The rapid advancement of AI has expanded its capabilities across domains, yet introduced critical technical vulnerabilities, such as algorithmic bias and adversarial sensitivity, that pose significant societal risks, including misinformation, inequity, security breaches, physical harm, and eroded public trust. These challenges highlight the urgent need for robust AI governance. We propose a comprehensive framework integrating technical and societal dimensions, structured around three interconnected pillars: Intrinsic Security (system reliability), Derivative Security (real-world harm mitigation), and Social Ethics (value alignment and accountability). Uniquely, our approach unifies technical methods, emerging evaluation benchmarks, and policy insights to promote transparency, accountability, and trust in AI systems. Through a systematic review of over 300 studies, we identify three core challenges: (1) the generalization gap, where defenses fail against evolving threats; (2) inadequate evaluation protocols that overlook real-world risks; and (3) fragmented regulations leading to inconsistent oversight. These shortcomings stem from treating governance as an afterthought, rather than a foundational design principle, resulting in reactive, siloed efforts that fail to address the interdependence of technical integrity and societal trust. To overcome this, we present an integrated research agenda that bridges technical rigor with social responsibility. Our framework offers actionable guidance for researchers, engineers, and policymakers to develop AI systems that are not only robust and secure but also ethically aligned and publicly trustworthy. The accompanying repository is available at this https URL.

A digital twin is a virtual replica of a real-world physical phenomena that uses mathematical modeling to characterize and simulate its defining features. By constructing digital twins for disease processes, we can perform in-silico simulations that mimic patients' health conditions and counterfactual outcomes under hypothetical interventions in a virtual setting. This eliminates the need for invasive procedures or uncertain treatment decisions. In this paper, we propose a method to identify digital twin model parameters using only noninvasive patient health data. We approach the digital twin modeling as a composite inverse problem, and observe that its structure resembles pretraining and finetuning in self-supervised learning (SSL). Leveraging this, we introduce a physics-informed SSL algorithm that initially pretrains a neural network on the pretext task of learning a differentiable simulator of a physiological process. Subsequently, the model is trained to reconstruct physiological measurements from noninvasive modalities while being constrained by the physical equations learned in pretraining. We apply our method to identify digital twins of cardiac hemodynamics using noninvasive echocardiogram videos, and demonstrate its utility in unsupervised disease detection and in-silico clinical trials.

The widespread use of Large Multimodal Models (LMMs) has raised concerns about model toxicity. However, current research mainly focuses on explicit toxicity, with less attention to some more implicit toxicity regarding prejudice and discrimination. To address this limitation, we introduce a subtler type of toxicity named dual-implicit toxicity and a novel toxicity benchmark termed MDIT-Bench: Multimodal Dual-Implicit Toxicity Benchmark. Specifically, we first create the MDIT-Dataset with dual-implicit toxicity using the proposed Multi-stage Human-in-loop In-context Generation method. Based on this dataset, we construct the MDIT-Bench, a benchmark for evaluating the sensitivity of models to dual-implicit toxicity, with 317,638 questions covering 12 categories, 23 subcategories, and 780 topics. MDIT-Bench includes three difficulty levels, and we propose a metric to measure the toxicity gap exhibited by the model across them. In the experiment, we conducted MDIT-Bench on 13 prominent LMMs, and the results show that these LMMs cannot handle dual-implicit toxicity effectively. The model's performance drops significantly in hard level, revealing that these LMMs still contain a significant amount of hidden but activatable toxicity. Data are available at this https URL

The Gaia Galactic survey mission is designed and optimized to obtain astrometry, photometry, and spectroscopy of nearly two billion stars in our Galaxy. Yet as an all-sky multi-epoch survey, Gaia also observes several million extragalactic objects down to a magnitude of G~21 mag. Due to the nature of the Gaia onboard selection algorithms, these are mostly point-source-like objects. Using data provided by the satellite, we have identified quasar and galaxy candidates via supervised machine learning methods, and estimate their redshifts using the low resolution BP/RP spectra. We further characterise the surface brightness profiles of host galaxies of quasars and of galaxies from pre-defined input lists. Here we give an overview of the processing of extragalactic objects, describe the data products in Gaia DR3, and analyse their properties. Two integrated tables contain the main results for a high completeness, but low purity (50-70%), set of 6.6 million candidate quasars and 4.8 million candidate galaxies. We provide queries that select purer sub-samples of these containing 1.9 million probable quasars and 2.9 million probable galaxies (both 95% purity). We also use high quality BP/RP spectra of 43 thousand high probability quasars over the redshift range 0.05-4.36 to construct a composite quasar spectrum spanning restframe wavelengths from 72-100 nm.

