arXiv subjects

Hillel Kugler

Publications and source records attributed to Hillel Kugler.

4 recordsLinked to original sources

arXiv2025

Bridging Efficiency and Safety: Formal Verification of Neural Networks with Early Exits

Ensuring the safety and efficiency of AI systems is a central goal of modern research. Formal verification provides guarantees of neural network robustness, while early exits improve inference efficiency by enabling intermediate predictions. Yet verifying networks with early exits introduces new challenges due to their conditional execution paths. In this work, we define a robustness property tailored to early exit architectures and show how off-the-shelf solvers can be used to assess it. We present a baseline algorithm, enhanced with an early stopping strategy and heuristic optimizations that maintain soundness and completeness. Experiments on multiple benchmarks validate our framework's effectiveness and demonstrate the performance gains of the improved algorithm. Alongside the natural inference acceleration provided by early exits, we show that they also enhance verifiability, enabling more queries to be solved in less time compared to standard networks. Together with a robustness analysis, we show how these metrics can help users navigate the inherent trade-off between accuracy and efficiency.

cs.LG↗

arXiv2024

Formal Verification of Deep Neural Networks for Object Detection

Deep neural networks (DNNs) are widely used in real-world applications, yet they remain vulnerable to errors and adversarial attacks. Formal verification offers a systematic approach to identify and mitigate these vulnerabilities, enhancing model robustness and reliability. While most existing verification methods focus on image classification models, this work extends formal verification to the more complex domain of emph{object detection} models. We propose a formulation for verifying the robustness of such models and demonstrate how state-of-the-art verification tools, originally developed for classification, can be adapted for this purpose. Our experiments, conducted on various datasets and networks, highlight the ability of formal verification to uncover vulnerabilities in object detection models, underscoring the need to extend verification efforts to this domain. This work lays the foundation for further research into formal verification across a broader range of computer vision applications.

cs.CV↗

arXiv2022

Solving the 3-SAT problem using network-based biocomputation

The 3-Satisfiability Problem (3-SAT) is a demanding combinatorial problem, of central importance among the non-deterministic polynomial (NP) complete problems, with applications in circuit design, artificial intelligence and logistics. Even with optimized algorithms, the solution space that needs to be explored grows exponentially with increasing size of 3-SAT instances. Thus, large 3-SAT instances require excessive amounts of energy to solve with serial electronic computers. Network-based biocomputation (NBC) is a multidisciplinary parallel computation approach with drastically reduced energy consumption. NBC uses biomolecular motors to propel cytoskeletal filaments through nanofabricated networks that encode the mathematical problems. By stochastically exploring possible paths through the networks, the cytoskeletal filaments find possible solutions to the encoded problem instance. Here we first report a novel algorithm that converts 3-SAT into NBC-compatible network format. We demonstrate that this algorithm works in practice, by experimentally solving four small 3-SAT instances (with up to 3 variables and 5 clauses) using the actin-myosin biomolecular motor system. This is a key step towards the broad general applicability of NBC because polynomial conversions to 3-SAT exist for a wide set of important NP-complete problems.

physics.bio-ph↗

arXiv2021

Physical requirements for scaling up network-based biocomputation

The high energy consumption of electronic data processors, together with physical challenges limiting their further improvement, has triggered intensive interest in alternative computation paradigms. Here we focus on network-based biocomputation (NBC), a massively parallel approach that benefits from the energy efficiency of biological agents, such as molecular motors or bacteria, and their availability in large numbers. We analyse and define the fundamental requirements that need to be fulfilled to scale up NBC computers to become a viable technology that can solve large NP-complete problem instances faster or with less energy consumption than electronic computers. Our work can serve as a guide for further efforts to contribute to elements of future NBC devices, and as the theoretical basis for a detailed NBC roadmap.

physics.bio-ph↗