Ask or search anything...
Exploring Behaviors of Hybrid Systems via the Voronoi Bias over Output Signals
05 Dec 2024
In this paper, we consider an analysis of temporal properties of hybrid systems based on simulations, so-called falsification of requirements. We present a novel exploration-based algorithm for falsification of black-box models of hybrid systems based on the Voronoi bias in the output space. This approach is inspired by techniques used originally in motion planning: rapidly exploring random trees. Instead of commonly employed exploration that is based on coverage of inputs, the proposed algorithm aims to cover all possible outputs directly. Compared to other state-of-the-art falsification tools, it also does not require robustness or other guidance metrics tied to a specific behavior that is being falsified. This allows our algorithm to falsify specifications for which robustness is not conclusive enough to guide the falsification procedure.
Aligning Generalisation Between Humans and Machines
University College LondonRecent advances in AI -- including generative approaches -- have resulted in
technology that can support humans in scientific discovery and forming
decisions, but may also disrupt democracies and target individuals. The
responsible use of AI and its participation in human-AI teams increasingly
shows the need for AI alignment, that is, to make AI systems act according to
our preferences. A crucial yet often overlooked aspect of these interactions is
the different ways in which humans and machines generalise. In cognitive
science, human generalisation commonly involves abstraction and concept
learning. In contrast, AI generalisation encompasses out-of-domain
generalisation in machine learning, rule-based reasoning in symbolic AI, and
abstraction in neurosymbolic AI. In this perspective paper, we combine insights
from AI and cognitive science to identify key commonalities and differences
across three dimensions: notions of, methods for, and evaluation of
generalisation. We map the different conceptualisations of generalisation in AI
and cognitive science along these three dimensions and consider their role for
alignment in human-AI teaming. This results in interdisciplinary challenges
across AI and cognitive science that must be tackled to provide a foundation
for effective and cognitively supported alignment in human-AI teaming
scenarios.
A comparative analysis of rank aggregation methods for the partial label ranking problem
08 Sep 2025
An evaluation of rank aggregation techniques for the partial label ranking problem identifies that adapted scoring-based methods, Borda and Copeland, demonstrate superior robustness to incomplete preference data and are approximately twice as fast as the state-of-the-art Bucket pivot algorithm. These findings suggest more efficient and reliable approaches for modeling preferences with ties, particularly relevant for applications like LLM fine-tuning.
View blog
Cosmic Birefringence from the Atacama Cosmology Telescope Data Release 6
the University of TokyoThe polarized light of the cosmic microwave background is sensitive to new physics that violates parity symmetry. For example, the interaction of photons with the fields of elusive dark matter and dark energy could cause a uniform rotation of the plane of linear polarization across the sky, an effect known as cosmic birefringence. We extract the cosmological rotation angle, , using Bayesian analysis of parity-violating correlations, and , of polarization data from the Atacama Cosmology Telescope (ACT) Data Release 6. We use prior probabilities for instrumental miscalibration angles derived from the optics model for the ACT telescope and instruments, and marginalize over a residual intensity-to-polarization leakage. We measure (68\% confidence level), which excludes with a statistical significance of . Although there remain systematics in the ACT data that are not understood and do not allow us to draw strong cosmological conclusions, this result is consistent with previous independent results from the \wmap\ and \planck\ missions. It is suggestive that independent data sets and analyses using different methodologies have yielded the same sign and comparable magnitudes for .
The Ethics of Advanced AI Assistants
28 Apr 2024
Joel Lehman
Victoria Krakovna
Google DeepMindResearchers from Google DeepMind and Google Research, in collaboration with academic institutions, present the first systematic ethical framework for advanced AI assistants. Their analysis details the opportunities and profound risks of these new systems, emphasizing challenges in human-AI interaction and societal integration, and calls for new sociotechnical evaluation methods.
