Improving DCNN Performance with Sparse Category-Selective Objective Function
Authors: Shizhou Zhang, Yihong Gong, Jinjun Wang
IJCAI 2016 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | As experimental results show, when applying the proposed method to the Quick model and NIN models, image classification performances are remarkably improved on four widely used benchmark datasets: CIFAR-10, CIFAR-100, MNIST and SVHN, which demonstrate the effectiveness of the presented method. and 4 Experimental Evaluations To evaluate the effectiveness of the proposed SCSOF for improving object recognition performances of CNN models, we conduct experimental evaluations using shallow and deep models, respectively. |
| Researcher Affiliation | Academia | Shizhou Zhang, Yihong Gong , Jinjun Wang Institute of Artificial Intelligence and Robotics, Xi an Jiaotong University, Xi an, 710049, P.R.China |
| Pseudocode | No | The paper explains the formulation and implementation details of the Sparse Category-Selective Objective Function (SCSOF) in Section 3.2 and 3.3, but does not provide a formal pseudocode block or algorithm. |
| Open Source Code | No | The implementation is based on the Caffe [Jia et al., 2014] package. The model is available in the Caffe package. (This refers to a third-party package and a pre-existing model, not the authors' specific code for SCSOF). |
| Open Datasets | Yes | We evaluate the proposed SCSOF on four widely used benchmark datasets, namely CIFAR-10, CIFAR-100, MNIST and SVHN. The reason for choosing these datasets is because they contain a large amount of small images (about 32x32 pixels), so that models can be trained by using computers with moderate configurations within reasonable time frames. Because of this, the four datasets have become very popular choices for deep network performance evaluations in the computer vision and pattern recognition research communities. |
| Dataset Splits | Yes | CIFAR-10 Dataset. The CIFAR-10 dataset is composed of 10 classes of natural images, 50,000 for training and 10,000 for testing. CIFAR-100 Dataset. this dataset contains 50,000 images for training and 10,000 images for testing. MNIST Dataset. There are 60,000 training samples and 10,000 testing samples in total. SVHN Dataset. ...while the remaining 598,388 images of the training and the extra sets were used for training, which is also the same with that in [Min Lin, 2014; Goodfellow et al., 2013; Lee et al., 2015]. The validation set was only used for tuning hyper-parameters and was not used for training the model. |
| Hardware Specification | No | The paper does not provide specific hardware details like CPU or GPU models. It mentions "computers with moderate configurations" and "CPU/GPU clusters" but no specific specifications. |
| Software Dependencies | No | The implementation is based on the Caffe [Jia et al., 2014] package. (No version number is specified for Caffe or any other software dependencies). |
| Experiment Setup | Yes | For Quick model, the weight decay coefficient λ is set to 0.004, the momentum is set to 0.9. The initial learning rate is set to 0.001 and decreased by a factor of 10 for every 8,000 iterations. The training process is finished over 30,000 iterations. For NIN model, we strictly follow the settings as in [Min Lin, 2014] for each dataset. The only hyper-parameter β introduced by the proposed SCSOF is empirically set to [10-6, 10-4]. |