RePr
A well-trained Convolutional Neural Network can easily be pruned without significant loss of performance. This is because of unnecessary overlap in the features captured by the network’s filters. Innovations in network architecture such as skip/dense connections and Inception units have mitigated this problem to some extent, but these improvements come with increased computation and memory requirements at run-time. We attempt to address this problem from another angle – not by changing the network structure but by altering the training method. We show that by temporarily pruning and then restoring a subset of the model’s filters, and repeating this process cyclically, overlap in the learned features is reduced, producing improved generalization. We show that the existing model-pruning criteria are not optimal for selecting filters to prune in this context and introduce inter-filter orthogonality as the ranking criteria to determine under-expressive filters. Our method is applicable both to vanilla convolutional networks and more complex modern architectures, and improves the performance across a variety of tasks, especially when applied to smaller networks. …

Graph Convolution Embedded Long Short Term Memory Network (GC-LSTM)
Dynamic link prediction is a research hot in complex networks area, especially for its wide applications in biology, social network, economy and industry. Compared with static link prediction, dynamic one is much more difficult since network structure evolves over time. Currently most researches focus on static link prediction which cannot achieve expected performance in dynamic network. Aiming at low AUC, high Error Rate, add/remove link prediction difficulty, we propose GC-LSTM, a Graph Convolution Network (GC) embedded Long Short Term Memory network (LTSM), for end-to-end dynamic link prediction. To the best of our knowledge, it is the first time that GCN embedded LSTM is put forward for link prediction of dynamic networks. GCN in this new deep model is capable of node structure learning of network snapshot for each time slide, while LSTM is responsible for temporal feature learning for network snapshot. Besides, current dynamic link prediction method can only handle removed links, GC-LSTM can predict both added or removed link at the same time. Extensive experiments are carried out to testify its performance in aspects of prediction accuracy, Error Rate, add/remove link prediction and key link prediction. The results prove that GC-LSTM outperforms current state-of-art method. …

Binarized Deep Neural Network (BDNN)
In this work we introduce a binarized deep neural network (BDNN) model. BDNNs are trained using a novel binarized back propagation algorithm (BBP), which uses binary weights and binary neurons during the forward and backward propagation, while retaining precision of the stored weights in which gradients are accumulated. At test phase, BDNNs are fully binarized and can be implemented in hardware with low circuit complexity. The proposed binarized networks can be implemented using binary convolutions and proxy matrix multiplications with only standard binary XNOR and population count (popcount) operations. BBP is expected to reduce energy consumption by at least two orders of magnitude when compared to the hardware implementation of existing training algorithms. We obtained near state-of-the-art results with BDNNs on the permutation-invariant MNIST, CIFAR-10 and SVHN datasets. …

FM-Pair
In this work, we propose FM-Pair, an adaptation of Factorization Machines with a pairwise loss function, making them effective for datasets with implicit feedback. The optimization model in FM-Pair is based on the BPR (Bayesian Personalized Ranking) criterion, which is a well-established pairwise optimization model. FM-Pair retains the advantages of FMs on generality, expressiveness and performance and yet it can be used for datasets with implicit feedback. We also propose how to apply FM-Pair effectively on two collaborative filtering problems, namely, context-aware recommendation and cross-domain collaborative filtering. By performing experiments on different datasets with explicit or implicit feedback we empirically show that in most of the tested datasets, FM-Pair beats state-of-the-art learning-to-rank methods such as BPR-MF (BPR with Matrix Factorization model). We also show that FM-Pair is significantly more effective for ranking, compared to the standard FMs model. Moreover, we show that FM-Pair can utilize context or cross-domain information effectively as the accuracy of recommendations would always improve with the right auxiliary features. Finally we show that FM-Pair has a linear time complexity and scales linearly by exploiting additional features. …