Search

We study the dynamics of the solar basin -- the accumulated population of weakly-interacting particles on bound orbits in the Solar System. We focus on particles starting off on Sun-crossing orbits, corresponding to initial conditions of production inside the Sun, and investigate their evolution over the age of the Solar System. A combination of analytic methods, secular perturbation theory, and direct numerical integration of orbits sheds light on the long- and short-term evolution of a population of test particles orbiting the Sun and perturbed by the planets. Our main results are that the effective lifetime of a solar basin at Earth's location is $\tau_{\rm eff} = 1.20\pm 0.09 \,\mathrm{Gyr}$, and that there is annual (semi-annual) modulation of the basin density with known phase and amplitude at the fractional level of 6.5% (2.2%). These results have important implications for direct detection searches of solar basin particles, and the strong temporal modulation signature yields a robust discovery channel. Our simulations can also be interpreted in the context of gravitational capture of dark matter in the Solar System, with consequences for any dark-matter phenomenon that may occur below the local escape velocity.

We introduce STAResNet, a ResNet architecture in Spacetime Algebra (STA) to solve Maxwell's partial differential equations (PDEs). Recently, networks in Geometric Algebra (GA) have been demonstrated to be an asset for truly geometric machine learning. In \cite{brandstetter2022clifford}, GA networks have been employed for the first time to solve partial differential equations (PDEs), demonstrating an increased accuracy over real-valued networks. In this work we solve Maxwell's PDEs both in GA and STA employing the same ResNet architecture and dataset, to discuss the impact that the choice of the right algebra has on the accuracy of GA networks. Our study on STAResNet shows how the correct geometric embedding in Clifford Networks gives a mean square error (MSE), between ground truth and estimated fields, up to 2.6 times lower than than obtained with a standard Clifford ResNet with 6 times fewer trainable parameters. STAREsNet demonstrates consistently lower MSE and higher correlation regardless of scenario. The scenarios tested are: sampling period of the dataset; presence of obstacles with either seen or unseen configurations; the number of channels in the ResNet architecture; the number of rollout steps; whether the field is in 2D or 3D space. This demonstrates how choosing the right algebra in Clifford networks is a crucial factor for more compact, accurate, descriptive and better generalising pipelines.

The Poincar\'e gauge theory (PGT) of gravity provides a viable formulation of general relativity (Einstein-Cartan theory), and a popular model-building framework for modified gravity with torsion. Notoriously, however, the PGT terms which propagate vector torsion lead to strongly-coupled ghosts: the modern view is that only scalar torsion can propagate. To fix this, we revisit the concept of embedding explicit mass scales in scale-invariant theories, showing how the Klein-Gordon theory naturally leads to a slowly-rolling inflaton. We then show that the unique scale-invariant embedding of PGT leads to two new terms, one of which is the Maxwell term for vector torsion. We provide the full spectrum of quantum particles in the resulting theory. Our result means that every PGT is conformal and - after a two-decade hiatus - vector torsion is back on the menu., Comment: Fixed typos, added text around Eq. (18), added two refs

This report provides an overview of the challenge hosted at the OpenSUN3D Workshop on Open-Vocabulary 3D Scene Understanding held in conjunction with ICCV 2023. The goal of this workshop series is to provide a platform for exploration and discussion of open-vocabulary 3D scene understanding tasks, including but not limited to segmentation, detection and mapping. We provide an overview of the challenge hosted at the workshop, present the challenge dataset, the evaluation methodology, and brief descriptions of the winning methods. For additional details, please see https://opensun3d.github.io/index_iccv23.html., Comment: Our OpenSUN3D workshop website for ICCV 2023: https://opensun3d.github.io/index_iccv23.html

We provide a complementary nested sampling analysis for the Atacama Cosmology Telescope lensing data release 6. This allows the quantification of global consistency statistics between ACT lensing and alternative datasets. In the context of flat $\Lambda$CDM, we find no inconsistency between ACT, Baryonic Acoustic Oscillations, Planck anisotropies, weak lensing datasets, or NPIPE lensing. As part of our analysis, we also investigate the effect of the prior widths used in the ACT analysis and find that the headline results are quantitatively but not qualitatively affected by the chosen priors. We use both Bayes factors and the suspiciousness statistic to quantify the possibility of tension, and find suspiciousness unsuitable in the case of strong agreement between ACT DR6 and NPIPE. Nested sampling provides a competitive alternative to Metropolis Hastings and we recommend it be used alongside existing analyses. We release the chains and plotting source for the analysis using anesthetic., Comment: 8 pages, 2 figures, 3 tables. Nested sampling chains and analysis code available at https://zenodo.org/records/10252886. Comments welcome!

A variational principle for gauge theories of gravity is presented, which maintains manifest covariance under the symmetries to which the action is invariant, throughout the calculation of the equations of motion and conservation laws. This is performed by deriving explicit manifestly covariant expressions for the Euler--Lagrange variational derivatives and Noether's theorems for a generic action of the form typically assumed in gauge theories of gravity. The approach is illustrated by application to two scale-invariant gravitational gauge theories, namely Weyl gauge theory (WGT) and the recently proposed `extended' Weyl gauge theory (eWGT), where the latter may be considered as a novel gauging of the conformal group, but the method can be straightforwardly applied to other theories with smaller or larger symmetry groups. The approach also enables one easily to establish the relationship between manifestly covariant forms of variational derivatives obtained when one or more of the gauge field strengths is set to zero either before or after the variation is performed. This is illustrated explicitly for both WGT and eWGT in the case where the translational gauge field strength (or torsion) is set to zero before and after performing the variation, respectively., Comment: 27 pages, no figures, submitted to PRD

