Investigating the Impact of Randomness on Reproducibility in Computer Vision: A Study on Applications in Civil Engineering and Medicine

Abstract--Reproducibility is essential for scientific research. CUDA's advantages for accelerating algorithm execution on GPUs, if not controlled, its behavior across multiple executions This reality emphasizes the importance of The reproducibility crisis in machine learning is a growing understanding CUDA-induced randomness and its impact on concern that questions the reliability and validity of reported model performance in real-world applications. One survey shows that not all researchers lack of such insight could lead researchers to overestimate or are aware of this problem [2]. This issue stems from the difficulty underestimate the capabilities of machine learning algorithms, in replicating results due to various unknown and poorly which in turn could misdirect the efforts of the research understood factors, including but not limited to differences community. Deep learning architectures, as they are widely implications on reproducibility, and its broader implications on used in computer vision, with their complex, multi-layered real-world computer vision applications.