Research map / Architectures and efficiency / RNNs and normalization
Unsupervised Post-Processing of Word Vectors via Conceptor Negation
Tianlin Liu, Lyle Ungar, João Sedoc
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 |
|---|---|---|---|
| Apply Conceptor Negation instead of the strongest reported baseline | GloVe word vectors · SimLex-999 | Spearman correlation × 100: 44.97 -> 48.53 | Improved |
| Apply Conceptor Negation instead of the strongest reported baseline | Word2Vec word vectors · STS 2012 | Pearson correlation × 100: 57.67 -> 54.31 | Hurt |
| Apply Conceptor Negation instead of the strongest reported baseline | Word2Vec word vectors · STS 2013 | Pearson correlation × 100: 57.98 -> 59.17 | Improved |
| Apply Conceptor Negation instead of the strongest reported baseline | GloVe word vectors · SICK | Pearson correlation × 100: 67.85 -> 66.42 | Hurt |
| Use Conceptor Negation for concept categorization | GloVe word vectors · AP | purity × 100: 87.19 -> 90.95 | Improved |
| Use Conceptor Negation for concept categorization | Word2Vec word vectors · ESSLLI 2008 | purity × 100: 100.0 -> 100.0 | No clear effect |
| Initialize the Neural Belief Tracker with CN vectors instead of the strongest baseline | GloVe word vectors · WOZ 2.0 food | goal accuracy: 86.4 -> 88.8 | Improved |
| Initialize the Neural Belief Tracker with CN vectors instead of the strongest baseline | Word2Vec word vectors · WOZ 2.0 food | goal accuracy: 84.7 -> 78.5 | Hurt |
| Estimate the conceptor using only words appearing at least 200 times in the Wikipedia dump (ablation) | Conceptor Negation · Word-vector post-processing | performance | Improved |
| Apply ABTT or CN to FastText vectors (ablation) | FastText · Word-vector post-processing | performance | No clear effect |
| Use ABTT or CN for text classification (ablation) | Word vectors · Text classification | performance | No clear effect |
Closest papers by meaning
- All-but-the-Top: Simple and Effective Postprocessing for Word Representations (ICLR 2018 · similarity 0.87)
- Causally Denoise Word Embeddings Using Half-Sibling Regression (AAAI 2020 · similarity 0.85)
- Beyond Single Concept Vector: Modeling Concept Subspace in LLMs with Gaussian Distribution (ICLR 2025 · similarity 0.84)
- Further results on latent discourse models and word embeddings (JMLR 2021 · similarity 0.84)
- ConceptNet 5.5: An Open Multilingual Graph of General Knowledge (AAAI 2017 · similarity 0.84)
- Observable Propagation: Uncovering Feature Vectors in Transformers (ICML 2024 · similarity 0.82)
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.