In this work, we introduce OpenIns3D, a new 3D-input-only framework for 3D open-vocabulary scene understanding. The OpenIns3D framework employs a "Mask-Snap-Lookup" scheme. The "Mask" module learns class-agnostic mask proposals in 3D point clouds, the "Snap" module generates synthetic scene-level images at multiple scales and leverages 2D vision-language models to extract interesting objects, and the "Lookup" module searches through the outcomes of "Snap" to assign category names to the proposed masks. This approach, yet simple, achieves state-of-the-art performance across a wide range of 3D open-vocabulary tasks, including recognition, object detection, and instance segmentation, on both indoor and outdoor datasets. Moreover, OpenIns3D facilitates effortless switching between different 2D detectors without requiring retraining. When integrated with powerful 2D open-world models, it achieves excellent results in scene understanding tasks. Furthermore, when combined with LLM-powered 2D models, OpenIns3D exhibits an impressive capability to comprehend and process highly complex text queries that demand intricate reasoning and real-world knowledge. Project page: https://zheninghuang.github.io/OpenIns3D/, Comment: ECCV 2024. Project page: https://zheninghuang.github.io/OpenIns3D/

We use the new QUIJOTE-MFI wide survey (11, 13, 17 and 19 GHz) to produce spectral energy distributions (SEDs), on an angular scale of 1 deg, of the supernova remnants (SNRs) CTB 80, Cygnus Loop, HB 21, CTA 1, Tycho and HB 9. We provide new measurements of the polarized synchrotron radiation in the microwave range. For each SNR, the intensity and polarization SEDs are obtained and modelled by combining QUIJOTE-MFI maps with ancillary data. In intensity, we confirm the curved power law spectra of CTB 80 and HB 21 with a break frequency $\nu_{\rm b}$ at 2.0$^{+1.2}_{-0.5}$ GHz and 5.0$^{+1.2}_{-1.0}$ GHz respectively; and spectral indices respectively below and above the spectral break of $-0.34\pm0.04$ and $-0.86\pm0.5$ for CTB 80, and $-0.24\pm0.07$ and $-0.60\pm0.05$ for HB 21. In addition, we provide upper limits on the Anomalous Microwave Emission (AME), suggesting that the AME contribution is negligible towards these remnants. From a simultaneous intensity and polarization fit, we recover synchrotron spectral indices as flat as $-0.24$, and the whole sample has a mean and scatter of $-0.44\pm0.12$. The polarization fractions have a mean and scatter of $6.1\pm1.9$\%. When combining our results with the measurements from other QUIJOTE studies of SNRs, we find that radio spectral indices are flatter for mature SNRs, and particularly flatter for CTB 80 ($-0.24^{+0.07}_{-0.06}$) and HB 21 ($-0.34^{+0.04}_{-0.03}$). In addition, the evolution of the spectral indices against the SNRs age is modelled with a power-law function, providing an exponent $-0.07\pm0.03$ and amplitude $-0.49\pm0.02$ (normalised at 10 kyr), which are conservative with respect to previous studies of our Galaxy and the Large Magellanic Cloud., Comment: 33 pages, 15 figure, 15 tables. Submitted to MNRAS. QUIJOTE data maps available at https://research.iac.es/proyecto/quijote

The Andromeda Galaxy (M31) is the Local Group galaxy that is most similar to the Milky Way (MW). The similarities between the two galaxies make M31 useful for studying integrated properties common to spiral galaxies. We use the data from the recent QUIJOTE-MFI Wide Survey, together with new raster observations focused on M31, to study its integrated emission. The addition of raster data improves the sensitivity of QUIJOTE-MFI maps by almost a factor 3. Our main interest is to confirm if anomalous microwave emission (AME) is present in M31, as previous studies have suggested. To do so, we built the integrated spectral energy distribution of M31 between 0.408 and 3000 GHz. We then performed a component separation analysis taking into account synchrotron, free-free, AME and thermal dust components. AME in M31 is modelled as a log-normal distribution with maximum amplitude, $A_{\rm AME}$, equal to $1.03\pm0.32$ Jy. It peaks at $\nu_{\rm AME}=17.2\pm3.2$ GHz with a width of $W_{\rm AME}=0.58\pm0.16$. Both the Akaike and Bayesian Information Criteria find the model without AME to be less than 1 % as probable as the one taking AME into consideration. We find that the AME emissivity per 100 $\mu$m intensity in M31 is $\epsilon_{\rm AME}^{\rm 28.4\,GHz}=9.6\pm3.1$ $\mu$K/(MJy/sr), similar to that computed for the MW. We also provide the first upper limits for the AME polarization fraction in an extragalactic object. M31 remains the only galaxy where an AME measurement has been made of its integrated spectrum., Comment: 17 pages, 9 figures. Accepted for publication in MNRAS. QUIJOTE data maps available at https://research.iac.es/proyecto/quijote

