Zero-shot Learning via Simultaneous Generating and Learning

Authors: Hyeonwoo Yu, Beomhee Lee

NeurIPS 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental In experimental results, we demonstrate that the proposed generating and learning strategy makes the model achieve the outperforming results compared to that trained only on the seen classes, and also to the several state-of-the-art methods.
Researcher Affiliation Academia Hyeonwoo Yu Beomhee Lee Automation and Systems Research Institute (ASRI) Dept. of Electrical and Computer Engineering Seoul National University {bgus2000,bhlee}@snu.ac.kr
Pseudocode Yes Algorithm 1 Simultaneously Generating-And-Learning Algorithm
Open Source Code No The paper does not provide any explicit statements or links indicating the availability of open-source code for the described methodology.
Open Datasets Yes We firstly use the two benchmark datasets: Aw A (Animals with Attributes) [16], which contains 30,745 images of 40/10(train/test) classes, and CUB (Caltech-UCSD Birds-200-2011) [26], comprised of 11,788 images of 150/50(train/test) species.
Dataset Splits Yes SUN is a scene-image dataset and consists of 14,340 images with 645/72/65(train/test/validation) classes.
Hardware Specification No The paper does not provide specific details about the hardware (e.g., CPU/GPU models, memory) used to run the experiments.
Software Dependencies No The paper mentions software like Inception-V2, ResNet101, and Adam, but does not specify version numbers for any software dependencies.
Experiment Setup Yes The number of iterations for the benchmarks are: for mm VAE and SGAL(EM), 170,000 and 1,300 for Aw A1, 64,000 and 900 for Aw A2, 17,000 and 2,000 for CUB1 and 1,450,000 and 1,500 for SUN1. In order to consider the model uncertainty, we also train the model adopting (9) when generating unseen datapoints. For one latent variables sampled from prior network, a total of 5 samples are generated while activating dropouts in the decoder. Unlike (9), in the actual implementation all the dropouts of the prior network are deactivated for the training stabilization.