Adversarial Localized Energy Network for Structured Prediction

Authors: Pingbo Pan, Ping Liu, Yan Yan, Tianbao Yang, Yi Yang5347-5354

AAAI 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We conduct extensive experiments to verify the effectiveness and efficiency of our proposed method. ... We conduct experiments on different problems for the validation, including multi-label classification, binary image segmentation, and 3-class face segmentation tasks. The experimental results indicate that our proposed method can not only refine the final results to a higher stage even with a smaller input resolution, but also improve the convergence in the training and inference stages.
Researcher Affiliation Collaboration Pingbo Pan,1,2 Ping Liu,2 Yan Yan,3 Tianbao Yang,3 Yi Yang2 1Baidu Research, 2The Re LER Lab, University of Technology Sydney, 3University of Iowa
Pseudocode Yes Algorithm 1 ALEN training Input: training data C 1: while not converged do 2: (x, y ) C 3: calculate y according to Equation 4. 4: update θv according to Equation 6. 5: update θg according to Equation 11. 6: end while Output:energy network weight θv and inference network weight θg
Open Source Code No The paper states 'Our implementation is based on Tensorflow (Abadi et al. 2016). We also implement the proposed method with Paddle Paddle and achieve similar performance.' but does not provide a link or explicit statement about the public release of their source code.
Open Datasets Yes We use standard benchmarks of this task, namely Bibtex and Bookmarks, introduced by (Katakis, Tsoumakas, and Vlahavas 2008). ... We utilize the Labeled Faces in the Wild (LFW) dataset (Huang et al. 2007) to evaluate our framework on 3-class face segmentation. ... Following the work of DVN (Gygli, Norouzi, and Angelova 2017), we utilize the Weizmann horses dataset (Borenstein and Ullman 2004) to compare the performance on binary image segmentation.
Dataset Splits Yes We follow the same training, validation, and testing splits proposed in (Kae et al. 2013; Tsogkas et al. 2015; Gygli, Norouzi, and Angelova 2017) and utilize the same network architecture and data augmentation strategy as them.
Hardware Specification No The paper does not provide specific hardware details (e.g., GPU/CPU models, memory, or processing units) used for running its experiments.
Software Dependencies No The paper mentions that its implementation is based on TensorFlow and PaddlePaddle and uses the Adam optimizer, but it does not specify version numbers for these software components.
Experiment Setup No The paper states that hyperparameters were found via grid search and mentions individually-tuned learning rates and network architectures, but it does not provide specific hyperparameter values (e.g., learning rate, batch size, number of epochs) or detailed training configurations in the main text.