Search

There is growing symbiosis between artificial and biological intelligence sciences: neural principles inspire new intelligent machines, which are in turn used to advance our theoretical understanding of the brain. To promote further collaboration between biological and artificial intelligence researchers, we introduce the 2025 edition of the Algonauts Project challenge: How the Human Brain Makes Sense of Multimodal Movies (https://algonautsproject.com/). In collaboration with the Courtois Project on Neuronal Modelling (CNeuroMod), this edition aims to bring forth a new generation of brain encoding models that are multimodal and that generalize well beyond their training distribution, by training them on the largest dataset of fMRI responses to movie watching available to date. Open to all, the 2025 challenge provides transparent, directly comparable results through a public leaderboard that is updated automatically after each submission to facilitate rapid model assessment and guide development. The challenge will end with a session at the 2025 Cognitive Computational Neuroscience (CCN) conference that will feature winning models. We welcome researchers interested in collaborating with the Algonauts Project by contributing ideas and datasets for future challenges.

We analyse the large time behaviour of the rate function that describes the probability of large fluctuations of an underlying microscopic model associated to the homogeneous Boltzmann equation, such as the Kac walk. We consider in particular the asymptotic of the number of collisions, per particle and per unit of time, and show it exhibits a phase transition in the joint limit in which the number of particles N and the time interval [0,T] diverge. More precisely, due to the existence of Lu-Wennberg solutions, the corresponding limiting rate function vanishes for subtypical values of the number of collisions. We also analyse the second order large deviations showing that the probability of subtypical fluctuations is exponentially small in N, independently on T. As a key point, we establish the controllability of the homogeneous Boltzmann equation.

Conformal higher-spin gravity is the log-divergent part of the effective action of the scalar field coupled to background fields via higher-spin currents, as was defined by Segal and Tseytlin, which can be worked out over the flat space background. We revisit the problem of the scalar field in a higher-spin background and propose a manifestly covariant version thereof. The construction utilizes the Fedosov quantization of the cotangent bundle and the action is written with the help of the trace on a curved phase space that is provided by the Feigin--Felder--Shoikhet cocycle. The same construction allows one to formulate quantum mechanics on a curved space, the phase space being the cotangent bundle.

The Kraichnan-Leith-Batchelor (KLB) inverse energy cascade is a hallmark of 2D turbulence, with its theoretical energy spectrum observed in both Direct Numerical Simulations (DNS) and laboratory experiments. Surprisingly, however, we show that the effective diffusivity of 2D turbulent flows significantly departs from the KLB scaling prediction. We illustrate this phenomenon based on a suite of DNS of 2D turbulent flows forced at intermediate wavenumber and damped by weak linear or quadratic drag. We derive alternate scaling predictions based on the emergence of intense, isolated vortices causing spatially intermittent frictional dissipation localized within the small vortex cores. The predictions quantitatively match DNS data. This study points to a universal large-scale organization of 2D turbulent flows in physical space, bridging standard 2D Navier-Stokes turbulence with large-scale geophysical turbulence., Comment: 6 pages, 3 figures

Airborne Wind Energy (AWE) is an emerging technology designed to harness the power of high-altitude winds, offering a solution to several limitations of conventional wind turbines. AWE is based on flying devices (usually gliders or kites) that, tethered to a ground station and driven by the wind, convert its mechanical energy into electrical energy by means of a generator. Such systems are usually controlled by manoeuvering the kite so as to follow a predefined path prescribed by optimal control techniques, such as model-predictive control. These methods are strongly dependent on the specific model at use and difficult to generalize, especially in unpredictable conditions such as the turbulent atmospheric boundary layer. Our aim is to explore the possibility of replacing these techniques with an approach based on Reinforcement Learning (RL). Unlike traditional methods, RL does not require a predefined model, making it robust to variability and uncertainty. Our experimental results in complex simulated environments demonstrate that AWE agents trained with RL can effectively extract energy from turbulent flows, relying on minimal local information about the kite orientation and speed relative to the wind.

