Sparsifying Neural Network Connections for Face Recognition

By Yi Sun, Xiaogang Wang, Xiaoou Tang. 2015

145 citations recorded by the source corpus.

This paper proposes to learn high-performance deep ConvNets with sparse neural connections, referred to as sparse ConvNets, for face recognition. The sparse ConvNets are learned in an iterative way, each time one additional layer is sparsified and the entire model is re-trained given the initial weights learned in previous iterations. One important finding is that directly training the sparse ConvNet from scratch failed to find good solutions for face recognition, while using a previously learned denser model to properly initialize a sparser model is critical to continue learning effective features for face recognition. This paper also proposes a new neural correlation-based weight selection criterion and empirically verifies its effectiveness in selecting informative connections from previously learned models in each iteration. When taking a moderately sparse structure (26%-76% of weights in the dense model), the proposed sparse ConvNet model significantly improves the face recognition performance of the previous state-of-the-art DeepID2+ models given the same training data, while it keeps the performance of the baseline model with only 12% of the original parameters.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Computer Vision and Pattern Recognition

Citations in the corpus

Continual Learning via Neural Pruning (2019, background): “the compression of neural network with relatively little loss of performance [7, 8] .”
SphereFace: Deep Hypersphere Embedding for Face Recognition (2017, methodology): “Most recent approaches [25, 28, 34] combine Euclidean margin based losses with softmax loss to construct a joint supervision.”
Selfless Sequential Learning (2018, methodology): “Other works iterate between pruning and retraining of a neural network as a post processing step (Liu et al., 2015; Sun et al., 2016; Aghasi et al., 2017; Louizos et al., 2017).”
Learning to Prune Deep Neural Networks via Layer-wise Optimal Brain Surgeon (2017, methodology): “Recent work mainly focuses on developing efﬁcient algorithms to obtain a near-optimal pruning solution [7, 8, 9, 10, 11].”
Continual Learning in Neural Networks (2019, methodology): “Other works iterate between pruning and retraining of a neural network as a post processing step [88, 149, 2, 93].”
From Facial Expression Recognition to Interpersonal Relation Prediction (2016, background): “2016), face recognition (Schroff et al. 2015; Parkhi et al. 2015; Sun et al. 2016), and face clustering (Zhang et al.”
SphereFace Revived: Unifying Hyperspherical Face Recognition (2021, methodology): “Prior to hyperspherical FR, well-performing deep FR methods [13], [15], [16] were mostly built on either contrastive loss or triplet loss, validating the importance of solving open-set FR as a metric learning problem [2].”
Privacy-Preserving Visual Learning Using Doubly Permuted Homomorphic Encryption (2017, background): “Another promising direction for the future work is learning much higher-dimensional models like sparse convolutional neural networks [6, 53].”