arXiv subjects

Oliver Hahn

Publications and source records attributed to Oliver Hahn.

At least 19 recordsLinked to original sources

arXiv2026

Contrastive Energy Fields for Inference-Time Procedure Planning in Instructional Videos

Procedure planning seeks to estimate a sequence of actions to transition from an observed initial state to a given goal state. Current procedure planning approaches directly predict action sequences from latent representations using feed-forward neural networks or diffusion-based inference. These paradigms treat every action as plausible, lacking the ability to enforce task-specific logical constraints that render certain actions irrelevant or not plausible. We propose CEFITO, a procedure planning approach that learns a predictor to express an action-conditioned representation space. Based on this representation space, we formulate procedure planning as a task-constrained optimization problem. Unlike prior methods, CEFITO explicitly reasons over the action space by omitting irrelevant actions during inference-time planning. This reformulation enables effective procedure planning and achieves state-of-the-art accuracy on two established procedure planning benchmarks.

cs.CV↗

arXiv2026

Bridging perturbation theory and simulations: initial conditions and fast integrators for cosmological simulations

These lecture notes provide an introduction to the generation of initial conditions for cosmological N-body simulations. Starting from the definition and properties of Gaussian random fields, we discuss their role in cosmology and the efficient generation of such fields using Fourier methods. The Vlasov-Poisson system is introduced as the governing framework for cold collisionless matter, and its solution via characteristics and Lagrangian perturbation theory (LPT) is detailed. We discuss the use of LPT for initializing N-body simulations, emphasizing the importance of high-order LPT and late-time starts to minimize truncation and discreteness errors. Finally, we discuss time integration schemes, including PT-informed integrators, and their role in accurately evolving the system. These notes aim to bridge the gap between theoretical perturbation methods and practical simulation techniques.

astro-ph.CO↗

arXiv2026

Decaying Dark Matter Halo Abundance from a Revised Spherical Collapse Model

We present a semi-analytical framework for the halo mass function (HMF) in decaying dark matter (DDM) cosmologies, in which dark matter decays into a massive daughter particle inheriting a velocity kick $v_k$ and a massless dark radiation component. Building on the Press-Schechter formalism, we encode the DDM physics through a spherical collapse model that explicitly tracks the decay-induced mass loss, yielding a modified, mass-dependent critical collapse threshold $\delta_c(M_0)$ and a mapping $M_{\rm coll}(M_0)$ between the initial Lagrangian mass and the collapsed halo mass. The critical threshold exhibits a characteristic transition between two analytically tractable plateaus: a large-mass limit, where all daughter particles are retained by the halo, and a small-mass limit, where all daughters escape and the collapse is equivalent to that of a dark matter species decaying entirely into dark radiation, making $\delta_c$ independent of $M_0$ and $v_k$. We provide semi-analytical results and fits for both limits and a fitting formula for the transition, whose single free parameter $M_1 \propto v_k^3\,\tilde\Gamma^{-1/2} t_{\rm ta}$ has a transparent physical interpretation: it is the mass scale at which the kick velocity equals the halo orbital velocity. We validate our predictions against a suite of N-body simulations at $z=0$ and $z\approx 1$, finding good agreement across models spanning mild to strong HMF suppression relative to $\Lambda$CDM. Residual deviations for the largest kick velocities at $z=0$ are observed. Via a halo-by-halo comparison between simulations, we trace the discrepancy to the definition of the halo mass when daughter orbits extend beyond the halo boundary. The resulting fitting functions for $\delta_c(M_0,\Gamma,v_k)$ and $M_{\rm coll}(M_0)$ provide an efficient and accurate route to DDM constraints from current and forthcoming probes of the halo mass function.