Anomalous Microwave Emission (AME) is an important emission component between 10 and 60 GHz that is not yet fully understood. It seems to be ubiquituous in our Galaxy and is observed at a broad range of angular scales. Here we use the new QUIJOTE-MFI wide survey data at 11, 13, 17 and 19 GHz to constrain the AME in the Galactic plane ($|b|<10^\circ$) on degree scales. We built the spectral energy distribution between 0.408 and 3000 GHz for each of the 5309 0.9$^\circ$ pixels in the Galactic plane, and fitted a parametric model by considering five emission components: synchrotron, free-free, AME, thermal dust and CMB anisotropies. We show that not including QUIJOTE-MFI data points leads to the underestimation (up to 50 %) of the AME signal in favour of free-free emission. The parameters describing these components are then intercompared, looking for relations that help to understand AME physical processes. We find median values for the AME width, $W_{\rm AME}$, and for its peak frequency, $\nu_{\rm AME}$, respectively of $0.560^{+0.059}_{-0.050}$ and $20.7^{+2.0}_{-1.9}$ GHz, slightly in tension with current theoretical models. We find spatial variations throughout the Galactic plane for $\nu_{\rm AME}$, but only with reduced statistical significance. We report correlations of AME parameters with certain ISM properties, such as that between the AME emissivity (which shows variations with the Galactic longitude) and the interstellar radiation field, and that between the AME peak frequency and dust temperature. Finally, we discuss the implications of our results on the possible molecules responsible for AME., Comment: 32 pages, 31 figures. Accepted for publication in MNRAS. Derived data products available at https://research.iac.es/proyecto/quijote

The short answer is $\textit{probably no}$. Specifically, this paper considers a recent body of work which suggests that general relativity requires neither the support of dark matter halos, nor unconventional baryonic profiles, nor any infrared modification, to be consistent after all with the anomalously rapid orbits observed in many galactic discs. In particular, the gravitoelectric flux is alleged to collapse nonlinearly into regions of enhanced force, in an analogue of the colour-confining chromoelectric flux tube model which has yet to be captured by conventional post-Newtonian methods. However, we show that the scalar gravity model underpinning this proposal is wholly inconsistent with the nonlinear Einstein equations, which themselves appear to prohibit the linear confinement-type potentials which could indicate a disordered gravitational phase. Our findings challenge the fidelity of the previous Euclidean lattice analyses. We confirm by direct calculation using a number of perturbation schemes and gauges that the next-to-leading order gravitoelectric correction to the rotation curve of a reasonable baryonic profile would be imperceptible. The `gravitoelectric flux collapse' programme was also supported by using intragalactic lensing near a specific galactic baryon profile as a field strength heuristic. We recalculate this lensing effect, and conclude that it has been overstated by three orders of magnitude. As a by-product, our analysis suggests fresh approaches to (i) the fluid ball conjecture and (ii) gravitational energy localisation, both to be pursued in future work. In summary, whilst it may be interesting to consider the possibility of confinement-type effects in gravity, we may at least conclude here that confinement-type effects $\textit{cannot play any significant part}$ in explaining flat or rising galactic rotation curves without dark matter halos., Comment: 33 pages, 19 figures

We investigate recent claims that gravitomagnetic effects in linearised general relativity can explain flat and rising rotation curves, such as those observed in galaxies, without the need for dark matter. If one models a galaxy as an axisymmetric, stationary, rotating, non-relativistic and pressureless 'dust' of stars in the gravitoelectromagnetic (GEM) formalism, we show that GEM effects on the circular velocity $v$ of a star are $O(10^{-6})$ smaller than the standard Newtonian (gravitoelectric) effects. Moreover, we find that gravitomagnetic effects are $O(10^{-6})$ too small to provide the vertical support necessary to maintain the dynamical equilibrium assumed. These issues are obscured if one constructs a single equation for $v$, as considered previously. We nevertheless solve this equation for a galaxy having a Miyamoto--Nagai density profile. We show that for the values of the mass, $M$, and semi-major and semi-minor axes, $a$ and $b$, typical for a dwarf galaxy, the rotation curve depends only very weakly on $M$. Moreover, for aspect ratios $a/b > 2$, the rotation curves are concave over their entire range, which does not match observations in any galaxy. Most importantly, we show that for the poloidal gravitomagnetic flux $\psi$ to provide the necessary vertical support, it must become singular at the origin. This originates from the unwitting, but forbidden, inclusion of free-space solutions of the Poisson-like equation that determines $\psi$, hence ruling out the methodology as a means of explaining flat galaxy rotation curves. We further show that recent deliberate attempts to leverage such free-space solutions against the rotation curve problem yield no deterministic modification outside the thin disk approximation, and that, in any case, the homogeneous contributions to $\psi$ are ruled out by the boundary value problem posed by any physical axisymmetric galaxy., Comment: Minor textual changes

We test the hypothesis that language models trained with reinforcement learning from human feedback (RLHF) have the capability to "morally self-correct" -- to avoid producing harmful outputs -- if instructed to do so. We find strong evidence in support of this hypothesis across three different experiments, each of which reveal different facets of moral self-correction. We find that the capability for moral self-correction emerges at 22B model parameters, and typically improves with increasing model size and RLHF training. We believe that at this level of scale, language models obtain two capabilities that they can use for moral self-correction: (1) they can follow instructions and (2) they can learn complex normative concepts of harm like stereotyping, bias, and discrimination. As such, they can follow instructions to avoid certain kinds of morally harmful outputs. We believe our results are cause for cautious optimism regarding the ability to train language models to abide by ethical principles.

We introduce CGA-PoseNet, which uses the 1D-Up approach to Conformal Geometric Algebra (CGA) to represent rotations and translations with a single mathematical object, the motor, for camera pose regression. We do so starting from PoseNet, which successfully predicts camera poses from small datasets of RGB frames. State-of-the-art methods, however, require expensive tuning to balance the orientational and translational components of the camera pose.This is usually done through complex, ad-hoc loss function to be minimized, and in some cases also requires 3D points as well as images. Our approach has the advantage of unifying the camera position and orientation through the motor. Consequently, the network searches for a single object which lives in a well-behaved 4D space with a Euclidean signature. This means that we can address the case of image-only datasets and work efficiently with a simple loss function, namely the mean squared error (MSE) between the predicted and ground truth motors. We show that it is possible to achieve high accuracy camera pose regression with a significantly simpler problem formulation. This 1D-Up approach to CGA can be employed to overcome the dichotomy between translational and orientational components in camera pose regression in a compact and elegant way.

