arXiv subjects

Kasimir Gabert

Publications and source records attributed to Kasimir Gabert.

2 recordsLinked to original sources

arXiv2026

Trojans in Artificial Intelligence (TrojAI) Final Report

The Intelligence Advanced Research Projects Activity (IARPA) launched the TrojAI program to confront an emerging vulnerability in modern artificial intelligence: the threat of AI Trojans. These AI trojans are malicious, hidden backdoors intentionally embedded within an AI model that can cause a system to fail in unexpected ways, or allow a malicious actor to hijack the AI model at will. This multi-year initiative helped to map out the complex nature of the threat, pioneered foundational detection methods, and identified unsolved challenges that require ongoing attention by the burgeoning AI security field. This report synthesizes the program's key findings, including methodologies for detection through weight analysis and trigger inversion, as well as approaches for mitigating Trojan risks in deployed models. Comprehensive test and evaluation results highlight detector performance, sensitivity, and the prevalence of "natural" Trojans. The report concludes with lessons learned and recommendations for advancing AI security research.

cs.CR↗

arXiv2022

Batch Dynamic Algorithm to Find k-Cores and Hierarchies

Finding $k$-cores in graphs is a valuable and effective strategy for extracting dense regions of otherwise sparse graphs. We focus on the important problem of maintaining cores on rapidly changing dynamic graphs, where batches of edge changes need to be processed quickly. Prior batch core algorithms have only addressed half the problem of maintaining cores, the problem of maintaining a core decomposition. This finds vertices that are dense, but not regions; it misses connectivity. To address this, we bring an efficient index from community search into the core domain, the Shell Tree Index. We develop a novel dynamic batch algorithm to maintain it that improves efficiency over processing edge-by-edge. We implement our algorithm and experimentally show that with it core queries can be returned on rapidly changing graphs quickly enough for interactive applications. For 1 million edge batches, on many graphs we run over $100\times$ faster than processing edge-by-edge while remaining under re-computing from scratch.

cs.DS↗