To 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., Comment: Collection of summaries of twelve topical panel discussions and individual thoughts of speakers and panelists, Nordita Scientific Program "Quantum Gravity: from gravitational EFTs to UV complete approaches". 62 pages + references, no figures

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., Comment: Lecture notes PhD school "Towards Quantum Gravity", Nordita Scientific Program "Quantum Gravity: from gravitational EFTs to UV complete approaches". 265 pages + references, 37 figures

Recent advances in multimodal training have significantly improved the integration of image understanding and generation within a unified model. This study investigates how vision-language models (VLMs) handle image-understanding tasks, specifically focusing on how visual information is processed and transferred to the textual domain. We compare VLMs that generate both images and text with those that output only text, highlighting key differences in information flow. We find that in models with multimodal outputs, image and text embeddings are more separated within the residual stream. Additionally, models vary in how information is exchanged from visual to textual tokens. VLMs that only output text exhibit a distributed communication pattern, where information is exchanged through multiple image tokens. In contrast, models trained for image and text generation rely on a single token that acts as a narrow gate for the visual information. We demonstrate that ablating this single token significantly deteriorates performance on image understanding tasks. Furthermore, modifying this token enables effective steering of the image semantics, showing that targeted, local interventions can reliably control the model's global behavior.

It appears to be difficult within string theory to obtain genuine scale separation between spacetime and the internal sector. In this paper, we propose a novel mechanism for scale-separated vacua which hinges on stringy effects that are invisible at the level of effective field theory. We show that (meta)stable vacua can form if a super no-scale one-loop potential combines with generic two-loop contributions to the vacuum energy, in a manner analogous to Banks-Zaks fixed points. Weak string coupling and scale separation arise from the accidental smallness of the one-loop term, which receives contributions only from massive states, relative to the two-loop term. We provide a proof of concept of this mechanism in explicit non-supersymmetric heterotic toroidal orbifolds by balancing the complete one-loop contribution against the estimated two-loop term and numerically minimizing the resulting effective potential in a restricted sector of moduli space. We note that this mechanism is generically possible within any fundamental theory that has a one-loop energy that gets contributions only from massive modes., Comment: 25 pages, 2 figures

Diffusion models (DMs) generate remarkable high quality images via the stochastic denoising process, which unfortunately incurs high sampling time. Post-quantizing the trained diffusion models in fixed bit-widths, e.g., 4 bits on weights and 8 bits on activation, is shown effective in accelerating sampling time while maintaining the image quality. Motivated by the observation that the cross-layer dependency of DMs vary across layers and sampling steps, we propose a mixed precision quantization scheme, MPQ-Diff, which allocates different bit-width to the weights and activation of the layers. We advocate to use the cross-layer correlation of a given layer, termed network orthogonality metric, as a proxy to measure the relative importance of a layer per sampling step. We further adopt a uniform sampling scheme to avoid the excessive profiling overhead of estimating orthogonality across all time steps. We evaluate the proposed mixed-precision on LSUN and ImageNet, showing a significant improvement in FID from 65.73 to 15.39, and 52.66 to 14.93, compared to their fixed precision quantization, respectively.

Common functional languages incentivize tail-recursive functions, as opposed to general recursive functions that consume stack space and may not scale to large inputs. This distinction occasionally requires writing functions in a tail-recursive style that may be more complex and slower than the natural, non-tail-recursive definition. This work describes our implementation of the *tail modulo constructor* (TMC) transformation in the OCaml compiler, an optimization that provides stack-efficiency for a larger class of functions -- tail-recursive *modulo constructors* -- which includes in particular the natural definition of `List.map` and many similar recursive data-constructing functions. We prove the correctness of this program transformation in a simplified setting -- a small untyped calculus -- that captures the salient aspects of the OCaml implementation. Our proof is mechanized in the Coq proof assistant, using the Iris base logic. An independent contribution of our work is an extension of the Simuliris approach to define simulation relations that support different calling conventions. To our knowledge, this is the first use of Simuliris to prove the correctness of a compiler transformation., Comment: Published at POPL 2025

