Max Planck Institute for Informatics

ABSTRACT Analysis of 60 northern pike ( Esox lucius ) populations, assembled from published studies spanning the species' range and restricted to non‐flowing (lentic) systems, and analyzed using the von Bertalanffy growth function (VBGF), revealed substantial heterogeneity in growth parameters. Growth coefficients ( k ) ranged from 0.01 to 0.72 year −1 , asymptotic lengths () from 53 to 976 cm, and maximum observed lengths () from 35.1 to 113.2 cm. Biologically defensible thresholds were identified at k ≥ 0.25 year −1 and ≤ 113.2 cm, lose biological interpretability under realistic demographic constraints. Within this framework, the VBGF served two complementary roles: as a predictive model of linear growth trajectories and as a diagnostic representation of growth‐trajectory shape. The ratio emerged as an informative descriptor, capturing ontogenetic growth patterns along a continuum from near‐linear ( ≈0.07) to fully asymptotic ( ≈1.00) growth. This metric links empirical size‐at‐age observations with the degree to which asymptotic growth is expressed within finite lifespans. Integrating VBGF parameters ( k , ) with empirical descriptors (, lifespan) and cumulative thermal exposure (growing degree‐days) revealed two coherent growth regimes in northern pike along latitudinal thermal gradients: fast‐growing populations reaching moderate asymptotic sizes rapidly, and slow‐growing populations characterized by low growth coefficients and very large theoretical despite similar observed maximum lengths. Taken together, the ratio, combined with biologically grounded thresholds for k and , provides a parsimonious screening tool for identifying non‐asymptotic growth trajectories and a temperature‐ and time‐explicit framework for evaluating the expression of asymptotic growth across climatic gradients.

We show that if the ground set of a matroid can be partitioned into \(k \ge 2\) bases, then for any given subset \(S\) of the ground set, there is a partition into k bases such that the sizes of the intersections of the bases with \(S\) may differ by at most one. This settles the matroid equitability conjecture by Fekete and Szabo (Electron. J. Comb. 2011) in the affirmative. We also investigate equitable splittings of two disjoint sets \(S_1\) and \(S_2\), and show that there is a partition into \(k\) bases such that the sizes of the intersections with \(S_1\) may differ by at most one and the sizes of the intersections with \(S_2\) may differ by at most two; this is the best one can hope for arbitrary matroids.

In this work, we study the limits of compressed data structures, i.e., structures that support various queries on an input text \(T \in \Sigma^n\) while using space proportional to the size of \(T\) in compressed form. Nearly all fundamental queries can currently be efficiently supported in \(\mathcal{O}(\delta(T)\log^{\mathcal{O}(1)} n)\) space, where \(\delta(T)\) is the substring complexity—a strong compressibility measure that lower-bounds the optimal space required to represent the text [Kociumaka, Navarro, Prezza; IEEE Trans. Inf. Theory 2023]. In contrast, the optimal query time for compressed data structures has been characterized only for the basic random access problem.

We revisit the (\(f,q\))-Clustering problem that we introduced in a recent work [SODA’25]. Here, \(f\) and \(g\) are symmetric, monotone norms called inner and outer norms, respectively. The task is to partition a given set of points in a metric space into \(k\) clusters each represented by a cluster center. Each cluster is assigned a cluster cost, determined by the norm \(f\) applied to the vector of point-center distances in the cluster. The goal is to minimize the value of the norm \(g\) when applied to the vector of cluster costs. This problem subsumes fundamental clustering problems such as \(k\)-Center (i.e., \(\mathcal L_\infty, \mathcal L_\infty\)-Clustering), \(k\)-Median (i.e., \(\mathcal L_1, \mathcal L_1\)-Clustering), Min-Sum of Radii (i.e., \(\mathcal L_\infty, \mathcal L_1\)-Clustering), and Min-Load \(k\)-Clustering (i.e., \(\mathcal L_1, \mathcal L_\infty\)-Clustering). In our previous work, we focused on certain special cases of this problem for which we designed constant-factor approximation algorithms. Our bounds for more general settings left, however, large gaps to the known bounds for the basic problems they capture.