astro-ph.CO↗

arXiv2026

Scene-Centric Unsupervised Video Panoptic Segmentation

Video panoptic segmentation (VPS) aims to jointly detect, segment, and track all objects while partitioning the video into semantically consistent regions. We introduce the task setting of unsupervised VPS, omitting any human supervision. Existing unsupervised scene understanding works mainly focused on image segmentation tasks; the video domain remains underexplored. We propose VideoCUPS, the first unsupervised VPS approach. VideoCUPS generates temporally consistent panoptic video pseudo-labels from scene-centric videos by exploiting unsupervised depth, motion, and visual cues. Training on these pseudo-labels using a novel Video DropLoss yields an accurate, unsupervised VPS model. To benchmark progress, we introduce a comprehensive evaluation protocol and four competitive baselines, extending state-of-the-art unsupervised panoptic image and instance video segmentation models to VPS. VideoCUPS outperforms all baselines and demonstrates strong label-efficient learning. With VideoCUPS, our evaluation protocol, and baselines, we provide a strong foundation for future research on unsupervised VPS.

cs.CV↗

arXiv2026

JZ-Tree: GPU friendly neighbour search and friends-of-friends with dual tree walks in JAX plus CUDA

Algorithms based on spatial tree traversal are widely regarded as among the most efficient and flexible approaches for many problems in CPU-based high-performance computing (HPC). However, directly transferring these algorithms to GPU architectures often yields substantially smaller performance gains than expected in light of the high computational throughput of modern GPUs. The branching nature of tree algorithms leads to thread divergence and irregular memory access patterns -- both of which may severely limit GPU performance. To address these challenges, we propose a Morton (z-order) 'plane-based tree hierarchy' that is specifically designed for GPU architectures. The resulting flattened data layout enables efficient dual-tree traversal with collaborative execution across thread groups, leading to highly coalesced memory access patterns. Based on this framework we present implementations of two important spatial algorithms -- exact $k$-nearest neighbour search and friends-of-friends (FoF) clustering. For both cases, we observe more than an order-of-magnitude performance improvement over the closest competing GPU libraries for large problem sizes ($N \gtrsim 10^7$), together with strong scaling to distributed multi-GPU systems. We provide an open-source implementation, 'JZ-Tree' (JAX z-order tree), which serves as a foundation for efficient GPU implementations of a broad class of tree-based algorithms.

cs.DC↗

arXiv2026

Weak distillation of quantum resources

Importance sampling based on quasi-probability decomposition is the backbone of many widely used techniques, such as error mitigation, circuit knitting, and, more generally, virtual quantum resource distillation, as it allows one to simulate operations that are not accessible in a given setting. However, this class of protocols faces a fundamental problem -- it only allows to estimate expectation values. Here, we provide a general framework that lifts any quasi-probability-based protocol from expectation value estimation to a weak simulator, realizing sampling from the desired distribution only using a restricted class of quantum resources. Our method runs with the sampling cost proportional to the negativity of the quasi-probability, in stark contrast to the naive estimation-based approach that requires a large number of samples even in the case of small negativity. We show that our method requires significantly fewer samples in a number of relevant scenarios, such as error mitigation, entanglement distillation and magic state distillation. Our framework realizes the weak simulation of quantum resources without actually distilling the state, introducing a new notion of quantum resource distillation.

quant-ph↗

arXiv2025

DISCO-DJ II: a differentiable particle-mesh code for cosmology

The mildly non-linear regime of cosmic structure formation holds much of the information that upcoming large-scale structure surveys aim to exploit, making fast and accurate predictions on these scales essential. We present the $N$-body module of DISCO-DJ (DIfferentiable Simulations for COsmology - Done with Jax), designed to deliver high-fidelity, GPU-accelerated, and differentiable particle-mesh simulations tailored for cosmological inference. Theory-informed time integrators such as the recently introduced BullFrog method allow for accurate predictions already with few time steps (e.g. $6$ steps for per-cent-level accuracy in terms of the present-day power spectrum at $k \approx 0.2 \, h / \mathrm{Mpc}$ using $N = 512^3$ particles, which takes just a few seconds). To control discreteness effects and achieve high accuracy, the code incorporates a suite of advanced techniques, for example a custom non-uniform FFT implementation for force evaluation. Both forward- and reverse-mode differentiation are supported, with memory requirements independent of the number of time steps; in the reverse case, this is achieved through an adjoint formulation. We extensively study the effect of various numerical parameters on the accuracy. As an application of DISCO-DJ, we perform field-level inference by recovering $\sigma_8$ and the initial conditions from a noisy Gadget matter density field. Coupled with our recently introduced Einstein--Boltzmann solver, the DISCO-DJ ecosystem provides a self-consistent, fully differentiable pipeline for modelling the large-scale structure of the universe. The code is available at https://github.com/cosmo-sims/DISCO-DJ.

