Deep MIML Network
Authors: Ji Feng, Zhi-Hua Zhou
AAAI 2017 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | The effectiveness of Deep MIML network is validated by experiments on various domains of data. |
| Researcher Affiliation | Academia | Ji Feng, Zhi-Hua Zhou National Key Laboratory for Novel Software Technology, Nanjing University, Nanjing 210023, China Collaborative Innovation Center of Novel Software Technology and Industrialization, Nanjing 210023, China {fengj, zhouzh}@lamda.nju.edu.cn |
| Pseudocode | No | The paper describes the network architecture and operations but does not provide any structured pseudocode or algorithm blocks. |
| Open Source Code | No | The paper states 'We implemented the model via Keras (keras.io)', which is a third-party library, but does not provide any link or explicit statement about releasing the source code for the Deep MIML network described in the paper. |
| Open Datasets | Yes | We performed experiments on 2016 Yelp dataset challenge 1. Each review belongs to one or more categories (such as restaurant , Thai Food ) and we extract 100 categories with the reviews been tagged with. ... For image task, we conducted our experiments on Microsoft COCO dataset (Lin et al. 2014). ... we used the benchmark dataset reported in these prior works by Zhou et. al (2012) among others, which has already been preprocessed using tf-idf (MIML News data) and SBN features (Wei and Zhou 2016) on patches (MIML Scene data).2 3 |
| Dataset Splits | No | We split the dataset into training and testing set, with the split ratio of 0.7. and During validating process, we found varying K does not affect the performance a lot and we report the result with K equal to 4. There is no clear explicit training/validation/test split with specific percentages or counts for a validation set. |
| Hardware Specification | Yes | We also used 2 Nvidia Titan-X GPU to speed up training time. |
| Software Dependencies | No | The paper mentions 'We implemented the model via Keras (keras.io)', but it does not specify the version number of Keras or any other software dependency. |
| Experiment Setup | Yes | The loss function we choose here is the mean binary cross-entropy. During training time, we use stochastic gradient descent with dropout. and We used mean binary cross-entropy as loss function and used SGD with dropout rate of 0.5. The only hyperparameters here is K, the number of sub-concepts. During validating process, we found varying K does not affect the performance a lot and we report the result with K equal to 4. |