Ask or search anything...
University of GroningenUniversity Medical Center GroningenJWST and ALMA discern the assembly of structural and obscured components in a high-redshift starburst galaxy
10 May 2024
CNRS
California Institute of TechnologyWe present observations and analysis of the starburst, PACS-819, at z=1.45 ( M), using high-resolution (; 0.8 kpc) ALMA and multi-wavelength JWST images from the COSMOS-Web program. Dissimilar to HST/ACS images in the rest-frame UV, the redder NIRCam and MIRI images reveal a smooth central mass concentration and spiral-like features, atypical for such an intense starburst. Through dynamical modeling of the CO J=5--4 emission with ALMA, PACS-819 is rotation-dominated thus has a disk-like nature. However, kinematic anomalies in CO and asymmetric features in the bluer JWST bands (e.g., F150W) support a more disturbed nature likely due to interactions. The JWST imaging further enables us to map the distribution of stellar mass and dust attenuation, thus clarifying the relationships between different structural components, not discernable in the previous HST images. The CO J = 5 -- 4 and FIR dust continuum emission are co-spatial with a heavily-obscured starbursting core (<1 kpc) which is partially surrounded by much less obscured star-forming structures including a prominent arc, possibly a tidally-distorted dwarf galaxy, and a clump, either a sign of an ongoing violent disk instability or a recently accreted low-mass satellite. With spatially-resolved maps, we find a high molecular gas fraction in the central area reaching (/) and short depletion times ( 120 Myrs) across the entire system. These observations provide insights into the complex nature of starbursts in the distant universe and underscore the wealth of complementary information from high-resolution observations with both ALMA and JWST.
A Global Perspective with Updated Constraints on the Ultra-hot Jupiter WASP-19b: Atmospheric Properties and Stellar Activity
University College LondonWe present a detailed reanalysis of the atmospheric properties of WASP-19b, an ultra-hot Jupiter (1.14 M Jup, 1.41 R Jup) orbiting an active Sun-like star every 0.79 day. We reanalyze a transit and secondary eclipse of WASP-19b observed by the Hubble Space Telescope's Wide Field Camera 3 spectrograph (1.1 - 1.7 microns). When combined with Spitzer photometry at longer wavelengths, our analyses indicate the presence of water absorption features in both the planet's transmission and emission spectra, consistent with results from previously published studies. We jointly fit WASP-19b's dayside emission and transmission spectra with a retrieval model in order to constrain its atmospheric composition, and explore the effect of stellar activity on its transmission spectrum in greater depth. We also compare our dayside emission spectrum to predictions from a general circulation model, and conclude that magnetic drag appears to be relatively unimportant in shaping WASP-19b's atmospheric circulation. Lastly, we compare the size of WASP-19b's dayside water absorption feature to the population of hot Jupiters with similar measurements, and show that it is located in the transitional irradiation regime where temperature inversions first begin to emerge. As in previous studies, we find that the current observations provide relatively weak constraints on this planet's atmospheric properties. These constraints could be significantly improved by the addition of spectroscopically resolved observations at longer wavelengths with JWST/NIRSpec PRISM.
Dark matter measurements combining stellar and HI kinematics: 30% outliers with low dark matter content at
Chinese Academy of Sciences
University of St AndrewsWe construct the Schwarzschild dynamical models for 11 early-type galaxies
with the SAURON and Mitchell stellar IFUs out to , and
construct dynamical models with combined stellar and HI kinematics for a
subsample of 4 galaxies with HI velocity fields out to
obtained from the Westerbork Synthesis Radio Telescope, thus robustly obtaining
the dark matter content out to large radii for these galaxies. Adopting a
generalised-NFW dark matter profile, we measure an NFW-like density cusp in the
dark matter inner slopes for all sample galaxies, with a mean value of
(rms scatter ). The mean dark matter fraction for the
sample is within , and increases to at $2
R_\mathrm{e}0.65 R_\mathrm{e}$. The dark matter fractions within
of these galaxies are systematically lower than the
predictions of both the TNG-100 and EAGLE simulations. For the dark matter
fractions within and , 40% and 70% galaxies
are consistent with either the TNG-100 or the EAGLE predictions,
while the remaining 60% and 30% galaxies lie below the region.
