Logo des Repositoriums

Auflistung nach Autor:in "Reers, Volker"

1 - 2 von 2
  • Konferenzbeitrag
    Comparative Analysis of Vulnerabilities in Classical and Quantum Machine Learning
    (INFORMATIK 2024, 2024) Reers, Volker; Maußner, Marc
    Machine learning has made some remarkable breakthroughs in recent years. It has entered many sectors of the economy and everyday topics and in some cases has led to significant disruptions. The emergence of quantum computing is expected to lead to further significant increases in the performance of machine learning – regarding speed up of the training process and expressivity of the resulting models. However, as with all technologies, both classical and quantum machine learning are associated with new risks and attack vectors. This paper conducts a thorough examination of the vulnerabilities exhibited by classical and quantum machine learning models. Through a review of pertinent literature, we examine the vulnerability of classical models to attacks such as adversarial examples, evasion attacks, and poisoning attacks. Concurrently, we delve into the emerging realm of quantum machine learning, analyzing the unique properties of quantum systems and their implications for security in machine learning applications. Our comparative analysis offers insights into the robustness, scalability, and computational complexity of classical and quantum models under different attack scenarios. Furthermore, we discuss potential defense mechanisms and mitigation strategies to enhance the resilience of both classical and quantum machine learning frameworks against adversarial attacks.
  • Konferenzbeitrag
    Towards Performance Benchmarking for Quantum Reinforcement Learning
    (INFORMATIK 2023 - Designing Futures: Zukünfte gestalten, 2023) Reers, Volker
    This research paper explores the potential benefits and challenges of using Quantum Computing to enhance Reinforcement Learning (RL), a subset of Machine Learning that involves learning through trial and error. RL algorithms aim to learn a policy that maximizes a reward signal by interacting with an environment. Quantum Computing, which uses quantum-mechanical phenomena to process information, has been applied to several models in the realm of Machine Learning (ML). It therefore has the potential to accelerate computations of ML models which may be used to improve the efficiency of RL. However, current hardware limitations of quantum computers and the difficulty of designing quantum algorithms that outperform classical algorithms are major challenges to overcome. This paper discusses different approaches to using Quantum Computing for RL, such as using quantum algorithms to solve specific subproblems. In particular, we aim towards a performance benchmark of Quantum Reinforcement Learning using an implementation of Quantum Deep Q Learning as an example. While there is still much research to be done, the potential benefits of using Quantum Computing for RL are significant and warrant further exploration as quantum computing technology continues to evolve.