selected

Sparse Autoencoders (SAEs) decompose model activations into sparse combinations of interpretable dictionary atoms. Although SAEs are …

Transformer components such as non-linear activations and normalization are inherently non-injective, suggesting that different inputs …

Model merging has recently emerged as a lightweight alternative to ensembling, combining multiple fine-tuned models into a single set …

Task Arithmetic has emerged as a simple yet effective method to merge models without additional training. However, by treating entire …

In this paper, we present a novel data-free method for merging neural networks in weight space. Differently from most existing works, …

Few-shot graph classification is a novel yet promising emerging research field that still lacks the soundness of well-established …