Combined with 36 galaxies with dark matter fractions measured out to $5
R_\mathrm{e}$ in the literature, about 10% of these 47 galaxies lie below the
region of the TNG-100 or EAGLE predictions.
Modification of the halo mass function by kurtosis associated with primordial non-Gaussianity
27 Jun 2011
Nagoya UniversityWe study the halo mass function in the presence of the kurtosis type of
primordial non-Gaussianity. The kurtosis corresponds to the trispectrum as
defined in Fourier space. The primordial trispectrum is commonly characterized
by two parameters, and . As applications of the
derived non-Gaussian mass function, we consider the effect on the abundance of
void structure, the effect on early star formation and on formation of the most
massive object at high redshift. We show that by comparing the effects of
primordial non-Gaussianity on cluster abundance with that on void abundance, we
can distinguish between the skewness and the kurtosis types of primordial
non-Gaussianity. As for early star formation, we show that the kurtosis type of
primordial non-Gaussianity seems not to affect the reionization history of the
Universe on average. However, at high redshifts (up to ) such
non-Gaussianity does somewhat affect the early stages of reionization.
A Primer on the Inner Workings of Transformer-based Language Models
13 Oct 2024
Gabriele Sarti
Javi Ferrando
MetaThis primer provides a comprehensive technical introduction to interpreting transformer-based language models, particularly generative decoder-only architectures, by consolidating current techniques and systematically mapping discovered internal mechanisms and behaviors across model components.
View blog
A Systematic Analysis of Hybrid Linear Attention
08 Jul 2025
ByteDanceThis work systematically analyzes hybrid linear attention architectures to balance computational efficiency with long-range recall in large language models. The research demonstrates that a linear attention model's standalone performance does not predict its effectiveness in hybrid setups and identifies selective gating, hierarchical recurrence, and controlled forgetting as crucial architectural properties enabling near-Transformer recall with substantial KV-cache memory reductions.
View blog
Aligning Generalisation Between Humans and Machines
Recent 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.
TRiSM for Agentic AI: A Review of Trust, Risk, and Security Management in LLM-based Agentic Multi-Agent Systems
15 Sep 2025
Ranjan Sapkota
Cornell UniversityResearchers from the Vector Institute, Cornell University, and the University of Groningen present a comprehensive Trust, Risk, and Security Management (TRiSM) framework tailored for LLM-based Agentic Multi-Agent Systems (AMAS). This framework delineates unique threats, a risk taxonomy, and novel evaluation metrics, while aligning AMAS development with international AI governance and regulatory standards.
View blog
EAGER: Entropy-Aware GEneRation for Adaptive Inference-Time Scaling
13 Oct 2025
CohereEAGER introduces an entropy-aware generation method for adaptively scaling large language model inference, which dynamically adjusts computational budget based on token-wise uncertainty. This approach achieved up to 37% higher Pass@k performance while using 65% fewer tokens on various reasoning benchmarks, effectively improving the efficiency-performance trade-off for LLM reasoning tasks.
View blog
Lifelong Robot Library Learning: Bootstrapping Composable and Generalizable Skills for Embodied Control with Language Models
15 Jul 2024
Lifelong Robot Library Learning (LRLL), developed by researchers at the University of Groningen, enables robots to continually expand their skill set by autonomously abstracting new, generalizable skills using large language models. The framework demonstrates superior generalization in simulated manipulation tasks, achieving 86.6% success on unseen instructions, and successfully performs zero-shot transfer of learned skills from simulation to a real robot without catastrophic forgetting.
View blog
Q-S5: Towards Quantized State Space Models
13 Jun 2024
University of CambridgeIn the quest for next-generation sequence modeling architectures, State Space Models (SSMs) have emerged as a potent alternative to transformers, particularly for their computational efficiency and suitability for dynamical systems. This paper investigates the effect of quantization on the S5 model to understand its impact on model performance and to facilitate its deployment to edge and resource-constrained platforms. Using quantization-aware training (QAT) and post-training quantization (PTQ), we systematically evaluate the quantization sensitivity of SSMs across different tasks like dynamical systems modeling, Sequential MNIST (sMNIST) and most of the Long Range Arena (LRA). We present fully quantized S5 models whose test accuracy drops less than 1% on sMNIST and most of the LRA. We find that performance on most tasks degrades significantly for recurrent weights below 8-bit precision, but that other components can be compressed further without significant loss of performance. Our results further show that PTQ only performs well on language-based LRA tasks whereas all others require QAT. Our investigation provides necessary insights for the continued development of efficient and hardware-optimized SSMs.