A central task in string processing is text indexing, where the goal is to preprocess a text (a string of length \(n\)) into an efficient index (a data structure) supporting queries about the text. While the most fundamental exact pattern matching queries ask to find all the occurrences of a pattern (a string of length \(m\)) as substrings of the text, many applications call for approximate pattern matching queries, where the pattern may differ slightly from the matching substrings. A breakthrough in the extensive study of approximate text indexing came from Cole, Gottlieb, and Lewenstein (STOC 2004), who proposed \(k\)-errata trees — a family of text indexes supporting several closely related flavors of approximate pattern matching queries. In particular, \(k\)-errata trees yield an elegant solution to \(k\)-mismatch queries, where the similarity is quantified using an upper bound \(k \ge 1\) on the Hamming distance between the pattern and its approximate occurrences. The resulting \(k\)-mismatch index uses \(\mathcal{O}(n \log^{k} n)\) space and answers a query for a length-\(m\) pattern in \(\mathcal{O}(\log^{k} n \log \log n + m + \texttt{occ})\) time, where \(\texttt{occ}\) is the number of approximate occurrences.

In this paper, we present PanoDreamer, a novel method for producing a coherent 360° 3D scene from a single input image. Unlike existing methods that generate the scene sequentially, we frame the problem as single-image panorama and depth estimation. Once the coherent panoramic image and its corresponding depth are obtained, the scene can be reconstructed by inpainting the small occluded regions and projecting them into 3D space. Our key contribution is formulating single-image panorama and depth estimation as two optimization tasks and introducing alternating minimization strategies to effectively solve their objectives. We demonstrate that our approach outperforms existing techniques in single-image 360° 3D scene reconstruction in terms of consistency and overall quality.

Rendering novel, relit views of a human head, given a monocular portrait image as input, is an inherently underconstrained problem. The traditional graphics solution is to explicitly decompose the input image into geometry, material and lighting via differentiable rendering; but this is constrained by the multiple assumptions and approximations of the underlying models and parameterizations of these scene components. We propose 3DPR, an image-based relighting model that leverages generative priors learnt from multi-view One-Light-at-A-Time (OLAT) images captured in a light stage. We introduce a new diverse and large-scale multi-view 4K OLAT dataset of 139 subjects to learn a high-quality prior over the distribution of high-frequency face reflectance. We leverage the latent space of a pre-trained generative head model that provides a rich prior over face geometry learnt from in-the-wild image datasets. The input portrait is first embedded in the latent manifold of such a model through an encoder-based inversion process. Then a novel triplane-based reflectance network trained on our lightstage data is used to synthesize high-fidelity OLAT images to enable image-based relighting. Our reflectance network operates in the latent space of the generative head model, crucially enabling a relatively small number of lightstage images to train the reflectance model. Combining the generated OLATs according to a given HDRI environment maps yields physically accurate environmental relighting results. Through quantitative and qualitative evaluations, we demonstrate that 3DPR outperforms previous methods, particularly in preserving identity and in capturing lighting effects such as specularities, self-shadows, and subsurface scattering.

Machine learned potentials based on artificial neural networks are becoming a popular tool to define an effective energy model for complex systems, either incorporating electronic structure effects at the atomistic resolution, or effectively renormalizing part of the atomistic degrees of freedom at a coarse-grained resolution. One main criticism regarding neural network potentials is that their inferred energy is less interpretable than in traditional approaches, which use simpler and more transparent functional forms. Here we address this problem by extending tools recently proposed in the nascent field of explainable artificial intelligence to coarse-grained potentials based on graph neural networks. With these tools, neural network potentials can be practically decomposed into n-body interactions, providing a human understandable interpretation without compromising predictive power. We demonstrate the approach on three different coarse-grained systems including two fluids (methane and water) and the protein NTL9. The obtained interpretations suggest that well-trained neural network potentials learn physical interactions, which are consistent with fundamental principles.

Cloud-hosted environments are increasingly being adopted in both industry and academia alike, with marketing material and popular belief holding a myriad of conjectures as to why this adoption progresses. There is, however, a limited understanding of the factors that influence the decision to move to the cloud and the challenges encountered on this migration journey, grounded in data, specifically in the context of higher education. To address this gap, we perform an in-depth qualitative investigation to better understand the human factors of cloud adoption in universities. We engaged with 18 participants via semi-structured interviews and analysed the data using thematic analysis. We present our findings around three main themes - the decision matrix, the people-centric transition and the guardians of data - to comprehend the landscape of cloud migration ( cloudscape ) in higher education. Our findings highlight that cloud migrations go beyond a technical transformation. While they regularly encompass shifts in people's roles and mindsets, cloud adoption in the absence of such a shift may lead to patterns that triggered the move away from self-hosted and on-premise solutions in the first place. Furthermore, we find interaction effects between the technical depth of decision makers in contrast to necessary risk acceptance in the context of cloud adoption, thereby further adding to the broader discussion regarding the place of academia in monopolistic markets.