ACORN

Advanced vector quantization technique that explores the Leech lattice's optimal sphere packing properties at 24 dimensions. Delivers state-of-the-art LLM quantization performance, outperforming recent methods like Quip#, QTIP, and PVQ for extreme vector compression.

Sparse Late Interaction Retrieval model that combines the benefits of sparse representations with late interaction mechanisms. Provides efficient storage and fast retrieval while maintaining the accuracy advantages of token-level matching in sparse embedding space.

Distributed disk-based approximate nearest neighbor system achieving near-linear throughput scaling. Delivers 6.21-6.49x throughput improvement over scatter-gather baseline with sub-6ms latency on 10 servers.

Manifold-Consistent Graph Indexing for billion-scale disk-resident vector search. Leverages Local Intrinsic Dimensionality to achieve 5.8x throughput improvement over DiskANN on high-dimensional datasets.

Efficient multi-vector retrieval system using sparsified late interaction with inverted indexes. Achieves 40% less storage and 83% lower latency than ColBERT-v2 while maintaining competitive accuracy.

Incremental in-place update system for billion-scale vector search from Microsoft Research. Maintains 2.41x lower P99.9 latency than baselines while supporting efficient vector updates with minimal resource overhead.