We present the catalogue of Q-U-I JOint TEnerife (QUIJOTE) Wide Survey radio sources extracted from the maps of the Multi-Frequency Instrument compiled between 2012 and 2018. The catalogue contains 786 sources observed in intensity and polarization, and is divided into two separate sub-catalogues: one containing 47 bright sources previously studied by the \emph{Planck} collaboration and an extended catalogue of 739 sources either selected from the \emph{Planck} Second Catalogue of Compact Sources or found through a blind search carried out with a Mexican Hat 2 wavelet. A significant fraction of the sources in our catalogue (38.7 per cent) are within the $|b| \leq 20^\circ$ region of the Galactic plane. We determine statistical properties for those sources that are likely to be extragalactic. We find that these statistical properties are compatible with currently available models, with a $\sim$1.8 Jy completeness limit at 11 GHz. We provide the polarimetric properties of (38, 33, 31, 23) sources with P detected above the $99.99\%$ significance level at (11, 13, 17, 19) GHz, respectively. Median polarization fractions are in the $2.8$-$4.7$\% range in the 11-19 GHz frequency interval. We do not distinguish between Galactic and extragalactic sources here. The results presented here are consistent with those reported in the literature for flat- and steep-spectrum radio sources., Comment: 21 pages, 13 figures, 7 tables

We derive linearly polarized astrophysical component maps in the Northern Sky from the QUIJOTE-MFI data at 11 and 13 GHz in combination with the WMAP K and Ka bands (23 and 33 GHz) and all Planck polarized channels (30-353 GHz), using the parametric component separation method B-SeCRET. The addition of QUIJOTE-MFI data significantly improves the parameter estimation of the low-frequency foregrounds, especially the estimation of the synchrotron spectral index, $\beta_s$. We present the first detailed $\beta_s$ map of the Northern Celestial Hemisphere at a smoothing scale of $2^{\circ}$. We find statistically significant spatial variability across the sky. We obtain an average value of $-3.08$ and a dispersion of $0.13$, considering only pixels with reliable goodness-of-fit. The power law model of the synchrotron emission provides a good fit to the data outside the Galactic plane but fails to track the complexity within this region. Moreover, when we assume a synchrotron model with uniform curvature, $c_s$, we find a value of $c_s = -0.0797 \pm 0.0012$. However, there is insufficient statistical significance to determine which model is favoured, either the power law or the power law with uniform curvature. Furthermore, we estimate the thermal dust spectral parameters in polarization. Our CMB, synchrotron, and thermal dust maps are highly correlated with the corresponding products of the PR4 Planck release, although some large-scale differences are observed in the synchrotron emission. Finally, we find that the $\beta_s$ estimation in the high signal-to-noise synchrotron emission areas is prior-independent while, outside these regions, the prior governs the $\beta_s$ estimation., Comment: 25 pages, 29 figures, Accepted for publication in MNRAS

The QUIJOTE-MFI Northern Hemisphere Wide-Survey has provided maps of the sky above declinations $-30^\circ$ at 11, 13, 17 and 19$\,$GHz. These data are combined with ancillary data to produce Spectral Energy Distributions in intensity in the frequency range 0.4--3\,000$\,$GHz on a sample of 52 candidate compact sources harbouring anomalous microwave emission (AME). We apply a component separation analysis at 1$^\circ$ scale on the full sample from which we identify 44 sources with high AME significance. We explore correlations between different fitted parameters on this last sample. QUIJOTE-MFI data contribute to notably improve the characterisation of the AME spectrum, and its separation from the other components. In particular, ignoring the 10--20\,GHz data produces on average an underestimation of the AME amplitude, and an overestimation of the free-free component. We find an average AME peak frequency of 23.6 $\pm$ 3.6$\,$GHz, about 4$\,$GHz lower than the value reported in previous studies. The strongest correlation is found between the peak flux density of the thermal dust and of the AME component. A mild correlation is found between the AME emissivity ($A_{\rm AME}/\tau_{250}$) and the interstellar radiation field. On the other hand no correlation is found between the AME emissivity and the free-free radiation Emission Measure. Our statistical results suggest that the interstellar radiation field could still be the main driver of the intensity of the AME as regards spinning dust excitation mechanisms. On the other hand, it is not clear whether spinning dust would be most likely associated with cold phases of the interstellar medium rather than with hot phases dominated by free-free radiation., Comment: 36 pages. Accepted for publication in MNRAS. Maps and derived data products available at https://research.iac.es/proyecto/quijote/

The Haze is an excess of microwave intensity emission surrounding the Galactic centre. It is spatially correlated with the $\gamma$-ray Fermi bubbles, and with the S-PASS radio polarization plumes, suggesting a possible common provenance. The models proposed to explain the origin of the Haze, including energetic events at the Galactic centre and dark matter decay in the Galactic halo, do not yet provide a clear physical interpretation. In this paper we present a re-analysis of the Haze including new observations from the Multi-Frequency Instrument (MFI) of the Q-U-I JOint TEnerife (QUIJOTE) experiment, at 11 and 13 GHz. We analyze the Haze in intensity and polarization, characterizing its spectrum. We detect an excess of diffuse intensity signal ascribed to the Haze. The spectrum at frequencies 11$\,\leq\nu\leq\,$70 GHz is a power-law with spectral index $\beta^{\rm H}=-2.79\pm0.08$, which is flatter than the Galactic synchrotron in the same region ($\beta^{\rm S}=-2.98\pm0.04$), but steeper than that obtained from previous works ($\beta^{\rm H}\sim-2.5$ at 23$\,\leq\,\nu\leq\,$70 GHz). We also observe an excess of polarized signal in the QUIJOTE-MFI maps in the Haze area. This is a first hint detection of polarized Haze, or a consequence of curvature of the synchrotron spectrum in that area. Finally, we show that the spectrum of polarized structures associated with Galactic centre activity is steep at low frequencies ($\beta \sim -3.2$ at 2.3 $\leq\nu\leq$ 23 GHz), and becomes flatter above 11 GHz., Comment: 31 pages. Accepted for publication in MNRAS

