Johann Maximilian Zollner

arXiv2025

Extending the Applicability of Bloom Filters by Relaxing their Parameter Constraints

These days, Key-Value Stores are widely used for scalable data storage. In this environment, Bloom filter (BF) serves as an efficient probabilistic data structure for representing sets of keys. They allow for set membership queries with no false negatives and with the right choice of the main parameters - length of the BF, number of hash functions used to map an element to the array's indices, and the number of elements inserted - the false positive rate is optimized. However, the number of hash functions is constrained to integer values, and the length of a BF is usually chosen to be a power of two to allow for efficient modulo operations using binary arithmetic. In this paper, we relax these constraints by proposing the Rational Bloom filter, which allows for non-integer numbers of hash functions. This results in optimized fraction-of-zero values for a known number of elements to be inserted. Based on this, we construct the Variably-Sized Block BF to allow for a flexible filter length, especially for large filters, with efficient computation.

cs.DS↗

arXiv2025

Enhancing Contrastive Learning for Geolocalization by Discovering Hard Negatives on Semivariograms

Accurate and robust image-based geo-localization at a global scale is challenging due to diverse environments, visually ambiguous scenes, and the lack of distinctive landmarks in many regions. While contrastive learning methods show promising performance by aligning features between street-view images and corresponding locations, they neglect the underlying spatial dependency in the geographic space. As a result, they fail to address the issue of false negatives -- image pairs that are both visually and geographically similar but labeled as negatives, and struggle to effectively distinguish hard negatives, which are visually similar but geographically distant. To address this issue, we propose a novel spatially regularized contrastive learning strategy that integrates a semivariogram, which is a geostatistical tool for modeling how spatial correlation changes with distance. We fit the semivariogram by relating the distance of images in feature space to their geographical distance, capturing the expected visual content in a spatial correlation. With the fitted semivariogram, we define the expected visual dissimilarity at a given spatial distance as reference to identify hard negatives and false negatives. We integrate this strategy into GeoCLIP and evaluate it on the OSV5M dataset, demonstrating that explicitly modeling spatial priors improves image-based geo-localization performance, particularly at finer granularity.

cs.CV↗