Parallel assemblies of slender structures forming brushes are common in our daily life from sweepers to pastry brushes and paintbrushes. This type of porous objects can easily trap liquid in their interstices when removed from a liquid bath. This property is exploited to transport liquids in many applications ranging from painting, dip-coating, brush-coating to the capture of nectar by bees, bats and honeyeaters. Rationalizing the viscous entrainment flow beyond simple scaling laws is complex due to its multiscale structure and the multidirectional flow. Here, we provide an analytical model, together with precision experiments with ideal rigid brushes, to fully characterize the flow through this anisotropic porous medium as it is withdrawn from a liquid bath. We show that the amount of liquid entrained by a brush varies non-monotonically during the withdrawal at low speed, is highly sensitive to the different parameters at play and is very well described by the model without any fitting parameter. Finally, an optimal brush geometry maximizing the amount of liquid captured at a given retraction speed is derived from the model and experimentally validated. These optimal designs open routes towards efficient liquid manipulating devices.

In this paper, we study the thermodynamic cost associated with erasing a static random access memory. By combining the stochastic thermodynamics framework of electronic circuits with machine learning-based optimization techniques, we show that it is possible to erase an electronic random access memory at arbitrarily fast speed and finite heat dissipation. This disproves a widely held belief that heat dissipation scales linearly with erasure speed. Furthermore, we find driving protocols that minimize the heat dissipation, leading to explicit design principles for future computer memories. This bridges an important gap between the theoretical framework of stochastic thermodynamics and applications in electronic engineering.

When examined through the lens of their residual streams, a puzzling property emerges in transformer networks: residual contributions (e.g., attention heads) sometimes specialize in specific tasks or input attributes. In this paper, we analyze this phenomenon in vision transformers, focusing on the spectral geometry of residuals, and explore its implications for modality alignment in vision-language models. First, we link it to the intrinsically low-dimensional structure of visual head representations, zooming into their principal components and showing that they encode specialized roles across a wide variety of input data distributions. Then, we analyze the effect of head specialization in multimodal models, focusing on how improved alignment between text and specialized heads impacts zero-shot classification performance. This specialization-performance link consistently holds across diverse pre-training data, network sizes, and objectives, demonstrating a powerful new mechanism for boosting zero-shot classification through targeted alignment. Ultimately, we translate these insights into actionable terms by introducing ResiDual, a technique for spectral alignment of the residual stream. Much like panning for gold, it lets the noise from irrelevant unit principal components (i.e., attributes) wash away to amplify task-relevant ones. Remarkably, this dual perspective on modality alignment yields fine-tuning level performances on different data distributions while modeling an extremely interpretable and parameter-efficient transformation, as we extensively show on more than 50 (pre-trained network, dataset) pairs.

Energy functionals produced by second-order homogenization of periodic elastic structures commonly feature negative gradient moduli. We show that this undesirable property is caused by the truncation of the energy expansion in powers of the small scale separation parameter. By revisiting Cholesky's LDLT decomposition, we propose an alternative truncation method that restores positivity while preserving the order of accuracy. We illustrate this method on a variety of periodic structures, both continuous and discrete, and derive compact analytical expressions of the homogenized energy that are positive and accurate to second order. The method can also cure the energy functionals produced by second-order dimension reduction, which suffer similar non-positivity issues. It naturally extends beyond second order.

The empirical wavelet transform is an adaptive multiresolution analysis tool based on the idea of building filters on a data-driven partition of the Fourier domain. However, existing 2D extensions are constrained by the shape of the detected partitioning. In this paper, we provide theoretical results that permits us to build 2D empirical wavelet filters based on an arbitrary partitioning of the frequency domain. We also propose an algorithm to detect such partitioning from an image spectrum by combining a scale-space representation to estimate the position of dominant harmonic modes and a watershed transform to find the boundaries of the different supports making the expected partition. This whole process allows us to define the empirical watershed wavelet transform. We illustrate the effectiveness and the advantages of such adaptive transform, first visually on toy images, and next on both unsupervised texture segmentation and image deconvolution applications.