We present cosmological results from the measurement of clustering of galaxy, quasar and Lyman-$\alpha$ forest tracers from the first year of observations with the Dark Energy Spectroscopic Instrument (DESI Data Release 1). We adopt the full-shape (FS) modeling of the power spectrum, including the effects of redshift-space distortions, in an analysis which has been validated in a series of supporting papers. In the flat $\Lambda$CDM cosmological model, DESI (FS+BAO), combined with a baryon density prior from Big Bang Nucleosynthesis and a weak prior on the scalar spectral index, determines matter density to $\Omega_\mathrm{m}=0.2962\pm 0.0095$, and the amplitude of mass fluctuations to $\sigma_8=0.842\pm 0.034$. The addition of the cosmic microwave background (CMB) data tightens these constraints to $\Omega_\mathrm{m}=0.3056\pm 0.0049$ and $\sigma_8=0.8121\pm 0.0053$, while further addition of the the joint clustering and lensing analysis from the Dark Energy Survey Year-3 (DESY3) data leads to a 0.4% determination of the Hubble constant, $H_0 = (68.40\pm 0.27)\,{\rm km\,s^{-1}\,Mpc^{-1}}$. In models with a time-varying dark energy equation of state, combinations of DESI (FS+BAO) with CMB and type Ia supernovae continue to show the preference, previously found in the DESI DR1 BAO analysis, for w_0>-1 and w_a<0 with similar levels of significance. DESI data, in combination with the CMB, impose the upper limits on the sum of the neutrino masses of \sum m_\nu < 0.071\,{\rm eV} at 95% confidence. DESI data alone measure the modified-gravity parameter that controls the clustering of massive particles, $\mu_0=0.11^{+0.45}_{-0.54}$, while the combination of DESI with the CMB and the clustering and lensing analysis from DESY3 constrains both modified-gravity parameters, giving $\mu_0 = 0.04\pm 0.22$ and $\Sigma_0 = 0.044\pm 0.047$, in agreement with general relativity. [Abridged.]

The Lyman-alpha forest provides strong constraints on both cosmological parameters and intergalactic medium astrophysics, which are forecast to improve further with the next generation of surveys including eBOSS and DESI. As is generic in cosmological inference, extracting this information requires a likelihood to be computed throughout a high-dimensional parameter space. Evaluating the likelihood requires a robust and accurate mapping between the parameters and observables, in this case the 1D flux power spectrum. Cosmological simulations enable such a mapping, but due to computational time constraints can only be evaluated at a handful of sample points; "emulators" are designed to interpolate between these. The problem then reduces to placing the sample points such that an accurate mapping is obtained while minimising the number of expensive simulations required. To address this, we introduce an emulation procedure that employs Bayesian optimisation of the training set for a Gaussian process interpolation scheme. Starting with a Latin hypercube sampling (other schemes with good space-filling properties can be used), we iteratively augment the training set with extra simulations at new parameter positions which balance the need to reduce interpolation error while focussing on regions of high likelihood. We show that smaller emulator error from the Bayesian optimisation propagates to smaller widths on the posterior distribution. Even with fewer simulations than a Latin hypercube, Bayesian optimisation shrinks the 95% credible volume by 90% and, e.g., the 1 sigma error on the amplitude of small-scale primordial fluctuations by 38%. This is the first demonstration of Bayesian optimisation applied to large-scale structure emulation, and we anticipate the technique will generalise to many other probes such as galaxy clustering, weak lensing and 21cm.

Face anti-spoofing (FAS) is an essential mechanism for safeguarding the integrity of automated face recognition systems. Despite substantial advancements, the generalization of existing approaches to real-world applications remains challenging. This limitation can be attributed to the scarcity and lack of diversity in publicly available FAS datasets, which often leads to overfitting during training or saturation during testing. In terms of quantity, the number of spoof subjects is a critical determinant. Most datasets comprise fewer than 2,000 subjects. With regard to diversity, the majority of datasets consist of spoof samples collected in controlled environments using repetitive, mechanical processes. This data collection methodology results in homogenized samples and a dearth of scenario diversity. To address these shortcomings, we introduce the Wild Face Anti-Spoofing (WFAS) dataset, a large-scale, diverse FAS dataset collected in unconstrained settings. Our dataset encompasses 853,729 images of 321,751 spoof subjects and 529,571 images of 148,169 live subjects, representing a substantial increase in quantity. Moreover, our dataset incorporates spoof data obtained from the internet, spanning a wide array of scenarios and various commercial sensors, including 17 presentation attacks (PAs) that encompass both 2D and 3D forms. This novel data collection strategy markedly enhances FAS data diversity. Leveraging the WFAS dataset and Protocol 1 (Known-Type), we host the Wild Face Anti-Spoofing Challenge at the CVPR2023 workshop. Additionally, we meticulously evaluate representative methods using Protocol 1 and Protocol 2 (Unknown-Type). Through an in-depth examination of the challenge outcomes and benchmark baselines, we provide insightful analyses and propose potential avenues for future research. The dataset is released under Insightface.

