arXiv subjects

Sebastiano Cultrera di Montesano

Publications and source records attributed to Sebastiano Cultrera di Montesano.

9 recordsLinked to original sources

arXiv2026

Score Distributions, Not Cells: Evaluating Single-Cell Perturbations Under Class Overlap

Most classification problems assume the classes are roughly separable, so that an individual sample can usually be assigned to one class. Single-cell perturbation data violates this assumption: two perturbations can produce different populations of cells while overlapping so much that an individual cell could belong to either. Per-cell accuracy then measures this overlap rather than model quality. We see this on Tahoe-100M and the Virtual Cell Challenge, where a linear classifier, an MLP, and a Transformer all plateau near macro-F1 0.2-0.3 even though almost every pair of perturbations is statistically distinguishable. The fix is to score perturbations across the whole population rather than cell by cell. We average a classifier's per-cell probability vectors over all cells of a perturbation to form a population profile, then rank candidate perturbations by this profile; we call the resulting score the Classifier Discrimination Score (CDS). Taking the top-ranked class recovers the winning perturbation. It needs no retraining, costs linear time in the number of cells, and recovers near-perfect identification from the same weak models. CDS differs from the pseudobulk-based Perturbation Discrimination Score (PDS) used in recent benchmarks only in where the average is taken, raw gene expression for PDS versus a learned discriminative space for CDS, and identifies the true perturbation more reliably on both datasets, with the gap widening as cells grow scarce. Because a metric that misranks the ground truth will misrank the models scored against it, per-cell accuracy and raw-pseudobulk scores should be used with caution when comparing perturbation models.

cs.LG↗

arXiv2026

scDataset: Scalable Data Loading for Deep Learning on Large-Scale Single-Cell Omics

Training deep learning models on single-cell datasets with hundreds of millions of cells requires loading data from disk, as these datasets exceed available memory. While random sampling provides the data diversity needed for effective training, it is prohibitively slow due to the random access pattern overhead, whereas sequential streaming achieves high throughput but introduces biases that degrade model performance. We present scDataset, a PyTorch data loader that enables efficient training from on-disk data with seamless integration across diverse storage formats. Our approach combines block sampling and batched fetching to achieve quasi-random sampling that balances I/O efficiency with minibatch diversity. On Tahoe-100M, a dataset of 100 million cells, scDataset achieves more than two orders of magnitude speedup compared to true random sampling while working directly with AnnData files. We provide theoretical bounds on minibatch diversity and empirically show that scDataset matches the performance of true random sampling across multiple classification tasks and model architectures.

cs.LG↗

arXiv2026

When Labels Have Structure: Improving Image Classification with Hierarchy-Aware Cross-Entropy

Standard cross-entropy is the default classification loss across virtually all of machine learning, yet it treats all misclassifications equally, ignoring the semantic distances that a class hierarchy encodes. We propose Hierarchy-Aware Cross-Entropy (HACE), a drop-in replacement for standard cross-entropy that incorporates a known class hierarchy directly into the loss. HACE combines two components: prediction aggregation, which propagates the model's probability mass upward through the class hierarchy to ensure that parent nodes accumulate the confidence of their children; and ancestral label smoothing, which distributes the ground-truth signal along the path from the true class to the root. We evaluate HACE on CIFAR-100, FGVC Aircraft, and NABirds in two regimes: end-to-end training across six architectures spanning convolutional and attention-based designs, and linear probing on frozen DINOv2-Large features. In end-to-end training, HACE improves accuracy over standard cross-entropy in 15 out of 18 architecture--dataset pairs, with a mean gain of 4.66\%. In linear probing on frozen DINOv2-Large features, HACE outperforms all competing methods on all three datasets, with a mean improvement of 2.18\% over the next best baseline.

cs.LG↗

arXiv2025

Banana Trees for the Persistence in Time Series Experimentally

In numerous fields, dynamic time series data require continuous updates, necessitating efficient data processing techniques for accurate analysis. This paper examines the banana tree data structure, specifically designed to efficiently maintain persistent homology -- a multi-scale topological descriptor -- for dynamically changing time series data. We implement this data structure and conduct an experimental study to assess its properties and runtime for update operations. Our findings indicate that banana trees are highly effective with unbiased random data, outperforming state-of-the-art static algorithms in these scenarios. Additionally, our results show that real-world time series share structural properties with unbiased random walks, suggesting potential practical utility for our implementation.

cs.DS↗

arXiv2024

The Euclidean MST-ratio for Bi-colored Lattices

Given a finite set, $A \subseteq \mathbb{R}^2$, and a subset, $B \subseteq A$, the \emph{MST-ratio} is the combined length of the minimum spanning trees of $B$ and $A \setminus B$ divided by the length of the minimum spanning tree of $A$. The question of the supremum, over all sets $A$, of the maximum, over all subsets $B$, is related to the Steiner ratio, and we prove this sup-max is between $2.154$ and $2.427$. Restricting ourselves to $2$-dimensional lattices, we prove that the sup-max is $2.0$, while the inf-max is $1.25$. By some margin the most difficult of these results is the upper bound for the inf-max, which we prove by showing that the hexagonal lattice cannot have MST-ratio larger than $1.25$.

cs.CG↗

arXiv2024

Chromatic Alpha Complexes

Motivated by applications in the medical sciences, we study finite chromatic sets in Euclidean space from a topological perspective. Based on the persistent homology for images, kernels and cokernels, we design provably stable homological quantifiers that describe the geometric micro- and macro-structure of how the color classes mingle. These can be efficiently computed using chromatic variants of Delaunay and alpha complexes, and code that does these computations is provided.

math.AT↗

arXiv2024

Dynamically Maintaining the Persistent Homology of Time Series

We present a dynamic data structure for maintaining the persistent homology of a time series of real numbers. The data structure supports local operations, including the insertion and deletion of an item and the cutting and concatenating of lists, each in time $O(\log n + k)$, in which $n$ counts the critical items and $k$ the changes in the augmented persistence diagram. To achieve this, we design a tailor-made tree structure with an unconventional representation, referred to as banana tree, which may be useful in its own right.

cs.DS↗

arXiv2024

Chromatic Topological Data Analysis

Exploring the shape of point configurations has been a key driver in the evolution of TDA (short for topological data analysis) since its infancy. This survey illustrates the recent efforts to broaden these ideas to model spatial interactions among multiple configurations, each distinguished by a color. It describes advances in this area and prepares the ground for further exploration by mentioning unresolved questions and promising research avenues while focusing on the overlap with discrete geometry.

cs.CG↗

arXiv2022

On the Size of Chromatic Delaunay Mosaics

Given a locally finite set $A \subseteq \mathbb{R}^d$ and a coloring $χ\colon A \to \{0,1,\ldots,s\}$, we introduce the chromatic Delaunay mosaic of $χ$, which is a Delaunay mosaic in $\mathbb{R}^{s+d}$ that represents how points of different colors mingle. Our main results are bounds on the size of the chromatic Delaunay mosaic, in which we assume that $d$ and $s$ are constants. For example, if $A$ is finite with $n = \#{A}$, and the coloring is random, then the chromatic Delaunay mosaic has $O(n^{\lceil{d/2}\rceil})$ cells in expectation. In contrast, for Delone sets and Poisson point processes in $\mathbb{R}^d$, the expected number of cells within a closed ball is only a constant times the number of points in this ball. Furthermore, in $\mathbb{R}^2$ all colorings of a dense set of $n$ points have chromatic Delaunay mosaics of size $O(n)$. This encourages the use of chromatic Delaunay mosaics in applications.

math.CO↗