Motivated by the question of whether Chow polynomials of matroids have only real roots, this article revisits the known relationship between Eulerian polynomials and the Hilbert series of Chow rings of permutohedral varieties. This is done using a quadratic Gr\"obner basis associated to a new presentation of those rings, which is obtained by iterating the semi-small decomposition of Chow rings of matroids. This Gr\"obner basis can also be applied to compute certain principal ideals in these rings, and ultimately reestablish the known connection between derangement polynomials and the Hilbert series of Chow rings for corank 1 uniform matroids. More broadly, this approach enables us to express the Hilbert series of Chow rings for any uniform matroid as polynomials related to the ascent statistics on particular sets of inversion sequences.

For the High-Luminosity (HL-LHC) phase, the upgrade of the Compact Muon Solenoid (CMS) experiment at CERN will include a novel MIP Timing Detector (MTD). The central part of MTD, the barrel timing layer (BTL), is designed to provide a measurement of the time of arrival of charged particles with a precision of 30 ps at the beginning of HL-LHC, progressively degrading to 60 ps while operating in an extremely harsh radiation environment for over a decade. In this paper we present a comparative analysis of the time resolution of BTL module prototypes made of LYSO:Ce crystal bars read out by silicon photo-multipliers (SiPMs). The timing performance measured in beam test campaigns is presented for prototypes with different construction and operation parameters, such as different SiPM cell sizes (15, 20, 25 and 30 $\rm \mu m$), SiPM manufacturers and crystal bar thicknesses. The evolution of time resolution as a function of the irradiation level has been studied using non-irradiated SiPMs as well as SiPMs exposed up to $2\times 10^{14}~n_{eq}/cm^2$ fluence. The key parameters defining the module time resolution such as SiPM characteristics (gain, photon detection efficiency, radiation induced dark count rate) and crystal properties (light output and dimensions) are discussed. These results have informed the final choice of the MTD barrel sensor configuration and offer a unique starting point for the design of future large-area scintillator-based timing detectors in either low or high radiation environments.

As large language models (LLMs) become increasingly integrated into society, their alignment with human morals is crucial. To better understand this alignment, we created a large corpus of human- and LLM-generated responses to various moral scenarios. We found a misalignment between human and LLM moral assessments; although both LLMs and humans tended to reject morally complex utilitarian dilemmas, LLMs were more sensitive to personal framing. We then conducted a quantitative user study involving 230 participants (N=230), who evaluated these responses by determining whether they were AI-generated and assessed their agreement with the responses. Human evaluators preferred LLMs' assessments in moral scenarios, though a systematic anti-AI bias was observed: participants were less likely to agree with judgments they believed to be machine-generated. Statistical and NLP-based analyses revealed subtle linguistic differences in responses, influencing detection and agreement. Overall, our findings highlight the complexities of human-AI perception in morally charged decision-making.

We investigate the slow, second order motion of thin flexible floating strips drifting in surface gravity waves. We introduce a diffractionless model (Froude-Krylov approximation) that neglects viscosity, surface tension, and radiation effects. This model predicts a mean yaw moment that favors a longitudinal orientation of the strip, along the direction of wave propagation, and a small reduction in the Stokes drift velocity. Laboratory experiments with thin rectangular strips of polypropylene show a systematic rotation of the strips towards the longitudinal position, in good agreement with our model. We finally observe that the mean angular velocity towards the stable longitudinal orientation decreases as the strip length increases, an effect likely due to dissipation, which is not accounted for in our inviscid model.

