single Ba atom is also observed. Significant recovery of fluorescence signal lost through photobleaching is demonstrated upon annealing of Ba deposits in the Xe ice. Following annealing, it is observed that Ba atoms in the hexa-vacancy site exhibit antibleaching while Ba atoms in the tetra-vacancy site exhibit bleaching. This may be evidence for a matrix site transfer upon laser excitation. Our findings offer a path of continued research toward tagging of Ba daughters in all significant sites in solid xenon., Comment: 9 pages, 8 figures

Curl lensing, also known as lensing field-rotation or shear B-modes, is a distinct post-Born observable caused by two lensing deflections at different redshifts (lens-lens coupling). For the Cosmic Microwave Background (CMB), the field-rotation is approximately four orders of magnitude smaller than the CMB lensing convergence. Direct detection is therefore challenging for near-future CMB experiments such as the Simons Observatory (SO) or CMB `Stage-4' (CMB-S4). Instead, the curl can be probed in cross-correlation between a direct reconstruction and a template formed using pairs of large-scale structure (LSS) tracers to emulate the lens-lens coupling. In this paper, we derive a new estimator for the optimal curl template specifically adapted for curved-sky applications, and test it against non-Gaussian complications using N-body cosmology simulations. We find non-foreground biases to the curl cross-spectrum are purely Gaussian at the sensitivity of SO. However, higher-order curl contractions induce non-Gaussian bias at the order of $1\sigma$ for CMB-S4 using quadratic estimators (QE). Maximum a-Posteriori (MAP) lensing estimators significantly reduce biases for both SO and CMB-S4, in agreement with our analytic predictions. We also show that extragalactic foregrounds in the CMB can bias curl measurements at order of the signal, and evaluate a variety of mitigation strategies to control these biases for SO-like experiments. Near-future observations will be able to measure post-Born lensing curl B-modes., Comment: 18 pages, 16 figures

We describe the planned near-term and potential longer-term upgrades of the Belle II detector at the SuperKEKB electron-positron collider operating at the KEK laboratory in Tsukuba, Japan. These upgrades will allow increasingly sensitive searches for possible new physics beyond the Standard Model in flavor, tau, electroweak and dark sector physics that are both complementary to and competitive with the LHC and other experiments., Comment: Editor: F. Forti 170 pages