We present new intensity and polarisation maps obtained with the QUIJOTE experiment towards the Galactic regions W49, W51 and IC443, covering the frequency range from 10 to 20 GHz at $\sim$ 1 deg angular resolution, with a sensitivity in the range 35-79 ${\mu}$K/beam for total intensity and 13-23 ${\mu}$K/beam for polarisation. For each region, we combine QUIJOTE maps with ancillary data at frequencies ranging from 0.4 to 3000 GHz, reconstruct the spectral energy distribution and model it with a combination of known foregrounds. We detect anomalous microwave emission (AME) in total intensity towards W49 at 4.7${\sigma}$ and W51 at 4.0${\sigma}$ with peak frequencies ${\nu}_{AME}$ = (20.0 $\pm$ 1.4) GHz and ${\nu}_{AME}$ = (17.7 $\pm$ 3.6) GHz respectively; this is the first detection of AME towards W51. The contamination from ultra-compact HII regions to the residual AME flux density is estimated at 10% in W49 and 5% in W51, and does not rule out the AME detection. The polarised SEDs reveal a synchrotron contribution with spectral indices ${\alpha}_s$ = -0.67 $\pm$ 0.10 in W49 and ${\alpha}_s$ = -0.51 $\pm$ 0.07 in W51, ascribed to the diffuse Galactic emission and to the local supernova remnant respectively. Towards IC443 in total intensity we measure a broken power-law synchrotron spectrum with cut-off frequency ${\nu}_{0,s}$ = (114 $\pm$ 73) GHz, in agreement with previous studies; our analysis, however, rules out any AME contribution which had been previously claimed towards IC443. No evidence of polarised AME emission is detected in this study., Comment: 29 pages, 11 figures. Accepted for publication in MNRAS

We present QUIJOTE intensity and polarization maps in four frequency bands centred around 11, 13, 17 and 19GHz, and covering approximately 29000 deg$^2$, including most of the Northern sky region. These maps result from 9000 h of observations taken between May 2013 and June 2018 with the first QUIJOTE instrument (MFI), and have angular resolutions of around $1^\circ$, and sensitivities in polarization within the range 35-40 $\mu$K per 1-degree beam, being a factor $\sim 2$-$4$ worse in intensity. We discuss the data processing pipeline employed, and the basic characteristics of the maps in terms of real space statistics and angular power spectra. A number of validation tests have been applied to characterise the accuracy of the calibration and the residual level of systematic effects, finding a conservative overall calibration uncertainty of 5%. We also discuss flux densities for four bright celestial sources (Tau A, Cas A, Cyg A and 3C274) which are often used as calibrators at microwave frequencies. The polarization signal in our maps is dominated by synchrotron emission. The distribution of spectral index values between the 11GHz and WMAP 23GHz map peaks at $\beta=-3.09$ with a standard deviation of 0.14. The measured BB/EE ratio at scales of $\ell=80$ is $0.26\pm 0.07$ for a Galactic cut $|b|>10^\circ$. We find a positive TE correlation for 11GHz at large angular scales ($\ell \lesssim 50$), while the EB and TB signals are consistent with zero in the multipole range $30 \lesssim \ell \lesssim 150$. The maps discussed in this paper are publicly available., Comment: 60 pages. Accepted for publication in MNRAS. Maps and derived data products available at https://research.iac.es/proyecto/quijote/

As AI systems become more capable, we would like to enlist their help to supervise other AIs. We experiment with methods for training a harmless AI assistant through self-improvement, without any human labels identifying harmful outputs. The only human oversight is provided through a list of rules or principles, and so we refer to the method as 'Constitutional AI'. The process involves both a supervised learning and a reinforcement learning phase. In the supervised phase we sample from an initial model, then generate self-critiques and revisions, and then finetune the original model on revised responses. In the RL phase, we sample from the finetuned model, use a model to evaluate which of the two samples is better, and then train a preference model from this dataset of AI preferences. We then train with RL using the preference model as the reward signal, i.e. we use 'RL from AI Feedback' (RLAIF). As a result we are able to train a harmless but non-evasive AI assistant that engages with harmful queries by explaining its objections to them. Both the SL and RL methods can leverage chain-of-thought style reasoning to improve the human-judged performance and transparency of AI decision making. These methods make it possible to control AI behavior more precisely and with far fewer human labels.

