LATIDIA · Investigación

CircuitTree: Composición de circuitos en un transcriptor de sílabas

arXiv:2610.12520v1 Announce Type: cross Resumen: El objetivo del descubrimiento de circuitos es descomponer una red neuronal monolítica en pequeñas subredes modulares que son interpretables en el sentido de que cada una hace una identificación

La noticia

arXiv:2610.12520v1 Announce Type: cross Abstract: The goal of circuit discovery is to decompose a monolithic neural network into tiny modular subnetworks which are interpretable in the sense that each does an identifiable subtask. We develop a practical method for composing arbitrarily many circuits into a single model. We propose a greedy, density-scaled composition algorithm that overcomes limitations of previous attempts to compose circuits by naive union, then generalize it to $N$ circuits via CircuitTree, a new data structure built on a $\Theta(N\log N)$ tree-compose schedule. We compose circuits pruned from HuBERT for classifying individual phonemes into a syllable classifier, a capability none of the 39 phoneme circuits has individually. Chaining