The field of video compression has developed some of the most sophisticated and efficient compression algorithms known in the literature, enabling very high compressibility for little loss of information. Whilst some of these techniques are domain specific, many of their underlying principles are universal in that they can be adapted and applied for compressing different types of data. In this work we present DeepCABAC, a compression algorithm for deep neural networks that is based on one of the state-of-the-art video coding techniques. Concretely, it applies a Context-based Adaptive Binary Arithmetic Coder (CABAC) to the network's parameters, which was originally designed for the H.264/AVC video coding standard and became the state-of-the-art for lossless compression. Moreover, DeepCABAC employs a novel quantization scheme that minimizes the rate-distortion function while simultaneously taking the impact of quantization onto the accuracy of the network into account. Experimental results show that DeepCABAC consistently attains higher compression rates than previously proposed coding techniques for neural network compression. For instance, it is able to compress the VGG16 ImageNet model by x63.6 with no loss of accuracy, thus being able to represent the entire network with merely 8.7MB. The source code for encoding and decoding can be found at https://github.com/fraunhoferhhi/DeepCABAC.

From all different proposed We present DeepCABAC, a novel contextadaptive methods, sparsification followed by weight quantization and binary arithmetic coder for compressing entropy coding arguably belong to the set of most popular deep neural networks. It quantizes each weight parameter approaches, since very high compression ratios can be by minimizing a weighted rate-distortion achieved under such paradigm (Han et al., 2015a; Louizos function, which implicitly takes the impact of et al., 2017; Wiedemann et al., 2018a;b). Whereas much of quantization on to the accuracy of the network research has focused on the sparsification part, a substantially into account. Subsequently, it compresses the less amount have focused on improving the later two quantized values into a bitstream representation steps. In fact, most of the proposed (post-sparsity) compression with minimal redundancies. We show that Deep-algorithms come with at least one of the following CABAC is able to reach very high compression caveats: 1) they decouple the quantization procedure from ratios across a wide set of different network architectures the subsequent lossless compression algorithm, 2) ignore and datasets. For instance, we are correlations between the parameters and 3) apply a lossless able to compress by x63.6 the VGG16 ImageNet compression algorithm that produce a bitstream with more model with no loss of accuracy, thus being able to redundancies than principally needed (e.g.