astro-ph.CO↗

arXiv2025

Motion-Refined DINOSAUR for Unsupervised Multi-Object Discovery

Unsupervised multi-object discovery (MOD) aims to detect and localize distinct object instances in visual scenes without any form of human supervision. Recent approaches leverage object-centric learning (OCL) and motion cues from video to identify individual objects. However, these approaches use supervision to generate pseudo labels to train the OCL model. We address this limitation with MR-DINOSAUR -- Motion-Refined DINOSAUR -- a minimalistic unsupervised approach that extends the self-supervised pre-trained OCL model, DINOSAUR, to the task of unsupervised multi-object discovery. We generate high-quality unsupervised pseudo labels by retrieving video frames without camera motion for which we perform motion segmentation of unsupervised optical flow. We refine DINOSAUR's slot representations using these pseudo labels and train a slot deactivation module to assign slots to foreground and background. Despite its conceptual simplicity, MR-DINOSAUR achieves strong multi-object discovery results on the TRI-PD and KITTI datasets, outperforming the previous state of the art despite being fully unsupervised.

cs.CV↗

arXiv2025

Measuring non-Gaussianity with Correlation

Quantum non-Gaussianity is a key resource for quantum advantage in continuous-variable systems. We introduce a general framework to quantify non-Gaussianity based on correlation generation: two copies of a state become correlated at a $50{:}50$ beam splitter if and only if the state is non-Gaussian, with correlations reducing to entanglement in the pure-state case. This connection enables operational measures of non-Gaussianity, defined through correlation quantifiers such as R\'enyi-$\alpha$ entropy for pure states and R\'enyi-$\alpha$ mutual information for mixed states. We prove that all such measures are monotonic under Gaussian channels. Building on this framework, we propose a sample-efficient experimental protocol to estimate non-Gaussianity using standard optical components, even in the state agnostic setting. Finally, we establish a lower bound on the sample complexity of estimating Wigner negativity, allowing a direct comparison with our protocol. Our results provide both a unifying theoretical framework for non-Gaussianity and a practical route toward its experimental quantification.

quant-ph↗

arXiv2025

Majorization theory for quasiprobabilities

The question of how to characterize disorder arises in a wide variety of settings, including information theory, economics, and quantum thermodynamics. The theory of majorization provides an elegant answer, playing a central role in these fields. However, the existing majorization framework is inapplicable to functions that both take negative values and are defined on infinite spaces. Yet such functions, in the form of quasiprobability distributions, are ubiquitous in fields such as quantum optics, signal analysis, and bosonic quantum computation. Here we develop a notion of majorization that is applicable to such functions, proving that it admits four equivalent characterizations that naturally reduce to the finite case, thereby generalizing a seminal theorem by Hardy, Littlewood, and P\'olya. Moreover, we extend this equivalence to the setting where majorization is considered relative to an arbitrary positive distribution. We give several applications of our results in the context of quantum resource theories. These include deriving families of resource monotones and constraining quantum state conversions. We analytically and numerically study examples using the Wigner and Husimi functions, which feature prominently in quantum optics. Our results provide a comprehensive majorization framework for assessing the disorder of integrable functions over infinite measure spaces.

quant-ph↗

arXiv2025

Feed-Forward SceneDINO for Unsupervised Semantic Scene Completion

Semantic scene completion (SSC) aims to infer both the 3D geometry and semantics of a scene from single images. In contrast to prior work on SSC that heavily relies on expensive ground-truth annotations, we approach SSC in an unsupervised setting. Our novel method, SceneDINO, adapts techniques from self-supervised representation learning and 2D unsupervised scene understanding to SSC. Our training exclusively utilizes multi-view consistency self-supervision without any form of semantic or geometric ground truth. Given a single input image, SceneDINO infers the 3D geometry and expressive 3D DINO features in a feed-forward manner. Through a novel 3D feature distillation approach, we obtain unsupervised 3D semantics. In both 3D and 2D unsupervised scene understanding, SceneDINO reaches state-of-the-art segmentation accuracy. Linear probing our 3D features matches the segmentation accuracy of a current supervised SSC approach. Additionally, we showcase the domain generalization and multi-view consistency of SceneDINO, taking the first steps towards a strong foundation for single image 3D scene understanding.

