Adding causal representation learning and latent diffusion to split federated learning made multi-task medical segmentation converge where baseline SplitFed did not
AspireAI community · arXiv
Federated learning aggregates model updates across clients without sharing raw data, and split federated learning further partitions the model between clients and a server to reduce client computation and communication. Decentralized medical institutions rarely operate on a single shared task, so standard federated and split-federated collaborations are poorly aligned with real clinical workflows; multi-task federated learning allows clients to handle different tasks but often introduces instability and privacy vulnerabilities. MuCALD-SplitFed is a multi-task split-federated framework that integrates causal representation learning and latent diffusion. Experiments show MuCALD-SplitFed consistently improving segmentation, while baseline SplitFed fails to converge, and the approach further reduces information leakage at the split points, mitigating reconstruction-based and membership inference attacks, outperforming state-of-the-art personalized federated learning and multi-task federated learning approaches. The limitation is that the evaluation is confined to medical image segmentation, so the stability benefit of combining causal representation learning with latent diffusion in a split setting has not been shown outside that domain.
Read the story and explore more from the AspireAI community.
Two sparse-autoencoder analyses have resisted scaling: matching semantically similar features across layers, and compressing large feature circuits into interpretable supernodes. This paper argues both are the same underlying problem, estimating semantic distances between SAE features that sit on different activation manifolds. Its representation treats each feature not as a single decoder vector but as an activation-weighted distribution over the hidden states expressing it, projects those distributions into a shared reference space, and compares them with Wasserstein distance. The representation is proved invariant to activation rescaling and stable under perturbation, and recovers true matches under finite-sample margin conditions. Empirically it outperformed decoder-vector and LLM-based baselines and captured subtle functional distinctions between related features, and it compressed large feature circuits into interpretable supernodes automatically. The limitation is that this is interpretability tooling: the paper demonstrates better feature matching and circuit compression, not that downstream interpretability claims become more accurate.