View blog
Beyond Magic Words: Sharpness-Aware Prompt Evolving for Robust Large Language Models with TARE
17 Nov 2025
Jinhe Bi
UCLA
University of ChicagoThe performance of Large Language Models (LLMs) hinges on carefully engineered prompts. However, prevailing prompt optimization methods, ranging from heuristic edits and reinforcement learning to evolutionary search, primarily target point-wise accuracy. They seldom enforce paraphrase invariance or searching stability, and therefore cannot remedy this brittleness in practice. Automated prompt search remains brittle: small, semantically preserving paraphrases often cause large performance swings. We identify this brittleness as the textual sharpness of the prompt landscape. In this work, we provide the first formal treatment of textual sharpness in the discrete, semantic space of prompts, together with an operational robustness criterion over a semantic neighborhood; the design is black-box or API-only, requiring no gradients to update the model's parameters. Then we introduce TARE (Textual Sharpness-Aware Evolving), a derivative-free framework that alternates between an inner, sampling-based adversarial search that stresses a prompt with hard paraphrases and an outer, robust selection that prefers candidates whose neighborhoods remain strong. We further propose ATARE, which learns anisotropic weights to shape the semantic neighborhood and adapts its radius over time to balance exploration and fidelity. Diverse tasks evaluate our methods, whose design for minimizing textual sharpness gap leads to prompts that preserve accuracy under paraphrasing, outperforming accuracy-only prompt search while remaining computationally practical.
X-ray panorama of the SS433/W50 complex by SRG/eROSITA
16 Oct 2025
The SRG/eROSITA mission has delivered the first full-extent X-ray map of the SS 433/W50 complex with high spatial resolution, distinguishing between diffuse thermal plasma, narrow filaments, and non-thermal Extended X-ray Jets. Findings reveal a three-stage energy flow, with the Eastern lobe containing ~4 x 10^50 erg of thermal energy and EXJs exhibiting a non-thermal power-law spectrum with Γ ranging from 1.4 to 2.2.
View blog
Which LIME should I trust? Concepts, Challenges, and Solutions
31 Mar 2025
This survey provides the first comprehensive overview and a novel two-dimensional taxonomy of Local Interpretable Model-agnostic Explanations (LIME) variants, categorizing them by the specific issues they address and their technical modifications. It identifies persistent challenges in LIME research, such as reproducibility and inconsistent evaluation, and offers an interactive resource for practitioners.
View blog
A General Constructive Upper Bound on Shallow Neural Nets Complexity
07 Oct 2025
We provide an upper bound on the number of neurons required in a shallow
neural network to approximate a continuous function on a compact set with a
given accuracy. This method, inspired by a specific proof of the
Stone-Weierstrass theorem, is constructive and more general than previous
bounds of this character, as it applies to any continuous function on any
compact set.
Is it Gaussian? Testing bosonic quantum states
08 Oct 2025
University of AmsterdamGaussian states are widely regarded as one of the most relevant classes of continuous-variable (CV) quantum states, as they naturally arise in physical systems and play a key role in quantum technologies. This motivates a fundamental question: given copies of an unknown CV state, how can we efficiently test whether it is Gaussian? We address this problem from the perspective of representation theory and quantum learning theory, characterizing the sample complexity of Gaussianity testing as a function of the number of modes. For pure states, we prove that just a constant number of copies is sufficient to decide whether the state is exactly Gaussian. We then extend this to the tolerant setting, showing that a polynomial number of copies suffices to distinguish states that are close to Gaussian from those that are far. In contrast, we establish that testing Gaussianity of general mixed states necessarily requires exponentially many copies, thereby identifying a fundamental limitation in testing CV systems. Our approach relies on rotation-invariant symmetries of Gaussian states together with the recently introduced toolbox of CV trace-distance bounds.
Planet Earth in reflected and polarized light -- III. Modeling and analysis of a decade-long catalog of Earthshine observations
European Southern ObservatoryEarthshine observations offer a unique opportunity to study Earth as an exoplanet seen from the Moon. As the Sun-Earth-Moon geometry changes, Earth can be observed as a spatially unresolved exoplanet at different phase angles, providing important context for future observations of Earth-like exoplanets. Here, we present a catalog of Earthshine polarization spectra obtained with FORS2 on the VLT, covering diverse scenes, surface conditions, cloud properties, and weather patterns for over a decade. For the first time, we model this extensive dataset in detail using a homogeneous modeling framework. Previous efforts to model some of these spectra struggled to reproduce the observed polarization continuum, even with advanced 3D radiative transfer models incorporating satellite-derived surface and atmospheric data. We improve upon this with a 3D model that includes subgrid cloud variability, wavelength-dependent surface albedo maps, and an accurate treatment of ocean glint. Our simulations successfully reproduce most observed spectra to a much higher precision than previously possible. Our statistical analysis reveals that the spectral slope in the visible can distinguish between ocean and mixed surfaces in both reflected and polarized light, which is not possible using broadband filters alone. Polarized light at large phase angles, beyond the Rayleigh scattering regime, is particularly effective in differentiating oceans from land, unlike reflected light. We also identify correlations between cloud optical thickness and the polarized spectral slope, and between cloud cover and broadband B-R differences in reflected light, demonstrating the diagnostic power of these observations. This work highlight the potential of polarization for characterizing Earth-like exoplanets. From polarization alone, we can infer oceans, vegetation, and an active water cycle, key indicators of a habitable planet.