In the contemporary business landscape, collaboration across multiple organizations offers a multitude of opportunities, including reduced operational costs, enhanced performance, and accelerated technological advancement. The application of process mining techniques in an inter-organizational setting, exploiting the recorded process event data, enables the coordination of joint effort and allows for a deeper understanding of the business. Nevertheless, considerable concerns pertaining to data confidentiality emerge, as organizations frequently demonstrate a reluctance to expose sensitive data demanded for process mining, due to concerns related to privacy and security risks. The presence of conflicting interests among the parties involved can impede the practice of open data sharing. To address these challenges, we propose our approach and toolset named CONFINE, which we developed with the intent of enabling process mining on process event data from multiple providers while preserving the confidentiality and integrity of the original records. To ensure that the presented interaction protocol steps are secure and that the processed information is hidden from both involved and external actors, our approach is based on a decentralized architecture and consists of trusted applications running in Trusted Execution Environments (TEE). In this demo paper, we provide an overview of the core components and functionalities as well as the specific details of its application.

Massless chiral fields of arbitrary spin in six spacetime dimensions, also known as higher spin singletons, admit a simple formulation in terms of $SL(2,\mathbb{H})$ tensors. We show that, paralleling the four-dimensional case, these fields can be described using a $0$-form and a gauge $2$-form, taking values in totally symmetric tensors of $SL(2,\mathbb{H})$. We then exhibit an example of interacting theory that couples a tower of singletons of all integer spin to a background of $\mathfrak{g}$-valued higher spin fields, for $\mathfrak{g}$ an arbitrary Lie algebra equipped with an invariant symmetric bilinear form. Finally, we discuss the formulation of these models in arbitrary even dimensions, as well as their partially-massless counterpart.

We motivate a relation between dark energy and the scale of new physics in weakly coupled string theory. This mixing between infrared and ultraviolet physics leads to a unique corner for real-world phenomenology: barring fine-tunings, we are naturally led to the ``dark dimension'' scenario, a single mesoscopic extra dimension of micron size with the standard model localized on D-branes. Our explicit top-down worldsheet derivation establishes it on a more solid grounding. Allowing some fine-tuning, such that the vacuum energy only arise at higher orders in string perturbation theory, the ``little string theory'' scenario with a very weakly coupled string is an alternative possibility. In this case, the string scale lies at the edge of detectability of particle accelerators., Comment: 6 pages, 1 figure

Over the last decades, theoretical photochemistry has produced multiple techniques to simulate the nonadiabatic dynamics of molecules. Surprisingly, much less effort has been devoted to adequately describing the first step of a photochemical or photophysical process: photoexcitation. Here, we propose a formalism to include the effect of a laser pulse in trajectory-based nonadiabatic dynamics at the level of the initial conditions, with no additional cost. The promoted density approach (PDA) decouples the excitation from the nonadiabatic dynamics by defining a new set of initial conditions, which include an excitation time. PDA with surface hopping leads to nonadiabatic dynamics simulations in excellent agreement with quantum dynamics using an explicit laser pulse and highlights the strong impact of a laser pulse on the resulting photodynamics and the limits of the (sudden) vertical excitation. Combining PDA with trajectory-based nonadiabatic methods is possible for any arbitrary-sized molecules using a code provided in this work., Comment: SI included in the pdf. Link to PDA/PDAW code 'promdens' (and guidelines) in the SI

Humans naturally build mental models of object interactions and dynamics, allowing them to imagine how their surroundings will change if they take a certain action. While generative models today have shown impressive results on generating/editing images unconditionally or conditioned on text, current methods do not provide the ability to perform object manipulation conditioned on actions, an important tool for world modeling and action planning. Therefore, we propose to learn an action-conditional generative models by learning from unlabeled videos of human hands interacting with objects. The vast quantity of such data on the internet allows for efficient scaling which can enable high-performing action-conditional models. Given an image, and the shape/location of a desired hand interaction, CosHand, synthesizes an image of a future after the interaction has occurred. Experiments show that the resulting model can predict the effects of hand-object interactions well, with strong generalization particularly to translation, stretching, and squeezing interactions of unseen objects in unseen environments. Further, CosHand can be sampled many times to predict multiple possible effects, modeling the uncertainty of forces in the interaction/environment. Finally, method generalizes to different embodiments, including non-human hands, i.e. robot hands, suggesting that generative video models can be powerful models for robotics.