We discuss the challenges of motivating, constructing, and quantizing a canonically normalized inflationary perturbation in spatially curved universes. We show that this has historically proved challenging due to the interaction of nonadiabaticity with spatial curvature. We construct a novel curvature perturbation that is canonically normalized in the sense of its equation of motion and is unique up to a single scalar parameter. With this construction it becomes possible to set initial conditions invariant under canonical transformations, overcoming known ambiguities in the literature. This corrected quantization has potentially observational consequences via modifications to the primordial power spectrum at large angular scales, as well as theoretical implications for quantization procedures in curved cosmologies filled with a scalar field., Comment: 15 pages, 3 figures; v3: As published in PRD

Solar forecasting from ground-based sky images has shown great promise in reducing the uncertainty in solar power generation. With more and more sky image datasets open sourced in recent years, the development of accurate and reliable deep learning-based solar forecasting methods has seen a huge growth in potential. In this study, we explore three different training strategies for solar forecasting models by leveraging three heterogeneous datasets collected globally with different climate patterns. Specifically, we compare the performance of local models trained individually based on single datasets and global models trained jointly based on the fusion of multiple datasets, and further examine the knowledge transfer from pre-trained solar forecasting models to a new dataset of interest. The results suggest that the local models work well when deployed locally, but significant errors are observed when applied offsite. The global model can adapt well to individual locations at the cost of a potential increase in training efforts. Pre-training models on a large and diversified source dataset and transferring to a target dataset generally achieves superior performance over the other two strategies. With 80% less training data, it can achieve comparable performance as the local baseline trained using the entire dataset.

Neural networks often pack many unrelated concepts into a single neuron - a puzzling phenomenon known as 'polysemanticity' which makes interpretability much more challenging. This paper provides a toy model where polysemanticity can be fully understood, arising as a result of models storing additional sparse features in "superposition." We demonstrate the existence of a phase change, a surprising connection to the geometry of uniform polytopes, and evidence of a link to adversarial examples. We also discuss potential implications for mechanistic interpretability., Comment: Also available at https://transformer-circuits.pub/2022/toy_model/index.html

For conformally invariant gravity theories defined on Riemannian spacetime and having the Schwarzschild--de-Sitter (SdS) metric as a solution in the Einstein gauge, we consider whether one may conformally rescale this solution to obtain flat rotation curves, such as those observed in galaxies, without the need for dark matter. Contrary to recent claims in the literature, we show that if one works in terms of quantities that can be physically measured, then in any conformal frame the trajectories followed by `ordinary' matter particles are merely the timelike geodesics of the SdS metric, as one might expect. This resolves the apparent frame dependence of physical predictions and unambiguously yields rotation curves with no flat region. We also show that attempts to model rising rotation curves by fitting the coefficient of the quadratic term in the SdS metric individually for each galaxy are precluded, since this coefficient is most naturally interpreted as proportional to a global cosmological constant. We further extend our analysis beyond static, spherically-symmetric systems to show that the invariance of particle dynamics to the choice of conformal frame holds for arbitrary metrics, again as expected. Moreover, we show that this conclusion remains valid for conformally invariant gravity theories defined on more general Weyl--Cartan spacetimes, which include Weyl, Riemann--Cartan and Riemannian spacetimes as special cases., Comment: 5 pages, no figures, accepted as a Letter by European Physical Journal C

Graph Self-Supervised Learning (GSSL) provides a robust pathway for acquiring embeddings without expert labelling, a capability that carries profound implications for molecular graphs due to the staggering number of potential molecules and the high cost of obtaining labels. However, GSSL methods are designed not for optimisation within a specific domain but rather for transferability across a variety of downstream tasks. This broad applicability complicates their evaluation. Addressing this challenge, we present "Molecular Graph Representation Evaluation" (MOLGRAPHEVAL), generating detailed profiles of molecular graph embeddings with interpretable and diversified attributes. MOLGRAPHEVAL offers a suite of probing tasks grouped into three categories: (i) generic graph, (ii) molecular substructure, and (iii) embedding space properties. By leveraging MOLGRAPHEVAL to benchmark existing GSSL methods against both current downstream datasets and our suite of tasks, we uncover significant inconsistencies between inferences drawn solely from existing datasets and those derived from more nuanced probing. These findings suggest that current evaluation methodologies fail to capture the entirety of the landscape., Comment: Camera ready, NeurIPS Benchmark 2023

Integration of intermittent renewable energy sources into electric grids in large proportions is challenging. A well-established approach aimed at addressing this difficulty involves the anticipation of the upcoming energy supply variability to adapt the response of the grid. In solar energy, short-term changes in electricity production caused by occluding clouds can be predicted at different time scales from all-sky cameras (up to 30-min ahead) and satellite observations (up to 6h ahead). In this study, we integrate these two complementary points of view on the cloud cover in a single machine learning framework to improve intra-hour (up to 60-min ahead) irradiance forecasting. Both deterministic and probabilistic predictions are evaluated in different weather conditions (clear-sky, cloudy, overcast) and with different input configurations (sky images, satellite observations and/or past irradiance values). Our results show that the hybrid model benefits predictions in clear-sky conditions and improves longer-term forecasting. This study lays the groundwork for future novel approaches of combining sky images and satellite observations in a single learning framework to advance solar nowcasting., Comment: Submitted to Renewable Energy

We review Skilling's nested sampling (NS) algorithm for Bayesian inference and more broadly multi-dimensional integration. After recapitulating the principles of NS, we survey developments in implementing efficient NS algorithms in practice in high-dimensions, including methods for sampling from the so-called constrained prior. We outline the ways in which NS may be applied and describe the application of NS in three scientific fields in which the algorithm has proved to be useful: cosmology, gravitational-wave astronomy, and materials science. We close by making recommendations for best practice when using NS and by summarizing potential limitations and optimizations of NS., Comment: 20 pages + supplementary information, 5 figures. preprint version; published version at https://www.nature.com/articles/s43586-022-00121-x

The true posterior distribution of a Bayesian neural network is massively multimodal. Whilst most of these modes are functionally equivalent, we demonstrate that there remains a level of real multimodality that manifests in even the simplest neural network setups. It is only by fully marginalising over all posterior modes, using appropriate Bayesian sampling tools, that we can capture the split personalities of the network. The ability of a network trained in this manner to reason between multiple candidate solutions dramatically improves the generalisability of the model, a feature we contend is not consistently captured by alternative approaches to the training of Bayesian neural networks. We provide a concise minimal example of this, which can provide lessons and a future path forward for correctly utilising the explainability and interpretability of Bayesian neural networks., Comment: 10 pages, 5 figures

