View Item 
        •   Utrecht University Student Theses Repository Home
        • UU Theses Repository
        • Theses
        • View Item
        •   Utrecht University Student Theses Repository Home
        • UU Theses Repository
        • Theses
        • View Item
        JavaScript is disabled for your browser. Some features of this site may not work without it.

        Browse

        All of UU Student Theses RepositoryBy Issue DateAuthorsTitlesSubjectsThis CollectionBy Issue DateAuthorsTitlesSubjects

        Testing Mixed-Precision for VGG, Inception and ResNet on the Dogs vs. Cats Dataset

        Thumbnail
        View/Open
        5532191_Steendam_BSc_KI_Thesis_Mixed_Precision.pdf (390.2Kb)
        Publication date
        2020
        Author
        Steendam, R.P.
        Metadata
        Show full item record
        Summary
        Mixed-precision floating points seem promising in reducing computation costs for deep neural networks. But does the technique live up to the promise and do all network architectures benefit equally from mixed-precision? Using the Dogs vs. Cats dataset we researched the effect of using mixed-precision on VGG, Inception and ResNet by measuring accuracy, training speed and inference speed. The results showed that the accuracy of mixed-precision was comparable with that of single-precision. Furthermore, all networks became faster, both in training and inference. The speedup between the different architectures varied between 32% and 45% for training and between 10% and 40% for inference.
        URI
        https://studenttheses.uu.nl/handle/20.500.12932/36152
        Collections
        • Theses
        Utrecht university logo