Adding Binary Search Connections to improve DenseNet Performance

<p>This paper introduces Binary Search Connections (BSC), a novel mechanism designed to enhance DenseNet by selectively reinforcing feature flow using binary search logic. While DenseNet connects each layer to all previous layers, BSC employs a recursive strategy to link non-adjacent layers based on a divide-and-conquer principle. This approach leads to the intentional repetition of important features from earlier layers, improving gradient flow and learning efficiency. Although BSC builds upon DenseNet, it enhances model performance by strategically duplicating features through depth-aware connections. Experiments on CIFAR-100 demonstrate that BSC-DenseNet-121 outperforms both DenseNet-121 and deeper variants, showcasing its effectiveness in boosting model performance. Source code: https://github.com/mr-ravin/BSC-DenseNet</p>