Early breast cancer detection (BCD) through mammography screening continues to be a major focus in radiology as it plays a critical role in reducing mortality rates (Coleman (2017); Ginsburg et al. (2020)). Although artificial intelligence (AI) models can help radiologists to evaluate mammograms (Sahu et al. (2023); Evans et al. (2013); Maxwell (1999)), training such models face the challenge of limited datasets that may not fully represent all subgroups or cover variations in data distributions. Historically, certain racial groups face barriers to healthcare access because of many socio-economic factors (Azin et al. (2023); Hershman et al. (2005); Hussain-Gambles et al. (2004)). This lack of access can result in datasets that do not adequately represent these groups, potentially cause AI models to show biases for these groups. Even with seemingly balanced datasets, subtle biases may persist in the collected data due to systemic inequalities in the quality of healthcare (Obermeyer et al. (2019)). Among these groups, African American patients are often underrepresented in both breast imaging and broader healthcare datasets (Yedjou et al. (2019); Newman and Kaljee (2017)).

Federated learning (FL) is a popular paradigm for private and collaborative model training on the edge. In centralized FL, the parameters of a global architecture (such as a deep neural network) are maintained and distributed by a central server/controller to clients who transmit model updates (gradients) back to the server based on local optimization. While many efforts have focused on reducing the communication complexity of gradient transmission, the vast majority of compression-based algorithms assume that each participating client is able to download and train the current and full set of parameters, which may not be a practical assumption depending on the resource constraints of smaller clients such as mobile devices. In this work, we propose a simple yet effective novel algorithm, Comfetch, which allows clients to train large networks using reduced representations of the global architecture via the count sketch, which reduces local computational and memory costs along with bi-directional communication complexity. We provide a nonconvex convergence guarantee and experimentally demonstrate that it is possible to learn large models, such as a deep convolutional network, through federated training on their sketched counterparts. The resulting global models exhibit competitive test accuracy over CIFAR10/100 classification when compared against un-compressed model training.