Research map / Architectures and efficiency / RNNs and normalization
Sparse Boltzmann Machines with Structure Learning as Applied to Text Analysis
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 |
|---|---|---|---|
| Learn sparse connectivity and hidden-unit count using SBM-SFC instead of fully connected Replicated Softmax | Replicated Softmax · NIPS test | per-word perplexity: 547 -> 488 | Helped |
| Learn sparse connectivity and hidden-unit count using SBM-SFC instead of fully connected Replicated Softmax | Replicated Softmax · CiteULike test | per-word perplexity: 636 -> 597 | Helped |
| Learn sparse connectivity and hidden-unit count using SBM-SFC instead of fully connected Replicated Softmax | Replicated Softmax · New York Times test | per-word perplexity: 1,809 -> 1,583 | Helped |
| Add hidden-unit tree connections to the sparse model (ablation) | SBM-SFC · NIPS test | per-word perplexity: 551 -> 488 | Helped |
| Add hidden-unit tree connections to the sparse model (ablation) | SBM-SFC · CiteULike test | per-word perplexity: 668 -> 597 | Helped |
| Add hidden-unit tree connections to the sparse model (ablation) | SBM-SFC · New York Times test | per-word perplexity: 1,910 -> 1,583 | Helped |
| Use learned SBM-SFC structure rather than iterative weight-magnitude pruning | Replicated Softmax with matching hidden-unit count and sparsity · CiteULike test | per-word perplexity: 584 vs 597 | Mixed |
| Use SBM-SFC rather than RS+ Pruned | Replicated Softmax with matching hidden-unit count · NIPS | interpretability score: 0.1006 -> 0.1235 | Helped |
| Use SBM-SFC rather than RS+ Pruned | Replicated Softmax with matching hidden-unit count · CiteULike | interpretability score: 0.1449 -> 0.1725 | Helped |
| Use SBM-SFC rather than RS+ Pruned | Replicated Softmax with matching hidden-unit count · New York Times | interpretability score: 0.1420 -> 0.1433 | Helped |
Closest papers by meaning
- Learning with Restricted Boltzmann Machines: Asymptotics of AMP and GD in High Dimensions (NeurIPS 2025 · similarity 0.87)
- Rethinking Evaluation of Sparse Autoencoders through the Representation of Polysemous Words (ICLR 2025 · similarity 0.84)
- From Flat to Hierarchical: Extracting Sparse Representations with Matching Pursuit (NeurIPS 2025 · similarity 0.84)
- Adaptive Layer Sparsity for Large Language Models via Activation Correlation Assessment (NeurIPS 2024 · similarity 0.84)
- Jacobian Sparse Autoencoders: Sparsify Computations, Not Just Activations (ICML 2025 · similarity 0.84)
- SASA: Subspace-Aware Sparse Autoencoders for Effective Mechanistic Interpretability (NeurIPS 2026 · similarity 0.83)
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.