search

hjf-hjf / TASPH

Two-stage Asymmetric Similarity Preserving Hashing for Cross-modal Retrieval (TASPH)
https://ieeexplore.ieee.org/document/10146484
3 stars 1 forks source link

Please advise on "An efficient iterative optimization algorithm" in the paper #1

Open Daydaylight opened 1 year ago

Daydaylight commented 1 year ago

Hi, thank you very much for sharing out your work and giving me the opportunity to learn more about your article.

I think your proposed method opens up a new direction for solving discrete constrained problems, but due to my level, I am unable to understand the iterative optimization algorithm you proposed. Could you please give me some guidance? I am not sure what material I should read to help me understand this algorithm of yours.

Thank you for checking it out and I look forward to your reply!

hjf-hjf commented 1 year ago

Thanks for your attention to our work. Actually, the core part of our iterative optimization algorithm is not proposed by us, but rather inspired by the papers ''Supervised Hashing with Latent Factor Models'' published in 2014'SIGIR and ''Discrete Latent Factor Model for Cross-Modal Hashing'' published in 2019'TIP. It develops a column-by-column optimization strategy to iteratively learn each bit of hash codes. More specifically, it construct a lower bound of the objective and optimize the lower bound by maximizing it, and the adopted stochastic learning strategy can reduce time complexity . It's difficult to explain it briefly and you can refer to these two papers for more details.