We investigate the effects of non-zero spatial curvature on cosmic inflation in the light of cosmic microwave background (CMB) anisotropy measurements from the Planck 2018 legacy release and from the 2015 observing season of BICEP2 and the Keck Array. Even a small percentage of non-zero curvature today would significantly limit the total number of e-folds of the scale factor during inflation, rendering just-enough inflation scenarios with a kinetically dominated or fast-roll stage prior to slow-roll inflation more likely. Finite inflation leads to oscillations and a cutoff towards large scales in the primordial power spectrum and curvature pushes them into the CMB observable window. Using nested sampling, we carry out Bayesian parameter estimations and model comparisons taking into account constraints from reheating and horizon considerations. We confirm the preference of CMB data for closed universes with Bayesian odds of over $100:1$ and with a posterior on the curvature density parameter of $\Omega_{K,0}=-0.051\pm0.017$ for a curvature extension of LCDM and $\Omega_{K,0}=-0.031\pm0.014$ for Starobinsky inflation. Model comparisons of various inflation models give similar results as for flat universes with the Starobinsky model outperforming most other models., Comment: 46 pages, 32 figures, 3 tables, submitted to PRD, from chapter 5 of L. T. Hergt's doctoral thesis at https://www.repository.cam.ac.uk/handle/1810/323129, data available at https://zenodo.org/record/6547872

Different ways of representing the group $SU(3)$ within a Geometric Algebra approach are explored. As part of this we consider characteristic multivectors for $SU(3)$, and how these are linked with decomposition of generators into commuting bivectors. The setting for this work is within a 6d Euclidean Clifford Algebra. We then go on to consider whether the fundamental forces of particle physics might arise from symmetry considerations in just the 4d geometric algebra of spacetime -- the STA. As part of this, a representation of $SU(3)$ is found wholly within the STA, involving preservation of a bivector norm. We also show how Octonions can be fully represented within the Spacetime Algebra, which we believe will be useful in making them understandable and accessible to a new community in Physics and Engineering. The two strands of the paper are drawn together in showing how preserving the octonion norm is the same as preserving the timelike part of the Dirac current of a particle. This suggests a new model for the symmetries preserved in particle physics. Following on from work by G\"unaydin and G\"ursey on the link between quarks, and octonions, and by Furey on chains of octonionic multiplications, we show how both of these fit well within our scheme, and give some wholly STA versions of the operations involved, which in the cases considered have easily understandable equivalents in terms of 4d geometry. Links with larger groups containing $SU(3)$, such as $G_2$ and $SU(8)$, are also considered., Comment: First parts of paper based on a talk given at the meeting AGACSE 2021, Brno, Czech Republic, September 2021

It is well-known that stars have the potential to be excellent dark matter detectors. Infalling dark matter that scatters within stars could lead to a range of observational signatures, including stellar heating, black hole formation, and modified heat transport. To make robust predictions for such phenomena, it is necessary to calculate the scattering rate for dark matter inside the star. As we show in this paper, for small enough momentum transfers, this requires taking into account collective effects within the dense stellar medium. These effects have been neglected in many previous treatments; we demonstrate how to incorporate them systematically, and show that they can parametrically enhance or suppress dark matter scattering rates depending on how dark matter couples to the Standard Model. We show that, as a result, collective effects can significantly modify the potential discovery or exclusion reach for observations of compact objects such as white dwarfs and neutron stars. While the effects are more pronounced for dark matter coupling through a light mediator, we show that even for dark matter coupling via a heavy mediator, scattering rates can differ by orders of magnitude from their naive values for dark matter masses <~ 100 MeV. We also illustrate how collective effects can be important for dark matter scattering in more dilute media, such as the Solar core. Our results demonstrate the need to systematically incorporate collective effects in a wide range of astroparticle contexts; to facilitate this, we provide expressions for in-medium self-energies for a variety of different media, which are applicable to many other processes of interest (such as particle production)., Comment: 66 pages, 14 figures; v3, matches version published in JCAP

We reconsider the status of the so-called second clock effect in Weyl gauge theories of gravity, which are invariant both under local Poincar\'e transformations and local changes of scale. In particular, we revisit and extend our previous demonstration that the second clock effect does not occur in such theories, in order both to clarify our argument and confirm our original findings in response to recent counterclaims in the literature., Comment: 5 pages, no figures, accepted for publication as a Letter in PRD. Minor edits to match published version. arXiv admin note: text overlap with arXiv:2009.06407

Translational invariance induced by pooling operations is an inherent property of convolutional neural networks, which facilitates numerous computer vision tasks such as classification. Yet to leverage rotational invariant tasks, convolutional architectures require specific rotational invariant layers or extensive data augmentation to learn from diverse rotated versions of a given spatial configuration. Unwrapping the image into its polar coordinates provides a more explicit representation to train a convolutional architecture as the rotational invariance becomes translational, hence the visually distinct but otherwise equivalent rotated versions of a given scene can be learnt from a single image. We show with two common vision-based solar irradiance forecasting challenges (i.e. using ground-taken sky images or satellite images), that this preprocessing step significantly improves prediction results by standardising the scene representation, while decreasing training time by a factor of 4 compared to augmenting data with rotations. In addition, this transformation magnifies the area surrounding the centre of the rotation, leading to more accurate short-term irradiance predictions., Comment: CVPR 2022 - OmniCV workshop (oral)

