VSE-ens: Visual-Semantic Embeddings with Efficient Negative Sampling

Authors: Guibing Guo, Songlin Zhai, Fajie Yuan, Yuan Liu, Xingwei Wang

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

Reproducibility Variable Result LLM Response
Research Type Experimental The experiments demonstrate that our approach converges 5.02x faster than the state-of-the-art approaches on Open Images, 2.5x on IAPR-TCI2 and 2.06x on NUS-WIDE datasets, as well as better ranking accuracy across datasets.
Researcher Affiliation Academia Guibing Guo, Songlin Zhai Fajie Yuan, Yuan Liu, Xingwei Wang Northeastern University, China University of Glasgow, UK
Pseudocode Yes Algorithm 1 sketches the pseudocodes of the improved learning algorithm.
Open Source Code No The paper provides URLs for the datasets used (Open Images, IAPR-TC12, NUS-WIDE), but it does not provide any links or explicit statements about the availability of the authors' own source code for the proposed methodology.
Open Datasets Yes Three real datasets are used in our evaluation, namely Open Images1, IAPR-TC122, NUS-WIDE3... 1https://github.com/openimages/dataset 2http://www.imageclef.org/photodata 3lms.comp.nus.edu.sg/research/NUS-WIDE.htm
Dataset Splits No The paper mentions tuning hyperparameters: 'Parameter λ for VSE-ens is tuned from 0.001 to 1.0 to find the best value. The learning rate and regularization settings of other models are tuned from 0.001 to 0.1 to search the optimal values.' However, it describes a 'leave-one-out evaluation protocol' where 'an annotation from each image for evaluation, and leave the rest for training', which primarily defines a training and testing split, without explicitly stating a separate validation split or how it was derived for hyperparameter tuning.
Hardware Specification No The paper does not provide any specific details about the hardware used for running the experiments (e.g., GPU models, CPU types, memory).
Software Dependencies No The paper does not specify any software dependencies with version numbers (e.g., programming languages, libraries, frameworks).
Experiment Setup Yes The hyperparameters for VSE-ens on all three datasets are: learning rate η = 0.01, regularization = 0.01, and variables are initialized by a normal distribution N(0, 0.01). Parameter λ for VSE-ens is tuned from 0.001 to 1.0 to find the best value. The learning rate and regularization settings of other models are tuned from 0.001 to 0.1 to search the optimal values.