Contextual RNN-GANs for Abstract Reasoning Diagram Generation

Authors: Viveka Kulharia, Arnab Ghosh, Amitabha Mukerjee, Vinay Namboodiri, Mohit Bansal

AAAI 2017 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We evaluate the Context-RNN-GAN model (and its variants) on a novel dataset of Diagrammatic Abstract Reasoning, where it performs competitively with 10th-grade human performance but there is still scope for interesting improvements as compared to college-grade human performance. We also evaluate our model on a standard video next-frame prediction task, achieving improved performance over comparable state-of-the-art.
Researcher Affiliation Academia Viveka Kulharia, 1 Arnab Ghosh, 1 Amitabha Mukerjee,1 Vinay Namboodiri,1 Mohit Bansal2 1IIT Kanpur 2UNC Chapel Hill
Pseudocode No The paper describes the architecture and training process of the Context-RNN-GAN model using mathematical formulations and textual descriptions, but it does not include any explicitly labeled pseudocode or algorithm blocks.
Open Source Code No The paper does not provide any explicit statements about making the source code available, nor does it include links to a code repository.
Open Datasets No We collected the data from several IQ test books and online resources (Aggarwal, 2016), (Sijwali and Sijwali, 2016), (Gupta, 2016) and (Jha and Siddiquee, 2011).
Dataset Splits Yes The models were validated (tuned) using the first 50% of the answer figure set and tested on the remaining unseen 50% of the answer figure set from the test/validation set of 100 questions.
Hardware Specification No The paper describes the model architectures and training procedures, but it does not specify any hardware details such as GPU models, CPU types, or memory used for the experiments.
Software Dependencies No The paper mentions using Adam Optimizer but does not specify software dependencies like programming languages (e.g., Python), deep learning frameworks (e.g., TensorFlow, PyTorch), or their specific version numbers.
Experiment Setup Yes The best (validated) hyperparameters for the Context-RNN-GAN was a GRU with two layers of 400 hidden units each for the generator and a GRU with a single layer and 500 hidden units for the discriminator... The final models of Context-RNN-GAN and the RNN-GAN models use λadv = 0.05, λp = 1, and they use p = 1 for the DAT-DAR task and p = 2 for the next-frame prediction task... In all of the above models a dropout of 0.50 was applied for each hidden layer. All of the models were trained using the Adam Optimizer (Kingma and Ba, 2014).