… The training loss function of these models has infinite … models to explain the generalization
of deep networks [4, 3, 2, 11, 20… on the output of a neural network in under-parameterized and …

… ) networks are popular in deep learning due to their ease of training and stateof-the-art
generalization… However for our purpose, we do not need to restrict l(.) to a specific choice, we only …

… the generalization measures are functions that are evaluated on a trained neural network and
… data points, we will refer to a set of neural networks trained on the same dataset as a task. …

… neural network generalization defies conventional explanations and requires new ones.
Neyshabur … NTK prediction function and apply even with infinite over-parametrization for both …

… order optimization to speed up training [Becker and LeCun… -convex function is an NP-complete
problem, and neural network … • We analyze the generalization properties of NGD, showing …

… neural networks. Because of this, we believe they will continue to play a transformative role
in … We quantified phenomena having to do with generalization, architecture dependendence, …

… on the generalization error of convolutional networks. The … to role of overparametrization
on generalization (Neyshabur … 2019). An explanation of this phenomenon that is consistent …

… wide deep neural networks with ReLU activation function are biased towards simple functions.
… [10] explores the generalization properties of deep neural networks trained on partially …

… Generalization for RL has recently grown to be an important … -linear function approximator
such as a neural network. On … in a synthetic environment and neural networks such as multi-…

… Consistent with the nature of the pruning algorithm playing a role in generalization, we …
generalization, 2019. [13] Zeyuan Allen-Zhu, Yuanzhi Li, and Yingyu Liang. Learning and …