Characterizing the photolysis processes undergone by transient volatile organic compounds in the troposphere requires the knowledge of their photoabsorption cross-section - quantities often challenging to determine experimentally, particularly due to the reactivity of these molecules. We present a computational tool coined AtmoSpec, which can predict a quantitative photoabsorption cross-section for volatile organic compounds by using computational photochemistry. The user enters the molecule of interest as a SMILES code and, after selecting a level of theory for the electronic structure (and waiting for the calculations to take place), is presented with a photoabsorption cross-section for the low-energy conformers and an estimate of the photolysis rate coefficient for different standardized actinic fluxes. More specifically, AtmoSpec is an automated workflow for the nuclear ensemble approach, an efficient technique to approximate the absolute intensities and excitation wavelengths of a photoabsorption cross-section for a molecule in the gas phase. This work provides background information on the nuclear ensemble approach, a guided example of a typical AtmoSpec calculation, details about the architecture of the code, and the current limitations and future developments of this tool.

We propose a realisation of partially-massless higher spin algebras in four dimensions in terms of bosonic and fermionic oscillators, using Howe duality between $sp(4,\mathbb R) \cong so(2,3)$ and $osp(1|2(\ell-1), \mathbb R)$. More precisely, we show that the centraliser of $osp(1|2(\ell-1),\mathbb R)$ in the Weyl--Clifford algebra generated by $4$ bosonic and $8(\ell-1)$ fermionic symbols, modulo $osp(1|2(\ell-1),\mathbb R)$ generators, is isomorphic to the higher spin algebra of the type-A$_\ell$ theory whose spectrum contains partially-massless fields of all spins and depths $t=1,3,\dots,2\ell-1$. We also discuss the possible existence of a deformation of this algebra, which would encode interaction for the type-A$_\ell$ theory.

We show that, given a continuous action $\alpha$ of a locally compact group $G$ on a factor $M$, the relative commutant $M'\cap(M\rtimes_{\alpha} G)$ is contained in $M\rtimes_{\alpha} H$ where $H$ is the subgroup of elements acting without spectral gap. As a corollary, we answer a question of Marrakchi and Vaes by showing that if $M$ is semifinite and $\alpha_g$ is not approximately inner for all $g\neq 1$, then $M'\cap (M\rtimes_{\alpha} G)=\mathbb{C}$., Comment: 12 pages

Rapid advancements in machine learning (ML) are transforming materials science by significantly speeding up material property calculations. However, the proliferation of ML approaches has made it challenging for scientists to keep up with the most promising techniques. This paper presents an empirical study on Geometric Graph Neural Networks for 3D atomic systems, focusing on the impact of different (1) canonicalization methods, (2) graph creation strategies, and (3) auxiliary tasks, on performance, scalability and symmetry enforcement. Our findings and insights aim to guide researchers in selecting optimal modeling components for molecular modeling tasks.

With the recent addition of Retrieval-Augmented Generation (RAG), the scope and importance of Information Retrieval (IR) has expanded. As a result, the importance of a deeper understanding of IR models also increases. However, interpretability in IR remains under-explored, especially when it comes to the models' inner mechanisms. In this paper, we explore the possibility of adapting Integrated Gradient-based methods in an IR context to identify the role of individual neurons within the model. In particular, we provide new insights into the role of what we call "relevance" neurons, as well as how they deal with unseen data. Finally, we carry out an in-depth pruning study to validate our findings., Comment: Accepted at ICTIR 2024

To gain insight into the mechanisms behind machine learning methods, it is crucial to establish connections among the features describing data points. However, these correlations often exhibit a high-dimensional and strongly nonlinear nature, which makes them challenging to detect using standard methods. This paper exploits the entanglement between intrinsic dimensionality and correlation to propose a metric that quantifies the (potentially nonlinear) correlation between high-dimensional manifolds. We first validate our method on synthetic data in controlled environments, showcasing its advantages and drawbacks compared to existing techniques. Subsequently, we extend our analysis to large-scale applications in neural network representations. Specifically, we focus on latent representations of multimodal data, uncovering clear correlations between paired visual and textual embeddings, whereas existing methods struggle significantly in detecting similarity. Our results indicate the presence of highly nonlinear correlation patterns between latent manifolds.

