English
Related papers

Related papers: On weight initialization in deep neural networks

200 papers

In this work, we propose to train a deep neural network by distributed optimization over a graph. Two nonlinear functions are considered: the rectified linear unit (ReLU) and a linear unit with both lower and upper cutoffs (DCutLU). The…

Machine Learning · Computer Science 2017-06-20 Guoqiang Zhang , W. Bastiaan Kleijn

Next generation deep neural networks for classification hosted on embedded platforms will rely on fast, efficient, and accurate learning algorithms. Initialization of weights in learning networks has a great impact on the classification…

Machine Learning · Computer Science 2016-07-21 Julius , Gopinath Mahale , Sumana T. , C. S. Adityakrishna

Empirical works show that for ReLU neural networks (NNs) with small initialization, input weights of hidden neurons (the input weight of a hidden neuron consists of the weight from its input layer to the hidden neuron and its bias term)…

Machine Learning · Computer Science 2022-10-20 Hanxu Zhou , Qixuan Zhou , Tao Luo , Yaoyu Zhang , Zhi-Qin John Xu

We present a simple linear regression based approach for learning the weights and biases of a neural network, as an alternative to standard gradient based backpropagation. The present work is exploratory in nature, and we restrict the…

Machine Learning · Computer Science 2023-07-17 Harshad Khadilkar

Motivated by the growing theoretical understanding of neural networks that employ the Rectified Linear Unit (ReLU) as their activation function, we revisit the use of ReLU activation functions for learning implicit neural representations…

Image and Video Processing · Electrical Eng. & Systems 2024-08-05 Joseph Shenouda , Yamin Zhou , Robert D. Nowak

Recently, neural networks in machine learning use rectified linear units (ReLUs) in early processing layers for better performance. Training these structures sometimes results in "dying ReLU units" with near-zero outputs. We first explore…

Machine Learning · Computer Science 2018-12-17 Scott C. Douglas , Jiutian Yu

Using back-propagation and its variants to train deep networks is often problematic for new users. Issues such as exploding gradients, vanishing gradients, and high sensitivity to weight initialization strategies often make networks…

Machine Learning · Computer Science 2018-07-16 Alexander G. Ororbia , Ankur Mali , Daniel Kifer , C. Lee Giles

Benign overfitting refers to how over-parameterized neural networks can fit training data perfectly and generalize well to unseen data. While this has been widely investigated theoretically, existing works are limited to two-layer networks…

Machine Learning · Computer Science 2024-10-28 Shuning Shang , Xuran Meng , Yuan Cao , Difan Zou

We introduce stochastic activations. This novel strategy randomly selects between several non-linear functions in the feed-forward layer of a large language model. In particular, we choose between SILU or RELU depending on a Bernoulli draw.…

Training neural networks with first order optimisation methods is at the core of the empirical success of deep learning. The scale of initialisation is a crucial factor, as small initialisations are generally associated to a feature…

Machine Learning · Computer Science 2025-09-16 Etienne Boursier , Nicolas Flammarion

Convergence rate of training algorithms for neural networks is heavily affected by initialization of weights. In this paper, an original algorithm for initialization of weights in backpropagation neural net is presented with application to…

Machine Learning · Computer Science 2020-04-07 Nadir Murru , Rosaria Rossini

Learning long term dependencies in recurrent networks is difficult due to vanishing and exploding gradients. To overcome this difficulty, researchers have developed sophisticated optimization techniques and network architectures. In this…

Neural and Evolutionary Computing · Computer Science 2015-04-09 Quoc V. Le , Navdeep Jaitly , Geoffrey E. Hinton

We identify and study two common failure modes for early training in deep ReLU nets. For each we give a rigorous proof of when it occurs and how to avoid it, for fully connected and residual architectures. The first failure mode,…

Machine Learning · Statistics 2018-11-14 Boris Hanin , David Rolnick

Rectified activation units (rectifiers) are essential for state-of-the-art neural networks. In this work, we study rectifier neural networks for image classification from two aspects. First, we propose a Parametric Rectified Linear Unit…

Computer Vision and Pattern Recognition · Computer Science 2015-02-09 Kaiming He , Xiangyu Zhang , Shaoqing Ren , Jian Sun

Deep learning, in the form of artificial neural networks, has achieved remarkable practical success in recent years, for a variety of difficult machine learning applications. However, a theoretical explanation for this remains a major open…

Machine Learning · Computer Science 2016-06-15 Itay Safran , Ohad Shamir

This paper provides an analysis of state-of-the-art activation functions with respect to supervised classification of deep neural network. These activation functions comprise of Rectified Linear Units (ReLU), Exponential Linear Unit (ELU),…

Machine Learning · Computer Science 2021-04-07 Anh Nguyen , Khoa Pham , Dat Ngo , Thanh Ngo , Lam Pham

Amongst others, the adoption of Rectified Linear Units (ReLUs) is regarded as one of the ingredients of the success of deep learning. ReLU activation has been shown to mitigate the vanishing gradient issue, to encourage sparsity in the…

Machine Learning · Statistics 2021-10-14 Nicola Picchiotti , Marco Gori

Activation functions play a critical role in deep neural networks by shaping gradient flow, optimization stability, and generalization. While ReLU remains widely used due to its simplicity, it suffers from gradient sparsity and dead-neuron…

Machine Learning · Computer Science 2025-12-03 Ashkan Shakarami , Yousef Yeganeh , Azade Farshad , Lorenzo Nicolè , Stefano Ghidoni , Nassir Navab

We develop a general duality between neural networks and compositional kernels, striving towards a better understanding of deep learning. We show that initial representations generated by common random initializations are sufficiently rich…

Machine Learning · Computer Science 2017-05-23 Amit Daniely , Roy Frostig , Yoram Singer

Traditional initialisation methods, e.g. He and Xavier, have been effective in avoiding the problem of vanishing or exploding gradients in neural networks. However, they only use simple pointwise distributions, which model one-dimensional…

Computer Vision and Pattern Recognition · Computer Science 2023-10-26 Christian Harder , Moritz Fuchs , Yuri Tolkach , Anirban Mukhopadhyay