Brief

Recent research in federated learning (FL) addresses critical challenges in privacy and data drift. One paper introduces TADI and Fulcrum to protect against topology-aware inference attacks by optimally allocating noise, showing privacy gains without utility loss. Another study proposes FlashbackCL, an extension to existing methods, which mitigates temporal forgetting in FL by using decayed label counts and a device-aware replay buffer, achieving significant improvements over prior work. Additional research explores personalized Bayesian FL approaches like pFedBayes, sFedBayes, and cFedbayes to handle data heterogeneity, alongside FedSAP for stable federated representation learning and FedMChain for multimodal FL by optimizing modalities sequentially. Finally, IntraShuffler is presented as a defense against privacy inference attacks in heterogeneous DP FL by shuffling client updates within privacy-compatible buckets. AI

Researchers have introduced Flashback, a novel reversible string decomposition method that operates by peeling maximal leading and trailing character runs from an input. This process records each pair as a bilateral token, with both decomposition and reconstruction achieving O(n) time and space complexity. A key finding is the run-pairing theorem, which establishes Flashback's equivalence to pairing runs from the string's ends inward, yielding a token count of 1+[r/2] for strings with r maximal runs. AI