Research map / Architectures and efficiency / RNNs and normalization
Near-Lossless Binarization of Word Embeddings
Julien Tissier, Christophe Gravier, Amaury Habrard
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 |
|---|---|---|---|
| binarize GloVe vectors to 256 bits | GloVe top-10 search · Top 10 query | execution time: 98.08 -> 3.25 | Improved |
| store binary rather than real-valued GloVe vectors | GloVe top-10 query including loading · Loading + Top 10 query | execution time: 23500 -> 310 | Improved |
| binarize fasttext vectors to 512 bits | fasttext · WS353 | Spearman correlation: 69.7 -> 70.3 | Improved |
| binarize GloVe vectors to 256 bits | GloVe · SimVerb | Spearman correlation: 22.7 -> 22.9 | Improved |
| binarize fasttext vectors to 512 bits | fasttext · DBpedia | accuracy: 95.0 -> 97.3 | Improved |
| binarize fasttext vectors to 512 bits | fasttext · Amazon Full | accuracy: 49.0 -> 49.8 | Improved |
| use the proposed encoder instead of naive binarization | dict2vec · Semantic similarity average, 64-bit | Fisher's average semantic score: 44.7 -> 46.5 | Improved |
| use the proposed encoder instead of naive binarization | dict2vec · Semantic similarity average, 128-bit | Fisher's average semantic score: 49.6 -> 54.1 | Improved |
| increase binary code size from 64 to 256 or 512 bits (ablation) | binarized embeddings · semantic similarity and classification tasks | task performance | Improved |
| remove the correlation regularization term (ablation) | autoencoder binarizer · training | convergence stability | Hurt |
| vary the regularization hyperparameter (ablation) | autoencoder binarizer · reported experiments | performance: ~2% variation | No clear effect |
| use binary instead of reconstructed vectors (ablation) | binarized embeddings · document classification and sentiment analysis | accuracy | No clear effect |
Closest papers by meaning
- All Word Embeddings from One Embedding (NeurIPS 2020 · similarity 0.88)
- A Simple but Tough-to-Beat Baseline for Sentence Embeddings (ICLR 2017 · similarity 0.86)
- Enriching Word Embeddings with a Regressor Instead of Labeled Corpora (AAAI 2019 · similarity 0.86)
- Repurposing Language Models into Embedding Models: Finding the Compute-Optimal Recipe (NeurIPS 2024 · similarity 0.85)
- Mixture of Scales: Memory-Efficient Token-Adaptive Binarization for Large Language Models (NeurIPS 2024 · similarity 0.84)
- NV-Embed: Improved Techniques for Training LLMs as Generalist Embedding Models (ICLR 2025 · 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.
