Inter-Class Angular Loss for Convolutional Neural Networks
Authors: Le Hui, Xiang Li, Chen Gong, Meng Fang, Joey Tianyi Zhou, Jian Yang3894-3901
AAAI 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Thorough experimental results on a series of vision and nonvision datasets confirm that ICAL critically improves the discriminative ability of various representative deep neural networks and generates superior performance to the original networks with conventional softmax loss. |
| Researcher Affiliation | Collaboration | PCA Lab, Key Lab of Intelligent Perception and Systems for High-Dimensional Information of Ministry of Education Jiangsu Key Lab of Image and Video Understanding for Social Security School of Computer Science and Engineering, Nanjing University of Science and Technology, Nanjing, China Tencent AI Lab, Shenzhen, China Institute of High Performance Computing , Singapore {le.hui, xiang.li.implus, chen.gong, csjyang}@njust.edu.cn, mfang@tencent.com, joey.tianyi.zhou@gmail.com |
| Pseudocode | No | The paper does not contain structured pseudocode or algorithm blocks. |
| Open Source Code | No | The paper does not provide concrete access to source code for the methodology described in this paper. A footnote links to word2vec, which is a third-party tool. |
| Open Datasets | Yes | Three challenging image classification datasets are used for the experiments, which include Fashion-MNIST (Xiao, Rasul, and Vollgraf 2017) is a dataset of clothes images... CIFAR (Krizhevsky and Hinton 2009) dataset consists of totally 60K colored natural scene images... Five popular text classification datasets are utilized for the experiments, which contain MR Movie Reviews dataset (Pang and Lee 2005)... CR Customer Reviews (Hu and Liu 2004)... Subj Subjectivity (Pang and Lee 2004)... MPQA Opinion polarity (Wiebe, Wilson, and Cardie 2005)... TREC (Socher et al. 2013). |
| Dataset Splits | Yes | CIFAR (Krizhevsky and Hinton 2009) dataset consists of totally 60K colored natural scene images with the resolution of 32 × 32. The training set and test set contain 50K images and 10K images, respectively. [...] For the datasets without providing a standard test set, we randomly select 10% of the data as the test set and conduct 10-fold cross-validation for all the compared methods. |
| Hardware Specification | No | The paper does not provide specific hardware details (e.g., GPU/CPU models, memory) used for running its experiments. |
| Software Dependencies | No | The paper mentions 'Caffe and Tensorflow' (in discussion of L-Softmax backpropagation) and 'Adadelta' optimizer, and 'word2vec', but does not provide specific version numbers for these or other software dependencies relevant to their implementation. |
| Experiment Setup | Yes | All the networks are trained using SGD. The initial learning rate is set to 0.1, and is divided by 10 at 50% and 75% of the pre-set total number of training epochs. Following (Huang et al. 2017), we use a weight decay of 10^-4 and momentum of 0.9 with dampening. The experimental settings for Fashion-MNIST are identical to those for CIFAR-10 dataset. The parameter λ for incorporating our ICAL is set to λ = 1.0 on CIFAR datasets, and λ = 1.5 on F-MNIST dataset. [...] For CNNs, we use: filter windows (h) of 3,4,5 with 150 feature maps for each window, 0.5 for dropout rate (p), and 50 for the min-batch size. |