Automatic Gating of Attributes in Deep Structure

Authors: Xiaoming Jin, Tao He, Cheng Wan, Lan Yi, Guiguang Ding, Dou Shen

IJCAI 2018 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experimental results on a manually-labeled subset of Image Net, a-Yahoo and a-Pascal data set justify the superiority of AG-DBN against several baselines including CNN model and other AG-DBN variants. Specifically, it outperforms the CNN model, VGG19, by significantly reducing the classification error from 26.70% to 13.56% on a-Pascal.
Researcher Affiliation Collaboration Xiaoming Jin1 , Tao He1 , Cheng Wan2, Lan Yi3, Guiguang Ding1, Dou Shen2 1 School of Software, Tsinghua University, Beijing, China 2 Baidu Corporation, Beijing, China 3 Department of Dev Net, Cisco Systems
Pseudocode Yes Algorithm 1 Overall Learning Algorithm of AG-DBN ... Algorithm 2 Learning the parameters of l-th RBM with given gate matrix
Open Source Code No The paper provides a link for a dataset (1https://github.com/Tsinghua-IDE/a-Image Net) but no explicit statement or link to the open-source code for the AG-DBN methodology itself.
Open Datasets Yes a-Image Net Image Net [Russakovsky et al., 2015] is a widely used image data set. ... 1https://github.com/Tsinghua-IDE/a-Image Net ... a-Yahoo data set is collected by [Farhadi et al., 2009] from the Yahoo image search. ... a-Pascal (this data set is also collected by [Farhadi et al., 2009], and attributes in a-Pascal is the same as a-Yahoo).
Dataset Splits No Validation data are used to help find the best G in N0 rounds. (The paper mentions validation data but does not provide specific split percentages or counts for reproducibility.)
Hardware Specification No The paper does not provide specific hardware details (e.g., GPU models, CPU types, or memory specifications) used for running the experiments.
Software Dependencies No The paper does not list any specific software dependencies with version numbers (e.g., Python, TensorFlow, or PyTorch versions).
Experiment Setup Yes Models all consist of 2 hidden layers of size 500 and 300, and an input layer accepting 50*50 gray scale raw image pixels. ... N0 is set to be 30 ... The batch size s is finally set to 50. ... All experimental results are averaged over 5 independent runs. ... 2 hidden layers are both of size 4096 to make the comparison fare.