Diffusion models represent a new paradigm in text-to-image generation. Beyond generating high-quality images from text prompts, models such as Stable Diffusion have been successfully extended to the joint generation of semantic segmentation pseudo-masks. However, current extensions primarily rely on extracting attentions linked to prompt words used for image synthesis. This approach limits the generation of segmentation masks derived from word tokens not contained in the text prompt. In this work, we introduce Open-Vocabulary Attention Maps (OVAM)-a training-free method for text-to-image diffusion models that enables the generation of attention maps for any word. In addition, we propose a lightweight optimization process based on OVAM for finding tokens that generate accurate attention maps for an object class with a single annotation. We evaluate these tokens within existing state-of-the-art Stable Diffusion extensions. The best-performing model improves its mIoU from 52.1 to 86.6 for the synthetic images' pseudo-masks, demonstrating that our optimized tokens are an efficient way to improve the performance of existing methods without architectural changes or retraining.

Turbulence is indispensable to redistribute nutrients for all life forms larger than microbial, on land and in the ocean. Yet, the development of deep-sea turbulence has not been studied in three dimensions (3D). As a disproportionate laboratory, an array of nearly 3000 high-resolution temperature sensors had been installed for three years on the flat 2500-m deep bottom of the Mediterranean Sea. The time series from the half-cubic hectometer 3D mooring-array allows for the creation of unique movies of deep-sea water motions. Although temperature differences are typically 0.001degrC, variable convection-turbulence is observed as expected from geothermal heating through the flat seafloor. During about 40% of the time, an additional turbulence, 3 times stronger in magnitude, is observed from slantwise advected warmer waters to pass in turbulent clouds. Besides turbulent clouds and seafloor heating, movies also reveal weakly turbulent interfacial-wave breakdown that commonly occurs in the open ocean far away from boundaries.

Machine unlearning methods have become increasingly important for selective concept removal in large pre-trained models. While recent work has explored unlearning in Euclidean contrastive vision-language models, the effectiveness of concept removal in hyperbolic spaces remains unexplored. This paper investigates machine unlearning in hyperbolic contrastive learning by adapting Alignment Calibration to MERU, a model that embeds images and text in hyperbolic space to better capture semantic hierarchies. Through systematic experiments and ablation studies, we demonstrate that hyperbolic geometry offers distinct advantages for concept removal, achieving near perfect forgetting with reasonable performance on retained concepts, particularly when scaling to multiple concept removal. Our approach introduces hyperbolic-specific components including entailment calibration and norm regularization that leverage the unique properties of hyperbolic space. Comparative analysis with Euclidean models reveals fundamental differences in unlearning dynamics, with hyperbolic unlearning reorganizing the semantic hierarchy while Euclidean approaches merely disconnect cross-modal associations. These findings not only advance machine unlearning techniques but also provide insights into the geometric properties that influence concept representation and removal in multimodal models. Source code available at this https URL

Physics-Informed Neural Networks (PINNs) are a novel computational approach for solving partial differential equations (PDEs) with noisy and sparse initial and boundary data. Although, efficient quantification of epistemic and aleatoric uncertainties in big multi-scale problems remains challenging. We propose \$PINN a novel method of computing global uncertainty in PDEs using a Bayesian framework, by combining local Bayesian Physics-Informed Neural Networks (BPINN) with domain decomposition. The solution continuity across subdomains is obtained by imposing the flux continuity across the interface of neighboring subdomains. To demonstrate the effectiveness of \$PINN, we conduct a series of computational experiments on PDEs in 1D and 2D spatial domains. Although we have adopted conservative PINNs (cPINNs), the method can be seamlessly extended to other domain decomposition techniques. The results infer that the proposed method recovers the global uncertainty by computing the local uncertainty exactly more efficiently as the uncertainty in each subdomain can be computed concurrently. The robustness of \$PINN is verified by adding uncorrelated random noise to the training data up to 15% and testing for different domain sizes.

Systemic drug administration often causes off-target effects limiting the efficacy of advanced therapies. Targeted drug delivery approaches increase local drug concentrations at the diseased site while minimizing systemic drug exposure. We present a magnetically guided microrobotic drug delivery system capable of precise navigation under physiological conditions. This platform integrates a clinical electromagnetic navigation system, a custom-designed release catheter, and a dissolvable capsule for accurate therapeutic delivery. In vitro tests showed precise navigation in human vasculature models, and in vivo experiments confirmed tracking under fluoroscopy and successful navigation in large animal models. The microrobot balances magnetic material concentration, contrast agent loading, and therapeutic drug capacity, enabling effective hosting of therapeutics despite the integration complexity of its components, offering a promising solution for precise targeted drug delivery.

Transfer learning is crucial for medical imaging, yet the selection of source datasets - which can impact the generalizability of algorithms, and thus patient outcomes - often relies on researchers' intuition rather than systematic principles. This study investigates these decisions through a task-based survey with machine learning practitioners. Unlike prior work that benchmarks models and experimental setups, we take a human-centered HCI perspective on how practitioners select source datasets. Our findings indicate that choices are task-dependent and influenced by community practices, dataset properties, and computational (data embedding), or perceived visual or semantic similarity. However, similarity ratings and expected performance are not always aligned, challenging a traditional "more similar is better" view. Participants often used ambiguous terminology, which suggests a need for clearer definitions and HCI tools to make them explicit and usable. By clarifying these heuristics, this work provides practical insights for more systematic source selection in transfer learning.