Gauge symmetry and the arrow of time: How to count what counts
18 Sep 2025
This thesis addresses two major problems in the philosophy of physics. The first is how to identify the minimal physical content of a theory; that is, what features of a theory are truly needed to make predictions, and what can be removed without changing its empirical consequences. The second is the problem of time's arrow: why time seems to have a direction, even though the fundamental laws of physics treat the past and future symmetrically.
I show that answering the first question leads to insights about the second. In particular, I argue that the overall size of the Universe is not used to make predictions in cosmology, and so should not count as part of the theory's minimal physical content. Describing the Universe without this feature leads to a striking result: the arrow of time becomes a local phenomenon. Observers like us who see a Universe full of matter clumped together to form structures like stars and planets are statistically much more likely to see increasing clumpiness into the future than into the past. This tendency helps explain our experience of time's direction.
Beyond Token Probes: Hallucination Detection via Activation Tensors with ACT-ViT
NVIDIAResearchers at Technion and Nvidia Research developed ACT-ViT, an architecture that re-conceptualizes large language model activation tensors as images to detect hallucinations with a Vision Transformer backbone. The method demonstrates superior performance and remarkable cross-LLM generalization, achieving inference times of approximately 10^-5 seconds per instance, vastly improving efficiency over existing techniques.
View blog
A million-solar-mass object detected at cosmological distance using gravitational imaging
University of California, DavisStructure on sub-galactic scales provides important tests of galaxy formation models and the nature of dark matter. However, such objects are typically too faint to provide robust mass constraints. Here, we report the discovery of an extremely low-mass object detected via its gravitational perturbation to a thin lensed arc observed with milli-arcsecond-resolution very long baseline interferometry (VLBI). The object was identified using a non-parametric gravitational imaging technique and confirmed using independent parametric modelling. It contains a mass of within a projected radius of 80 parsecs at an assumed redshift of 0.881. This detection is extremely robust and precise, with a statistical significance of 26, a 3.3 per cent fractional uncertainty on , and an astrometric uncertainty of 194 as. This is the lowest-mass object known to us, by two orders of magnitude, to be detected at a cosmological distance by its gravitational effect. This work demonstrates the observational feasibility of using gravitational imaging to probe the million-solar-mass regime far beyond our local Universe.
The Anatomy of Alignment: Decomposing Preference Optimization by Steering Sparse Features
25 Sep 2025
Prevailing alignment methods induce opaque parameter changes, making it difficult to audit what the model truly learns. To address this, we introduce Feature Steering with Reinforcement Learning (FSRL), a framework that trains a lightweight adapter to steer model behavior by modulating interpretable sparse features. First, we theoretically show that this mechanism is principled and expressive enough to approximate the behavioral shifts of post-training processes. Then, we apply this framework to the task of preference optimization and perform a causal analysis of the learned policy. We find that the model relies on stylistic presentation as a proxy for quality, disproportionately steering features related to style and formatting over those tied to alignment concepts like honesty. Despite exploiting this heuristic, FSRL proves to be an effective alignment method, achieving a substantial reduction in preference loss. Overall, FSRL offers an interpretable control interface and a practical way to diagnose how preference optimization pressures manifest at the feature level.
PARSE-Ego4D: Personal Action Recommendation Suggestions for Egocentric Videos
25 Jul 2024
Google
University of Central FloridaGoogle researchers introduce PARSE-Ego4D, a dataset designed to enable proactive AI assistance for egocentric videos by providing personal action recommendation annotations. The dataset was created by leveraging large language models to generate initial suggestions, which were then rigorously validated and refined through extensive human annotation, demonstrating the viability of this hybrid approach and setting new benchmarks for intelligent AR/VR systems.