The GECKOS Survey: Resolved, multiphase observations of mass-loading and gas density in the galactic wind of NGC 4666
22 Sep 2025
CNRSWe present a multiphase, resolved study of the galactic wind extending from the nearby starburst galaxy NGC 4666. For this we use VLT/MUSE observations from the GECKOS program and HI data from the WALLABY survey. We identify both ionised and HI gas in a biconical structure extending to at least 8 kpc from the galaxy disk, with increasing velocity offsets above the midplane in both phases, consistent with a multiphase wind. The measured electron density, using [SII], differs significantly from standard expectations of galactic winds. We find electron density declines from the galaxy centre to kpc, then rises again, remaining high ( cm) out to 5 kpc. We find that HI dominates the mass loading. The total HI mass outflow rate (above kpc) is between , accounting for uncertainties from disk-blurring and group interactions. The total ionised mass outflow rate (traced by H) is between and , depending on assumptions. From ALMA/ACA observations, we place an upper-limit on CO flux in the outflow which correlates to . We also show that the entire outflow is not limited to the bicone, but a secondary starburst at the edge generates a more widespread outflow, which should be included in simulations. The cool gas in NGC 4666 wind has insufficient velocity to escape the halo of a galaxy of its mass, especially because most of the mass is present in the slower atomic phase. This strong biconical wind contributes to gas cycling around the galaxy.
Turn up the light: Radiative efficiency of protostars at birth
18 Sep 2025
Early stages of stellar birth comprise of a two-step process involving the formation of two hydrostatic cores. The second step of gravitational collapse sets the radiative efficiency and accretion rate of the young protostar. These two parameters, of prime importance for protostellar evolution, dictate the luminosities and thus play a key role in deciphering the current discrepancy between observational surveys and theoretical models. In this letter, we provide quantitative estimates on the evolution of the radiative efficiency and accretion rate obtained from self-consistent, high-resolution, radiative hydrodynamic simulations performed using the codes PLUTO and RAMSES. The main highlight of our result is that the radiative efficiency reaches unity, that is, supercriticality, relatively quickly after protostellar birth. Supercriticality at the accretion shock is a necessary condition for cold accretion. Our results thus support a rapid transition to the cold accretion scenario, which is one of the assumptions used in Pre-Main Sequence (PMS) models working towards solutions to explain observational data. We briefly discuss the implications of the time evolution of the radiative efficiency factor in the context of the luminosity problem, the Protostellar Luminosity Function (PLF), PMS evolution, accurate sink properties, and the stellar Initial Mass Function (IMF).
A DECADE of dwarfs: first detection of weak lensing around spectroscopically confirmed low-mass galaxies
24 Sep 2025
University of WashingtonWe present the first detection of weak gravitational lensing around spectroscopically confirmed dwarf galaxies, using the large overlap between DESI DR1 spectroscopic data and DECADE/DES weak lensing catalogs. A clean dwarf galaxy sample with well-defined redshift and stellar mass cuts enables excess surface mass density measurements in two stellar mass bins ( and ), with signal-to-noise ratios of and respectively. This signal-to-noise drops to and respectively for measurements without applying individual inverse probability (IIP) weights, which mitigates fiber incompleteness from DESI's targeting. The measurements are robust against variations in stellar mass estimates, photometric shredding, and lensing calibration systematics. Using a simulation-based modeling framework with stellar mass function priors, we constrain the stellar mass-halo mass relation and find a satellite fraction of , which is higher than previous photometric studies but lower than CDM predictions. We find that IIP weights have a significant impact on lensing measurements and can change the inferred by a factor of two, highlighting the need for accurate fiber incompleteness corrections for dwarf galaxy samples. Our results open a new observational window into the galaxy-halo connection at low masses, showing that future massively multiplexed spectroscopic observations and weak lensing data will enable stringent tests of galaxy formation models and CDM predictions.
