On the Quality of the Initial Basin in Overspecified Neural Networks

Authors: Itay Safran, Ohad Shamir

ICML 2016 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Theoretical However, a theoretical explanation for this remains a major open problem, since training neural networks involves optimizing a highly non-convex objective function, and is known to be computationally hard in the worst case. In this work, we study the geometric structure of the associated non-convex objective function, in the context of Re LU networks and starting from a random initialization of the network parameters. Before continuing, we emphasize that our observations are purely geometric in nature, independent of any particular optimization procedure.
Researcher Affiliation Academia Itay Safran ITAY.SAFRAN@WEIZMANN.AC.IL Ohad Shamir OHAD.SHAMIR@WEIZMANN.AC.IL Weizmann Institute of Science, Israel
Pseudocode No The paper does not contain any structured pseudocode or algorithm blocks.
Open Source Code No The paper does not provide any concrete access information for source code.
Open Datasets No The paper refers to 'data' or 'training data' using generic symbols like S = (xt, yt)m t=1, but does not provide any concrete access information, such as specific names of public datasets, links, DOIs, or formal citations.
Dataset Splits No The paper does not provide specific dataset split information for training, validation, or testing.
Hardware Specification No The paper does not explicitly describe the hardware used to run experiments, as it is a theoretical paper.
Software Dependencies No The paper does not provide specific ancillary software details with version numbers.
Experiment Setup No The paper does not contain specific experimental setup details such as hyperparameter values or training configurations.