cs.CV↗

arXiv2025

Differentiable Halo Mass Prediction and the Cosmology-Dependence of Halo Mass Functions

Modern cosmological inference increasingly relies on differentiable models to enable efficient, gradient-based parameter estimation and uncertainty quantification. Here, we present a novel approach for predicting the abundance of dark matter haloes and their cosmology dependence using a differentiable, field-level neural network (NN) model, and study how well the cosmology dependence is captured by common parametrisations of the halo mass function (HMF), and by our NN-based approach. By training a 3D U-Net on initial density fields from fast N-body simulations with varying cosmological parameters, we enable direct, differentiable mapping from the linear density field to protohalo patches and their mass bins. Our method achieves competitive accuracy in identifying protohalo regions and in capturing the dependence of the HMF on cosmological parameters. Our NN derivatives agree well with finite differences of both analytical and emulated HMFs, at the level of the disagreement among the different models. We further demonstrate how the NN model can additionally be used to investigate the response of the HMF to changes in the initial Gaussian random field. Finally, we also demonstrate that a differentiable model can be used to extrapolate existing models at very high precision.

astro-ph.CO↗

arXiv2025

Accurate $N$-body simulations with local Primordial non-Gaussianities: initial conditions and aliasing

New generation galaxy surveys targeting constraints on local primordial non-Gaussianity (PNG) demand $N$-body simulations that accurately reproduce its effects. In this work, we explore various prescriptions for the initial conditions of simulations with PNG, aiming to optimise accuracy and minimise numerical errors, particularly due to aliasing. We have used $186$ runs that vary the starting redshift, LPT order, and non-Gaussianities ($f^{\rm local}_{\rm NL}$ and $g^{\rm local}_{\rm NL}$). Starting with $3$LPT at a redshift as low as $z_{\rm ini}\simeq 11.5$ reproduces to $<1 \%$ the power spectrum, bispectrum and halo mass function of a high-resolution reference simulation. The aliasing induced by the PNG terms in the power spectrum produces a $ \leq 3 \%$ excess small-scale power at the initial conditions, dropping below $0.1\%$ by $z=0$. State-of-the-art initial condition generators show a sub-percent agreement. We show that initial conditions for simulations with PNG should be established at a lower redshift using higher-order LPT schemes. We also show that removing the PNG aliasing signal is unnecessary for current simulations. The methodology proposed here can accelerate the generation of simulations with PNG while enhancing their accuracy.

astro-ph.CO↗

arXiv2025

Scene-Centric Unsupervised Panoptic Segmentation

Unsupervised panoptic segmentation aims to partition an image into semantically meaningful regions and distinct object instances without training on manually annotated data. In contrast to prior work on unsupervised panoptic scene understanding, we eliminate the need for object-centric training data, enabling the unsupervised understanding of complex scenes. To that end, we present the first unsupervised panoptic method that directly trains on scene-centric imagery. In particular, we propose an approach to obtain high-resolution panoptic pseudo labels on complex scene-centric data, combining visual representations, depth, and motion cues. Utilizing both pseudo-label training and a panoptic self-training strategy yields a novel approach that accurately predicts panoptic segmentation of complex scenes without requiring any human annotations. Our approach significantly improves panoptic quality, e.g., surpassing the recent state of the art in unsupervised panoptic segmentation on Cityscapes by 9.4% points in PQ.

cs.CV↗

arXiv2025

Boosting Omnidirectional Stereo Matching with a Pre-trained Depth Foundation Model

