{"about":{"site":"https://codewithpapers.app","non_affiliation":"Code with Papers and Syntology are not affiliated with, endorsed by, or sponsored by Papers with Code, Meta, or the pwc-archive mirror.","licence":"CC BY-SA 4.0","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","attribution":"https://codewithpapers.app/attribution","modified":"archive material modified by Syntology; see the attribution page"},"url":"/paper/how-does-topology-of-neural-architectures","title":"How does topology of neural architectures impact gradient propagation and model performance?","arxiv_id":null,"date":"2021-06-16","proceeding":"CVPR 2021 6","authors":["Kartikeya Bhardwa","Guihong Li2","Radu Marculescu"],"abstract":"DenseNets introduce concatenation-type skip connections that achieve state-of-the-art accuracy in several computer vision tasks. In this paper, we reveal that the topology of the concatenation-type skip connections is closely related to the gradient propagation which, in turn, enables a pre\u0002dictable behavior of DNNs’ test performance. To this end, we introduce a new metric called NN-Mass to quantify how effectively information flows through DNNs. Moreover, we empirically show that NN-Mass also works for other types of skip connections, e.g., for ResNets, Wide-ResNets (WRNs), and MobileNets, which contain addition-type skip connections (i.e., residuals or inverted residuals). As such, for both DenseNet-like CNNs and ResNets/WRNs/MobileNets, our theoretically grounded NN-Mass can identify models with similar accuracy, despite having significantly different size/compute requirements. Detailed experiments on both synthetic and real datasets (e.g., MNIST, CIFAR-10, CIFAR\u0002100, ImageNet) provide extensive evidence for our insights. Finally, the closed-form equation of our NN-Mass enables us to design significantly compressed DenseNets (for CIFAR\u000210) and MobileNets (for ImageNet) directly at initialization without time-consuming training and/or searching.","url_abs":"https://openaccess.thecvf.com/content/CVPR2021/html/Bhardwaj_How_Does_Topology_Influence_Gradient_Propagation_and_Model_Performance_of_CVPR_2021_paper.html","url_pdf":"https://arxiv.org/pdf/1910.00780.pdf","source":{"archive":"pwc-archive (Hugging Face), CC BY-SA 4.0","snapshot":"2025-07-28","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","row_kind":"abstracts"},"code_links":[{"paper_slug":"how-does-topology-of-neural-architectures","repo_url":"https://github.com/SLDGroup/NN_Mass","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":0,"framework":"pytorch","reach":null}],"tasks":[{"task_slug":"model-compression","task_name":"Model Compression"},{"task_slug":"architecture-search","task_name":"Neural Architecture Search"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[{"leaderboard":"/sota/neural-architecture-search-on-cifar-10","task":"Neural Architecture Search","dataset":"CIFAR-10","model":"NN-MASS- CIFAR-A","rank_in_archive_order":35,"of":41,"metrics":{"FLOPS":"1.95G","Parameters":"5.02M","Search Time (GPU days)":"0","Top-1 Error Rate":"3.0%"},"uses_additional_data":false},{"leaderboard":"/sota/neural-architecture-search-on-cifar-10","task":"Neural Architecture Search","dataset":"CIFAR-10","model":"NN-MASS- CIFAR-C","rank_in_archive_order":36,"of":41,"metrics":{"FLOPS":"1.2G","Parameters":"3.82M","Search Time (GPU days)":"0","Top-1 Error Rate":"3.18%"},"uses_additional_data":false},{"leaderboard":"/sota/neural-architecture-search-on-imagenet","task":"Neural Architecture Search","dataset":"ImageNet","model":"NN-MASS-B","rank_in_archive_order":123,"of":135,"metrics":{"Accuracy":"73.3","FLOPs":"393M","MACs":"393M","Params":"3.7M","Top-1 Error Rate":"26.7"},"uses_additional_data":false},{"leaderboard":"/sota/neural-architecture-search-on-imagenet","task":"Neural Architecture Search","dataset":"ImageNet","model":"NN-MASS-A","rank_in_archive_order":126,"of":135,"metrics":{"Accuracy":"72.9","FLOPs":"200M","MACs":"200M","Params":"2.3M","Top-1 Error Rate":"27.1"},"uses_additional_data":false}],"syntology":{"syntology_url":null,"atlas_url":null,"mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}