Research map / Learning theory and algorithms / Combinatorial and search optimization
Deep Hashing via Discrepancy Minimization
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use discrepancy-minimizing series expansion instead of HashNet | deep hashing with AlexNet · CIFAR-10, 64 bits | MAP: 0.7187 -> 0.7373 | Helped |
| use discrepancy-minimizing series expansion instead of HashNet | deep hashing with AlexNet · NUS-WIDE, 64 bits | MAP@5000: 0.7762 -> 0.7892 | Helped |
| use discrepancy-minimizing series expansion instead of HashNet | deep hashing with AlexNet · ImageNet, 32 bits | MAP@1000: 0.6219 -> 0.6123 | Hurt |
| add higher-order discrepancy terms rather than using only the first-order term (ablation) | DMDH · CIFAR-10, 16, 32, 48, and 64 bits | MAP: DMDH-L: 0.6705, 0.6955, 0.7044, 0.6975; DMDH: 0.7037, 0.7191, 0.7319, 0.7373 | Helped |
| gradually increase objective coefficients rather than fixing them (ablation) | DMDH · CIFAR-10, 16, 32, 48, and 64 bits | MAP: DMDH-F: 0.6850, 0.7027, 0.7108, 0.7128; DMDH: 0.7037, 0.7191, 0.7319, 0.7373 | Helped |
| add higher-order discrepancy terms and gradually increase their coefficients rather than using only a fixed first-order term (ablation) | DMDH · CIFAR-10, 16, 32, 48, and 64 bits | MAP: DMDH-LF: 0.6670, 0.6816, 0.6845, 0.6641; DMDH: 0.7037, 0.7191, 0.7319, 0.7373 | Helped |
| use asymmetric weighted similarity measurement controlled by β instead of symmetric weighting | DMDH · CIFAR-10, 64 bits | MAP | Helped |
Closest papers by meaning
- Deep Supervised Discrete Hashing (NeurIPS 2017 · similarity 0.87)
- Optimizing affinity-based binary hashing using auxiliary coordinates (NeurIPS 2016 · similarity 0.87)
- Greedy Hash: Towards Fast Optimization for Accurate Hash Coding in CNN (NeurIPS 2018 · similarity 0.86)
- One Loss for All: Deep Hashing with a Single Cosine Similarity based Learning Objective (NeurIPS 2021 · similarity 0.86)
- KALAHash: Knowledge-Anchored Low-Resource Adaptation for Deep Hashing (AAAI 2025 · similarity 0.84)
- Deep Unsupervised Hashing via External Guidance (ICML 2025 · similarity 0.83)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.