Artificial intelligence (AI) provides a promising substitution for streamlining COVID-19 diagnoses. However, concerns surrounding security and trustworthiness impede the collection of large-scale representative medical data, posing a considerable challenge for training a well-generalised model in clinical practices. To address this, we launch the Unified CT-COVID AI Diagnostic Initiative (UCADI), where the AI model can be distributedly trained and independently executed at each host institution under a federated learning framework (FL) without data sharing. Here we show that our FL model outperformed all the local models by a large yield (test sensitivity /specificity in China: 0.973/0.951, in the UK: 0.730/0.942), achieving comparable performance with a panel of professional radiologists. We further evaluated the model on the hold-out (collected from another two hospitals leaving out the FL) and heterogeneous (acquired with contrast materials) data, provided visual explanations for decisions made by the model, and analysed the trade-offs between the model performance and the communication costs in the federated training process. Our study is based on 9,573 chest computed tomography scans (CTs) from 3,336 patients collected from 23 hospitals located in China and the UK. Collectively, our work advanced the prospects of utilising federated learning for privacy-preserving AI in digital health., Comment: Nature Machine Intelligence

Molecular graph representation learning is a fundamental problem in modern drug and material discovery. Molecular graphs are typically modeled by their 2D topological structures, but it has been recently discovered that 3D geometric information plays a more vital role in predicting molecular functionalities. However, the lack of 3D information in real-world scenarios has significantly impeded the learning of geometric graph representation. To cope with this challenge, we propose the Graph Multi-View Pre-training (GraphMVP) framework where self-supervised learning (SSL) is performed by leveraging the correspondence and consistency between 2D topological structures and 3D geometric views. GraphMVP effectively learns a 2D molecular graph encoder that is enhanced by richer and more discriminative 3D geometry. We further provide theoretical insights to justify the effectiveness of GraphMVP. Finally, comprehensive experiments show that GraphMVP can consistently outperform existing graph SSL methods.

In recent years, a growing experimental program has begun to search for sub-GeV dark matter through its scattering with electrons. An associated theoretical challenge is to compute the dark matter scattering rate in experimental targets, and to find materials with large scattering rates. In this paper we point out that, if dark matter scatters through a mediator that couples to EM charge, then electromagnetic sum rules place limits on the achievable scattering rates. These limits serve as a useful sanity check for calculations, as well as setting a theoretical target for proposed detection methods. Motivated by this analysis, we explore how conductor-dielectric heterostructures can result in enhanced scattering rates compared to bulk conductors, for dark matter masses $\lesssim$ MeV. These effects could be especially important in computing the scattering rates from thin-film targets, e.g. superconducting detectors such as SNSPDs, TESs or MKIDs, for which the scattering rate could be enhanced by orders of magnitude at low enough dark matter masses, as well as introducing or enhancing directional dependence., Comment: 17 pages, 10 figures; v3, approximately matches version published in PRD

Uncovering the nature of dark matter is one of the most important goals of particle physics. Light bosonic particles, such as the dark photon, are well-motivated candidates: they are generally long-lived, weakly-interacting, and naturally produced in the early universe. In this work, we report on LAMPOST (Light $A'$ Multilayer Periodic Optical SNSPD Target), a proof-of-concept experiment searching for dark photon dark matter in the eV mass range, via coherent absorption in a multi-layer dielectric haloscope. Using a superconducting nanowire single-photon detector (SNSPD), we achieve efficient photon detection with a dark count rate (DCR) of $\sim 6\times10^{-6}$ counts/s. We find no evidence for dark photon dark matter in the mass range of $\sim 0.7$-$0.8$ eV with kinetic mixing $\epsilon \gtrsim 10^{-12}$, improving existing limits in $\epsilon$ by up to a factor of two. With future improvements to SNSPDs, our architecture could probe significant new parameter space for dark photon and axion dark matter in the meV to 10 eV mass range.

The singularities of serial robotic manipulators are those configurations in which the robot loses the ability to move in at least one direction. Hence, their identification is fundamental to enhance the performance of current control and motion planning strategies. While classical approaches entail the computation of the determinant of either a 6x n or nxn matrix for an n degrees of freedom serial robot, this work addresses a novel singularity identification method based on modelling the twists defined by the joint axes of the robot as vectors of the six-dimensional and three-dimensional geometric algebras. In particular, it consists of identifying which configurations cause the exterior product of these twists to vanish. In addition, since rotors represent rotations in geometric algebra, once these singularities have been identified, a distance function is defined in the configuration space C such that its restriction to the set of singular configurations S allows us to compute the distance of any configuration to a given singularity. This distance function is used to enhance how the singularities are handled in three different scenarios, namely motion planning, motion control and bilateral teleoperation., Comment: 35 pages, 6 figures

This work addresses the inverse kinematics of serial robots using conformal geometric algebra. Classical approaches include either the use of homogeneous matrices, which entails high computational cost and execution time or the development of particular geometric strategies that cannot be generalized to arbitrary serial robots. In this work, we present a compact, elegant and intuitive formulation of robot kinematics based on conformal geometric algebra that provides a suitable framework for the closed-form resolution of the inverse kinematic problem for manipulators with a spherical wrist. For serial robots of this kind, the inverse kinematics problem can be split in two subproblems: the position and orientation problems. The latter is solved by appropriately splitting the rotor that defines the target orientation into three simpler rotors, while the former is solved by developing a geometric strategy for each combination of prismatic and revolute joints that forms the position part of the robot. Finally, the inverse kinematics of 7 DoF redundant manipulators with a spherical wrist is solved by extending the geometric solutions obtained in the non-redundant case., Comment: 33 pages, 10 figures