Improving $(\alpha, f)$-Byzantine Resilience in Federated Learning via layerwise aggregation and cosine distance