Research map / Probabilistic and scientific modeling / Molecules and proteins
Benchmarks, Algorithms, and Metrics for Hierarchical Disentanglement
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 |
|---|---|---|---|
| Use COFHAE with a learned hierarchy and assignments rather than flat baseline representations | Representation learning on the nine synthetic benchmarks · Spaceshapes and Chopsticks variants | R^4_c | Helped |
| Add the ground-truth hierarchy to a flat autoencoder (ablation) | COFHAE ablation · Chopsticks depth-2 both | R^4_c | Helped |
| Add ground-truth assignment supervision to a model with the hierarchy (ablation) | COFHAE ablation · Chopsticks depth-2 both | R^4_c | Helped |
| Add a marginal total-correlation penalty (ablation) | COFHAE with hierarchy and assignment supervision · Chopsticks depth-2 both | R^4_c | No clear effect |
| Use conditional rather than marginal total-correlation regularization (ablation) | COFHAE with hierarchy and assignment supervision · Chopsticks depth-2 both | R^4_c | Helped |
| Remove assignment supervision while retaining the hierarchy and conditional TC penalty (ablation) | COFHAE · Chopsticks depth-2 both | R^4_c | Hurt |
| Use a smooth Softplus activation instead of ReLU in MIMOSA's initial autoencoder (ablation) | MIMOSA · Chopsticks | hierarchy recovery | Helped |
| Use contagion dynamics when building manifold components (ablation) | MIMOSA · Chopsticks variants | purity | Helped |
| Use RANSAC-style robust local SVD fitting (ablation) | MIMOSA · Chopsticks variants | purity | Helped |
| Increase disentanglement penalty strength in flat baselines (ablation) | TCVAE, FactorVAE, CascadeVAE, and JointVAE · Synthetic hierarchical benchmarks | R^4_c | Hurt |
| Sample slope/intercept variables from Gaussian rather than uniform distributions (ablation) | COFHAE · Chopsticks | R^4_c | Hurt |
| Add noise to the benchmark data (ablation) | MIMOSA · Synthetic benchmarks | hierarchy recovery | Hurt |
Closest papers by meaning
- PROGRESSIVE LEARNING AND DISENTANGLEMENT OF HIERARCHICAL REPRESENTATIONS (ICLR 2020 · similarity 0.91)
- On Causally Disentangled Representations (AAAI 2022 · similarity 0.90)
- Enriching Disentanglement: From Logical Definitions to Quantitative Metrics (NeurIPS 2024 · similarity 0.89)
- Transferring disentangled representations: bridging the gap between synthetic and real images (NeurIPS 2024 · similarity 0.89)
- Disentangled Representation Learning via Flow Matching (NeurIPS 2026 · similarity 0.89)
- Disentanglement Beyond Static vs. Dynamic: A Benchmark and Evaluation Framework for Multi-Factor Sequential Representations (NeurIPS 2025 · similarity 0.88)
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.