Searcharxiv⌕ Search

arXiv subjects

Amey Pasarkar

Publications and source records attributed to Amey Pasarkar.

2 recordsLinked to original sources

Vendi Anomaly Scores for Efficient and Accurate Anomaly Detection

Anomaly detection (AD) is a critical for trustworthy machine learning and scientific discovery. Existing methods typically define anomalies through local density, isolation heuristics, or distance from a fitted distribution, making them sensitive to neighborhood scale, partitioning choices, or restrictive distributional assumptions. In this work, we introduce a new paradigm by formulating anomaly detection from a diversity perspective. We propose the $\textit{Vendi Anomaly Score}$ (VAS), which detects anomalies by quantifying how much a sample changes the diversity of the dataset, as measured by the Vendi score, when removed. VAS captures both local redundancy and global data structure without relying on density estimation or partitioning heuristics. Moreover, VAS requires no hyperparameter tuning, instead adapting automatically to the spectral structure of the data. VAS is non-parametric and scales linearly with dataset size. Across 148 benchmark embedding datasets derived from 10 source datasets and 4 embedding architectures, VAS achieves state-of-the-art AD performance. We further validate VAS on large-scale ImageNet experiments, where it remains robust as both contamination rate and dataset scale increase.

cs.LG↗

A Unified and Predictive Measure of Functional Diversity

Despite the critical role of functional diversity (FD) in understanding ecological systems and processes, its robust quantification remains a significant challenge. A long-held view in the field is that it is not possible to capture its three facets -- functional richness, functional divergence, and functional evenness -- in a single index. This perspective has prompted recent proposals for FD measurement to use three separate indices, one for each aspect. Here, we challenge this paradigm by demonstrating that the probability-weighted Vendi Score (pVS), first introduced by Friedman and Dieng (2023), can serve as a powerful functional diversity index that can capture its three facets. We adapt pVS to functional ecology by defining it as the exponential of the Rényi entropy of the eigenvalues of the abundance-weighted trait similarity matrix. This formulation allows pVS to be applicable at any biological level. It can be defined at the species level, at which most existing FD metrics are defined, and at the individual level to naturally incorporate intraspecific trait variation (ITV) when detailed data are available. We theoretically and empirically demonstrate the robustness of pVS. We first mathematically prove it satisfies several essential desiderata for FD metrics, including invariance to functional redundancy, set monotonicity, distance monotonicity, and concavity. We then show that pVS consistently exhibits the expected ground-truth behavior on simulated ecosystem scenarios under which many FD metrics fail. By integrating abundances and trait similarities within a single, theoretically sound framework, pVS provides a generally applicable index for ecology.

q-bio.PE↗