Aligning data from different domains is a fundamental problem in machine learning with broad applications across very different areas, most notably aligning experimental readouts in single-cell multiomics. Mathematically, this problem can be formulated as the minimization of disagreement of pair-wise quantities such as distances and is related to the Gromov-Hausdorff and Gromov-Wasserstein distances. Computationally, it is a quadratic assignment problem (QAP) that is known to be NP-hard. Prior works attempted to solve the QAP directly with entropic or low-rank regularization on the permutation, which is computationally tractable only for modestly-sized inputs, and encode only limited inductive bias related to the domains being aligned. We consider the alignment of metric structures formulated as a discrete Gromov-Wasserstein problem and instead of solving the QAP directly, we propose to learn a related well-scalable linear assignment problem (LAP) whose solution is also a minimizer of the QAP. We also show a flexible extension of the proposed framework to general non-metric dissimilarities through differentiable ranks. We extensively evaluate our approach on synthetic and real datasets from single-cell multiomics and neural latent spaces, achieving state-of-the-art performance while being conceptually and computationally simple.

We propose a covariant method of constructing entire trajectories of physical states in superstring theory in the critical dimension. It is inspired by a recently developed covariant technology of excavating bosonic string trajectories, that is facilitated by the observation that the Virasoro constraints can be written as linear combinations of lowering operators of a bigger algebra, namely a symplectic algebra, which is Howe dual to the spacetime Lorentz algebra. For superstrings, it is the orthosymplectic algebra that appears instead, with its lowest weight states forming the simplest class of physical trajectories in the NS sector. To construct the simplest class in the R sector, the lowest weight states need to be supplemented with other states, which we determine. Deeper trajectories are then constructed by acting with suitable combinations of the raising operators of the orthosymplectic algebra, which we illustrate with several examples., Comment: minor changes, matches published version

Accurate reconstruction of complex dynamic scenes from just a single viewpoint continues to be a challenging task in computer vision. Current dynamic novel view synthesis methods typically require videos from many different camera viewpoints, necessitating careful recording setups, and significantly restricting their utility in the wild as well as in terms of embodied AI applications. In this paper, we propose $\textbf{GCD}$, a controllable monocular dynamic view synthesis pipeline that leverages large-scale diffusion priors to, given a video of any scene, generate a synchronous video from any other chosen perspective, conditioned on a set of relative camera pose parameters. Our model does not require depth as input, and does not explicitly model 3D scene geometry, instead performing end-to-end video-to-video translation in order to achieve its goal efficiently. Despite being trained on synthetic multi-view video data only, zero-shot real-world generalization experiments show promising results in multiple domains, including robotics, object permanence, and driving environments. We believe our framework can potentially unlock powerful applications in rich dynamic scene understanding, perception for robotics, and interactive 3D video viewing experiences for virtual reality., Comment: Accepted to ECCV 2024. Project webpage is available at: https://gcd.cs.columbia.edu/

In this paper, we present the major results from the Covid Radiographic imaging System based on AI (Co.R.S.A.) project, which took place in Italy. This project aims to develop a state-of-the-art AI-based system for diagnosing Covid-19 pneumonia from Chest X-ray (CXR) images. The contributions of this work are manyfold: the release of the public CORDA dataset, a deep learning pipeline for Covid-19 detection, and the clinical validation of the developed solution by expert radiologists. The proposed detection model is based on a two-step approach that, paired with state-of-the-art debiasing, provides reliable results. Most importantly, our investigation includes the actual usage of the diagnosis aid tool by radiologists, allowing us to assess the real benefits in terms of accuracy and time efficiency. Project homepage: https://corsa.di.unito.it/, Comment: Accepted at 21st IEEE International Symposium on Biomedical Imaging (ISBI)

