Hashing with Uncertainty Quantification via Sampling-based Hypothesis Testing

Yucheng Wang, Mingyuan Zhou, Xiaoning Qian

Abstract:

To quantify different types of uncertainty when deriving hash-codes for image retrieval, we develop a probabilistic hashing model(ProbHash).
Sampling-based hypothesis testing is then derived for hashing with uncertainty quantification(HashUQ) in ProbHash to improve the granularity of hashing-based retrieval by prioritizing the data with confident hash-codes. HashUQ can drastically improve the retrieval performance without sacrificing computational efficiency. For efficient deployment of HashUQ in real-world applications, we discretize the quantified uncertainty to reduce the potential storage overhead. Experimental results show that our HashUQ can achieve state-of-the-art retrieval performance on three image datasets. Ablation experiments on model hyperparameters, different model components, and effects of UQ are also provided with performance comparisons. Our code is available at https://github.com/QianLab/HashUQ.

Submission Length: Regular submission (no more than 12 pages of main content)

Changes Since Last Submission:

1. We discuss about the criterion for selecting likelihood function in *Appendix A.3*. As most of the commonly adopted likelihood function cannot be used in our proposed framework, as we explain in the Author Response to Reviewer 3tbV, we instead experimentally compare our adopted likelihood with another Boltzmann distribution which takes a similar format as Gaussian distribution.
2. Some minor revisions on grammar and format.

Code: https://github.com/QianLab/HashUQ

Assigned Action Editor: ~Shiyu_Chang2

Submission Number: 2963