Research map / Architectures and efficiency / RNNs and normalization
Structured Embedding Models for Grouped Data
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 |
|---|---|---|---|
| Replace one global embedding model with group-specific embeddings and shared context vectors | EFE · ArXiv papers | held-out pseudo log-likelihood: -2.176± 0.005 -> -2.287± 0.007 | Hurt |
| Replace one global embedding model with group-specific embeddings and shared context vectors | EFE · Senate speeches | held-out pseudo log-likelihood: -2.239 ± 0.002 -> -2.645± 0.002 | Hurt |
| Replace one global embedding model with group-specific embeddings and shared context vectors | EFE · Shopping data | held-out pseudo log-likelihood: -0.772± 0.000 -> -0.770± 0.001 | Helped |
| Add a hierarchical prior tying group embeddings to global embeddings (ablation) | SEFE · ArXiv papers | held-out pseudo log-likelihood: -2.287± 0.007 -> -2.170± 0.003 | Helped |
| Add a hierarchical prior tying group embeddings to global embeddings (ablation) | SEFE · Senate speeches | held-out pseudo log-likelihood: -2.645± 0.002 -> -2.217 ± 0.001 | Helped |
| Add a hierarchical prior tying group embeddings to global embeddings (ablation) | SEFE · Shopping data | held-out pseudo log-likelihood: -0.770± 0.001 -> -0.767± 0.000 | Helped |
| Use feed-forward amortization to generate group embeddings (ablation) | SEFE · ArXiv papers | held-out pseudo log-likelihood: -2.287± 0.007 -> -2.153± 0.004 | Helped |
| Use feed-forward amortization to generate group embeddings (ablation) | SEFE · Senate speeches | held-out pseudo log-likelihood: -2.645± 0.002 -> -2.484± 0.002 | Helped |
| Use feed-forward amortization to generate group embeddings (ablation) | SEFE · Shopping data | held-out pseudo log-likelihood: -0.770± 0.001 -> -0.774± 0.000 | Hurt |
| Use residual-network rather than feed-forward amortization (ablation) | Amortized SEFE · ArXiv papers | held-out pseudo log-likelihood: -2.153± 0.004 -> -2.120± 0.004 | Helped |
| Use residual-network rather than feed-forward amortization (ablation) | Amortized SEFE · Senate speeches | held-out pseudo log-likelihood: -2.484± 0.002 -> -2.249± 0.002 | Helped |
| Use residual-network rather than feed-forward amortization (ablation) | Amortized SEFE · Shopping data | held-out pseudo log-likelihood: -0.774± 0.000 -> -0.762± 0.000 | Helped |
Closest papers by meaning
- Exponential Family Embeddings (NeurIPS 2016 · similarity 0.91)
- By Tying Embeddings You Are Assuming the Distributional Hypothesis (ICML 2024 · similarity 0.85)
- On the Emergence of Linear Analogies in Word Embeddings (NeurIPS 2025 · similarity 0.85)
- The Scandinavian Embedding Benchmarks: Comprehensive Assessment of Multilingual and Monolingual Text Embedding (NeurIPS 2024 · similarity 0.85)
- MMTEB: Massive Multilingual Text Embedding Benchmark (ICLR 2025 · similarity 0.84)
- Demystifying Embedding Spaces using Large Language Models (ICLR 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.