In the pursuit of advancing natural language processing for the Italian language, we introduce a state-of-the-art Large Language Model (LLM) based on the novel Meta LLaMA-3 model: LLaMAntino-3-ANITA-8B-Inst-DPO-ITA. We fine-tuned the original 8B parameters instruction tuned model using the Supervised Fine-tuning (SFT) technique on the English and Italian language datasets in order to improve the original performance. Consequently, a Dynamic Preference Optimization (DPO) process has been used to align preferences, avoid dangerous and inappropriate answers, and limit biases and prejudices. Our model leverages the efficiency of QLoRA to fine-tune the model on a smaller portion of the original model weights and then adapt the model specifically for the Italian linguistic structure, achieving significant improvements in both performance and computational efficiency. Concurrently, DPO is employed to refine the model's output, ensuring that generated content aligns with quality answers. The synergy between SFT, QLoRA's parameter efficiency and DPO's user-centric optimization results in a robust LLM that excels in a variety of tasks, including but not limited to text completion, zero-shot classification, and contextual understanding. The model has been extensively evaluated over standard benchmarks for the Italian and English languages, showing outstanding results. The model is freely available over the HuggingFace hub and, examples of use can be found in our GitHub repository. https://huggingface.co/swap-uniba/LLaMAntino-3-ANITA-8B-Inst-DPO-ITA

We study infinite-distance limits in the moduli space of perturbative string vacua. The remarkable interplay of string dualities seems to determine a highly non-trivial dichotomy, summarized by the emergent string conjecture, by which in some duality frame either internal dimensions decompactify or a unique critical string becomes tensionless. We investigate whether this pattern persists in potentially non-geometric settings, showing that (a proxy for) the cutoff of the gravitational effective field theory in perturbative type II vacua extracted from a graviton scattering amplitude vanishes if and only if a light tower of states appears. Moreover, under some technical assumptions on the spectrum of conformal weights, the cutoff scales with the spectral gap of the internal conformal field theory in the same manner as in decompactification or emergent string limits, regardless of supersymmetry or whether the internal sector is geometric. As a byproduct, we elucidate the role of the species scale in (de)compactifications and show compatibility between effective field theory and worldsheet approaches in geometric settings with curvature., Comment: 41 pages. v2: matches published version

Enforcing security requirements in networked information systems relies on security controls to mitigate the risks from increasingly dangerous threats. Configuring security controls is challenging; even nowadays, administrators must perform it without adequate tool support. Hence, this process is plagued by errors that translate to insecure postures, security incidents, and a lack of promptness in answering threats. This paper presents the Security Capability Model (SCM), a formal model that abstracts the features that security controls offer for enforcing security policies, which includes an Information Model that depicts the basic concepts related to rules (i.e., conditions, actions, events) and policies (i.e., conditions' evaluation, resolution strategies, default actions), and a Data Model that covers the capabilities needed to describe different types of filtering and channel protection controls. Following state-of-the-art design patterns, the model allows for generating abstract versions of the security controls' languages and a model-driven approach for translating abstract policies into device-specific configuration settings. By validating its effectiveness in real-world scenarios, we show that SCM enables the automation of different and complex security tasks, i.e., accurate and granular security control comparison, policy refinement, and incident response. Lastly, we present opportunities for extensions and integration with other frameworks and models.

While heat transport by baroclinic turbulence in oceans and planetary atmospheres is well described by a two-layer model, the relative depth of the two layers varies greatly depending on the situation of interest, making it an important parameter governing the transport properties of the system. Focusing on the low-drag turbulent regime, we extend the vortex-gas scaling theory to address the case of arbitrary layer depths. To wit, we map the arbitrary-layer-depth system onto an equivalent equal-depth system with rescaled parameters, establishing the asymptotic validity of the mapping for weak bottom drag. This approach leads to quantitative predictions for the turbulent transport by two-layer baroclinic turbulence with arbitrary layer depths, without additional free parameters. We validate these predictions using an extended suite of numerical simulations with either linear or quadratic bottom drag.