Introduction to Quantum Noise, Measurement and Amplification
15 Apr 2010
McGill University
Yale UniversityThe topic of quantum noise has become extremely timely due to the rise of quantum information physics and the resulting interchange of ideas between the condensed matter and AMO/quantum optics communities. This review gives a pedagogical introduction to the physics of quantum noise and its connections to quantum measurement and quantum amplification. After introducing quantum noise spectra and methods for their detection, we describe the basics of weak continuous measurements. Particular attention is given to treating the standard quantum limit on linear amplifiers and position detectors using a general linear-response framework. We show how this approach relates to the standard Haus-Caves quantum limit for a bosonic amplifier known in quantum optics, and illustrate its application for the case of electrical circuits, including mesoscopic detectors and resonant cavity detectors.
Lectures in Quantum Gravity
08 Oct 2025
Formulating a quantum theory of gravity lies at the heart of fundamental theoretical physics. This collection of lecture notes encompasses a selection of topics that were covered in six mini-courses at the Nordita PhD school "Towards Quantum Gravity". The scope was to provide a coherent picture, from its foundation to forefront research, emphasizing connections between different areas. The lectures begin with perturbative quantum gravity and effective field theory. Subsequently, two ultraviolet-complete approaches are presented: asymptotically safe gravity and string theory. Finally, elements of quantum effects in black hole spacetimes are discussed.
Atlas: A Novel Pathology Foundation Model by Mayo Clinic, Charité, and Aignostics
10 Jan 2025
Recent advances in digital pathology have demonstrated the effectiveness of foundation models across diverse applications. In this report, we present Atlas, a novel vision foundation model based on the RudolfV approach. Our model was trained on a dataset comprising 1.2 million histopathology whole slide images, collected from two medical institutions: Mayo Clinic and Charité - Universtätsmedizin Berlin. Comprehensive evaluations show that Atlas achieves state-of-the-art performance across twenty-one public benchmark datasets, even though it is neither the largest model by parameter count nor by training dataset size.
A Cold and Super-Puffy Planet on a Polar Orbit
30 Sep 2025
University of OxfordScientists report the discovery of TOI-4507 b, a young, long-period exoplanet classified as a "super-puff" due to its exceptionally low density, orbiting an F-type host star on a nearly polar path. The study provides precise measurements of its physical and orbital parameters, including a 3D stellar obliquity of 82.0 degrees, and identifies it as a prime target for atmospheric characterization with JWST due to its high TSM of over 100.
View blog
Ewald-based Long-Range Message Passing for Molecular Graphs
Google ResearchNeural architectures that learn potential energy surfaces from molecular data have undergone fast improvement in recent years. A key driver of this success is the Message Passing Neural Network (MPNN) paradigm. Its favorable scaling with system size partly relies upon a spatial distance limit on messages. While this focus on locality is a useful inductive bias, it also impedes the learning of long-range interactions such as electrostatics and van der Waals forces. To address this drawback, we propose Ewald message passing: a nonlocal Fourier space scheme which limits interactions via a cutoff on frequency instead of distance, and is theoretically well-founded in the Ewald summation method. It can serve as an augmentation on top of existing MPNN architectures as it is computationally inexpensive and agnostic to architectural details. We test the approach with four baseline models and two datasets containing diverse periodic (OC20) and aperiodic structures (OE62). We observe robust improvements in energy mean absolute errors across all models and datasets, averaging 10% on OC20 and 16% on OE62. Our analysis shows an outsize impact of these improvements on structures with high long-range contributions to the ground truth energy.
Visions in Quantum Gravity
08 Oct 2025
University of CambridgeTo deepen our understanding of Quantum Gravity and its connections with black holes and cosmology, building a common language and exchanging ideas across different approaches is crucial. The Nordita Program "Quantum Gravity: from gravitational effective field theories to ultraviolet complete approaches" created a platform for extensive discussions, aimed at pinpointing both common grounds and sources of disagreements, with the hope of generating ideas and driving progress in the field. This contribution summarizes the twelve topical discussions held during the program and collects individual thoughts of speakers and panelists on the future of the field in light of these discussions.
There are no more papers matching your filters at the moment.