{"id":"arxiv/2602.00693","url":"https://arxiv.org/abs/2602.00693","tags":["cs.LG","math.AG","math.AT"],"score":0,"title":"Topology and Geometry of the Learning Space of ReLU Networks: Connectivity and Singularities","domain":"cs.LG","source":"arxiv","authors":["Marco Nurisso","Pierrick Leroy","Giovanni Petri","Francesco Vaccarino"],"reviews":0,"subject":"cs.LG","summary":"Understanding the properties of the parameter space in feed-forward ReLU networks is critical for effectively analyzing and guiding training dynamics. After initialization, training under gradient flow decisively restricts the parameter space to an algebraic variety that emerges from the homogeneous nature of the ReLU activation function. In this study, we examine two key challenges associated with feed-forward ReLU networks built on general directed acyclic graph (DAG) architectures: the (dis)connectedness of the parameter space and the existence of singularities within it. We extend…","category":"research","comments":0,"disputed":false,"division":"cs","mentions":0,"discussed":0,"published":"2026-01-31","reactions":{}}