LinkedIn's ESP partitions embeddings into per-objective subspaces, tunable at serving time
_reachsumit · x · 2026-09-28
LinkedIn researchers propose Embedding Subspace Partitioning (ESP) for multi-objective retrieval. Standard bi-encoders collapse relevance, engagement, and revenue signals into one static embedding, so trade-offs can't shift without retraining.
ESP decomposes the embedding into task-aware subspaces scored via a weighted sum, with weights tunable at serving time. It uses the model's end-of-sequence token as a delimiter with segment-aware attention masking and position-encoding resets to keep subspaces isolated in one forward pass, and serves via GPU-accelerated exhaustive kNN over a single concatenated index—no per-objective ANN infrastructure needed.
More from Research
- Reversible computing veteran poses open problem: universal BARCS primitives from unbiased Josephson junctions — MikePFrank · 2026-10-23
- Arc Institute uses genomic language model to map 376M prokaryotic noncoding sequences into 5.14M clusters — tedgreenwald · 2026-10-23
- The rise of AI-assisted peer review: reviewers are making it into the appendix — abursuc · 2026-10-23
- Anthropic uses Claude to formally verify security properties in 100+ open-source libraries at RISC-V assembly level — AnthropicAI · 2026-10-23
- Agentic interviews filter misaligned models, keeping LLM teams 81-83% clean — shangbinfeng · 2026-10-23
- Researchers test 9 strategies to pick LLM teams, beating random selection by up to 36.1% — shangbinfeng · 2026-10-23