View blog
Are You Doubtful? Oh, It Might Be Difficult Then! Exploring the Use of Model Uncertainty for Question Difficulty Estimation
17 Apr 2025
In an educational setting, an estimate of the difficulty of multiple-choice
questions (MCQs), a commonly used strategy to assess learning progress,
constitutes very useful information for both teachers and students. Since human
assessment is costly from multiple points of view, automatic approaches to MCQ
item difficulty estimation are investigated, yielding however mixed success
until now. Our approach to this problem takes a different angle from previous
work: asking various Large Language Models to tackle the questions included in
three different MCQ datasets, we leverage model uncertainty to estimate item
difficulty. By using both model uncertainty features as well as textual
features in a Random Forest regressor, we show that uncertainty features
contribute substantially to difficulty prediction, where difficulty is
inversely proportional to the number of students who can correctly answer a
question. In addition to showing the value of our approach, we also observe
that our model achieves state-of-the-art results on the USMLE and CMCQRD
publicly available datasets.
The warm outer layer of a Little Red Dot as the source of [Fe II] and collisional Balmer lines with scattering wings
University of CopenhagenThe population of the Little Red Dots (LRDs) may represent a key phase of supermassive black hole (SMBH) growth. A cocoon of dense excited gas is emerging as key component to explain the most striking properties of LRDs, such as strong Balmer breaks and Balmer absorption, as well as the weak IR emission. To dissect the structure of LRDs, we analyze new deep JWST/NIRSpec PRISM and G395H spectra of FRESCO-GN-9771, one of the most luminous known LRDs at . These reveal a strong Balmer break, broad Balmer lines and very narrow [O III] emission. We unveil a forest of optical [Fe II] lines, which we argue is emerging from a dense ( cm) warm layer with electron temperature K. The broad wings of H and H have an exponential profile due to electron scattering in this same layer. The high flux ratio of is an indicator of collisional excitation and resonant scattering dominating the Balmer line emission. A narrow H component, unseen in the other two Balmer lines due to outshining by the broad components, could trace the ISM of a normal host galaxy with a star formation rate M yr. The warm layer is mostly opaque to Balmer transitions, producing a characteristic P-Cygni profile in the line centers suggesting outflowing motions. This same layer is responsible for shaping the Balmer break. The broad-band spectrum can be reasonably matched by a simple photoionized slab model that dominates the Å continuum and a low mass ( M) galaxy that could explain the narrow [O III], with only subdominant contribution to the UV continuum. Our findings indicate that Balmer lines are not directly tracing gas kinematics near the SMBH and that the BH mass scale is likely much lower than virial indicators suggest.
BabyBabelLM: A Multilingual Benchmark of Developmentally Plausible Training Data
11 Oct 2025
We present BabyBabelLM, a multilingual collection of datasets modeling the language a person observes from birth until they acquire a native language. We curate developmentally plausible pretraining data aiming to cover the equivalent of 100M English words of content in each of 45 languages. We compile evaluation suites and train baseline models in each language. BabyBabelLM aims to facilitate multilingual pretraining and cognitive modeling.
A Minimal Formulation of Session Types: The Sessions of Trios in Concert
12 May 2025
Session types are a type-based approach to the verification of message-passing programs. They specify communication structures essential to enforcing program correctness; by relying on sequencing constructs, a session type can precisely describe the intended order of communication actions through a channel.
In this paper we study a fragment of session types that makes a very limited use of sequencing; we call it minimal session types. In the context of a core process calculus with sessions and higher-order concurrency, we establish two technical results. First, we prove that every process P typable with standard session types can be compiled down into a process D(P) typable with minimal session types. Second, we prove that P and D(P) are behaviorally equivalent. These results show that having sequencing in both processes and session types is convenient, but that only sequencing in processes is truly indispensable, as it can correctly codify sequencing in types.
Our developments draw inspiration from work by Parrow on behavior-preserving decompositions of untyped processes using trios, i.e., processes with exactly three nested prefixes. By casting Parrow's approach in the realm of typed processes, our developments reveal a conceptually simple formulation of session types, supported by static and dynamic correctness results.
There are no more papers matching your filters at the moment.