Research map / Architectures and efficiency / RNNs and normalization
SPINE: SParse Interpretable Neural Embeddings
Anant Subramanian, Danish Pruthi, Harsh Jhamtani, Taylor Berg-Kirkpatrick, Eduard Hovy
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 |
|---|---|---|---|
| Transform GloVe embeddings with SPINE instead of using the original embeddings | GloVe · Word Intrusion Detection | precision: 22.97 -> 68.35 | Improved |
| Transform word2vec embeddings with SPINE instead of using the original embeddings | word2vec · Word Intrusion Detection | precision: 26.08 -> 74.83 | Improved |
| Add the PSL loss term (λ2 = 1) instead of omitting it (λ2 = 0) (ablation) | SPINE · GloVe embeddings | sparsity: 22% more sparsity | Improved |
| Add the PSL loss term (λ2 = 1) instead of omitting it (λ2 = 0) (ablation) | SPINE · word2vec embeddings | sparsity: 67% more sparsity | Improved |
| Add denoising noise to the autoencoder (ablation) | SPINE · Word embeddings | sparsity: 6% more sparse | Improved |
| Use the ASL and PSL combination instead of an ℓ1 sparsity regularizer (ablation) | SPINE · Word embeddings | automated topic coherence | Improved |
| Increase the hidden dimension size to 2000 or more (ablation) | SPINE · Word embeddings | downstream task performance | Hurt |
| Reduce the hidden dimension size to 500 (ablation) | SPINE · Word embeddings | interpretability | Hurt |
Closest papers by meaning
- Understanding the Semantic Content of Sparse Word Embeddings Using a Commonsense Knowledge Base (AAAI 2020 · similarity 0.88)
- Rethinking Evaluation of Sparse Autoencoders through the Representation of Polysemous Words (ICLR 2025 · similarity 0.86)
- Automatically Interpreting Millions of Features in Large Language Models (ICML 2025 · similarity 0.86)
- Scaling and evaluating sparse autoencoders (ICLR 2025 · similarity 0.85)
- SAEBench: A Comprehensive Benchmark for Sparse Autoencoders in Language Model Interpretability (ICML 2025 · similarity 0.85)
- Compute Optimal Inference and Provable Amortisation Gap in Sparse Autoencoders (ICML 2025 · similarity 0.85)
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.