Omnidirectional depth perception is essential for mobile robotics applications that require scene understanding across a full 360{\deg} field of view. Camera-based setups offer a cost-effective option by using stereo depth estimation to generate dense, high-resolution depth maps without relying on expensive active sensing. However, existing omnidirectional stereo matching approaches achieve only limited depth accuracy across diverse environments, depth ranges, and lighting conditions, due to the scarcity of real-world data. We present DFI-OmniStereo, a novel omnidirectional stereo matching method that leverages a large-scale pre-trained foundation model for relative monocular depth estimation within an iterative optimization-based stereo matching architecture. We introduce a dedicated two-stage training strategy to utilize the relative monocular depth features for our omnidirectional stereo matching before scale-invariant fine-tuning. DFI-OmniStereo achieves state-of-the-art results on the real-world Helvipad dataset, reducing disparity MAE by approximately 16% compared to the previous best omnidirectional stereo method.

cs.CV↗

arXiv2024

Classical simulation of circuits with realistic odd-dimensional Gottesman-Kitaev-Preskill states

Classically simulating circuits with bosonic codes is challenging due to the prohibitive cost of simulating quantum systems with many, possibly infinite, energy levels. We propose an algorithm to simulate circuits with encoded Gottesman-Kitaev-Preskill (GKP) states, specifically for odd-dimensional encoded qudits. Our approach is tailored to be especially effective in the most challenging but practically relevant regime, where the codeword states exhibit high (but finite) squeezing. Our algorithm leverages the Zak-Gross Wigner function introduced by J. Davis et al. [arXiv:2407.18394], which represents infinitely squeezed encoded stabilizer states positively. The runtime of the algorithm scales with the negativity of the Wigner function, allowing for efficient simulation of certain large-scale circuits - namely, input stabilizer GKP states undergoing generalized GKP-encoded Clifford operations followed by modular measurements - with a high degree of squeezing. For stabilizer GKP states exhibiting 12 dB of squeezing, our algorithm can simulate circuits with up to 1,000 modes with less than double the number of samples required for a single input mode, in stark contrast to existing simulators. Therefore, this approach holds significant potential for benchmarking early implementations of quantum computing architectures utilizing bosonic codes.

quant-ph↗

arXiv2024

Assessing non-Gaussian quantum state conversion with the stellar rank

State conversion is a fundamental task in quantum information processing. Quantum resource theories allow for analyzing and bounding conversions that use restricted sets of operations. In the context of continuous-variable systems, state conversions restricted to Gaussian operations are crucial for both fundamental and practical reasons, particularly in state preparation and quantum computing with bosonic codes. However, previous analysis did not consider the relevant case of approximate state conversion. In this work, we introduce a framework for assessing approximate Gaussian state conversion by extending the stellar rank to the approximate stellar rank, which serves as an operational measure of non-Gaussianity. We derive bounds for Gaussian state conversion and distillation under approximate and probabilistic conditions, yielding new no-go results for non-Gaussian state preparation and enabling a reliable assessment of the performance of Gaussian conversion protocols. We also provide an open-source Python library to compute stellar-rank-related quantities and to assess Gaussian conversion.

quant-ph↗

arXiv2024

BullFrog: Multi-step perturbation theory as a time integrator for cosmological simulations

Modelling the cosmic large-scale structure can be done through numerical N-body simulations or by using perturbation theory. Here, we present an N-body approach that effectively implements a multi-step forward model based on Lagrangian Perturbation Theory (LPT) in a $\Lambda$CDM Universe. This is achieved by introducing the second-order accurate BullFrog integrator, which automatically performs 2LPT time steps to second order without requiring the explicit computation of 2LPT displacements. Importantly, we show that BullFrog trajectories rapidly converge to the exact solution as the number of time steps increases, at any moment in time, even though 2LPT becomes invalid after shell-crossing. As a validation test, we compare BullFrog against other N-body integrators and high-order LPT, both for a realistic $\Lambda$CDM cosmology and for simulations with a sharp UV cutoff in the initial conditions. The latter scenario enables controlled experiments against LPT and, in practice, is particularly relevant for modelling coarse-grained fluids arising in the context of effective field theory. We demonstrate that BullFrog significantly improves upon other LPT-inspired integrators, such as FastPM and COLA, without incurring any computational overhead compared to standard N-body integrators. Implementing BullFrog in any existing N-body code is straightforward, particularly if FastPM is already integrated.

astro-ph.CO↗