Stochastic Generative Hashing

Authors: Bo Dai, Ruiqi Guo, Sanjiv Kumar, Niao He, Le Song

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

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive experiments on a variety of large-scale datasets show that the proposed method achieves better retrieval results than the existing state-of-the-art methods.
Researcher Affiliation Collaboration Bo Dai* 1 Ruiqi Guo* 2 Sanjiv Kumar 2 Niao He 3 Le Song 1 1Georgia Institute of Technology. 2Google Research, NYC 3University of Illinois at Urbana-Champaign.
Pseudocode Yes Algorithm 1 Distributional-SGD
Open Source Code Yes Our implementation of stochastic generative hashing as well as the whole training procedure was done in Tensor Flow. We have released our code on Git Hub2. For the competing methods, we di- rectly used the code released by the authors. 2https://github.com/doubling/StochasticGenerativeHashing
Open Datasets Yes We used several benchmarks datasets, i.e., (1) MNIST which contains 60,000 digit images of size 28 28 pixels, (2) CIFAR-10 which contains 60,000 32 32 pixel color images in 10 classes, (3) SIFT-1M and (4) SIFT-1B which contain 106 and 109 samples, each of which is a 128 dimensional vector, and (5) GIST-1M which contains 106 samples, each of which is a 960 dimensional vector.
Dataset Splits No The paper mentions training data and datasets, but does not explicitly state the specific training, validation, and test dataset splits (e.g., percentages, sample counts, or predefined split references) required for reproduction in the main text.
Hardware Specification Yes Our experiments were run on AMD 2.4GHz Opteron CPUs 4 and 32G memory.
Software Dependencies No The paper states, 'Our implementation of stochastic generative hashing as well as the whole training procedure was done in Tensor Flow.', but does not specify a version number for TensorFlow or any other software dependencies.
Experiment Setup Yes where we fix hyperparameters for all our experiments and used a batch size of 500 and learning rate of 0.01 with stepsize decay.