Research map / Architectures and efficiency / RNNs and normalization
AdANNS: A Framework for Adaptive Semantic Search
Aniket Rege, Aditya Kusupati, Sharan Ranjit S, Alan Fan, Qingqing Cao, Sham M. Kakade, Prateek Jain, Ali Farhadi
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 |
|---|---|---|---|
| decouple clustering dimension d_c from scan dimension d_s using matryoshka representations | IVF · ImageNet-1K retrieval | top-1 accuracy: 68.51 -> 70.02 | Improved |
| use AdANNS-IVF instead of rigid IVF at matched compute | IVF · ImageNet-1K retrieval | compute: up to 100x lesser FLOPS/query | Improved |
| matryoshka instead of rigid representation for IVF | IVF · ImageNet-1K | accuracy-compute trade-off | Improved |
| post-hoc SVD compression of rigid representation instead of matryoshka | IVF/OPQ · ImageNet-1K | top-1 accuracy | Hurt |
| AdANNS-OPQ with matryoshka instead of OPQ on rigid | OPQ · ImageNet-1K | top-1 accuracy: up to 4% | Improved |
| AdANNS-OPQ at 32 bytes vs rigid OPQ at 64 bytes | OPQ · ImageNet-1K | storage and FLOPS: 64 bytes -> 32 bytes | Improved |
| AdANNS-IVFOPQ composite index | IVFOPQ · ImageNet-1K | top-1 accuracy: 1-4% | Improved |
| AdANNS-DiskANN with 8-byte OPQ | DiskANN · ImageNet-1K | cost: 32 -> 8 bytes | Improved |
| replace PQ with OPQ at large dimension | exact search · ImageNet-1K d=2048, m=8 | top-1 accuracy: 62.19 -> 66.89 | Improved |
| replace PQ with OPQ at small dimension | exact search · ImageNet-1K d=64, m=8 | top-1 accuracy: 68.39 -> 68.38 | No clear effect |
| use low-dimensional matryoshka prefix for scan on index built with 2048-d (AdANNS-IVF-D) | IVF · ImageNetV2, d=8 | top-1 accuracy: 50.44 -> 53.51 | Improved |
| AdANNS-IVF-D at high dimension | IVF · ImageNetV2, d=2048 | top-1 accuracy: 58.04 -> 58.04 | No clear effect |
Closest papers by meaning
- NUDGE: Lightweight Non-Parametric Fine-Tuning of Embeddings for Retrieval (ICLR 2025 · similarity 0.86)
- Exploiting Distribution Constraints for Scalable and Efficient Image Retrieval (ICLR 2025 · similarity 0.85)
- Adaptive Retrieval and Scalable Indexing for k-NN Search with Cross-Encoders (ICLR 2024 · similarity 0.85)
- AdaVideoRAG: Omni-Contextual Adaptive Retrieval-Augmented Efficient Long Video Understanding (NeurIPS 2025 · similarity 0.85)
- Semi-Parametric Retrieval via Binary Bag-of-Tokens Index (ICLR 2025 · similarity 0.84)
- einspace: Searching for Neural Architectures from Fundamental Operations (NeurIPS 2024 · similarity 0.84)
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.
