Research map / Image generation / Autoregressive and latent generation
Hierarchical Conditioning of Diffusion Models Using Tree-of-Life for Studying Species Evolution
Mridul Khurana, Arka Daw, M. Maruf, Josef C. Uyeda, Wasila Dahdul, Caleb Charpentier, Yasin Bakış, Henry L. Bart Jr., Paula M. Mabee, Hilmar Lapp, James P. Balhoff, Wei-Lun Chao and 3 more
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 |
|---|---|---|---|
| condition LDM with hierarchical phylogenetic embeddings instead of class-label embeddings | LDM · GLIN fish | FID: 11.46 -> 11.38 | No clear effect |
| use HIER-Embed instead of CLIP-encoded scientific names | LDM conditioning · GLIN fish generated images classification | F1-score (%): 70.16 -> 82.21 | Improved |
| use LDM instead of encoder-decoder Phylo-NN | Phylo-Diffusion · GLIN fish | FID: 28.08 -> 11.38 | Improved |
| mask level-4 embedding with Gaussian noise (ablation) | Phylo-Diffusion · Trait masking, level 3 nodes | mean probability change: in-subtree 0.1070 vs out-of-subtree 0.0027 | Improved |
| mask levels 3 and 4 with noise (ablation) | Phylo-Diffusion · Trait masking, level 2 nodes | mean probability change: in-subtree 0.0316 vs out-of-subtree 0.0107; Node 5: -0.0434 vs 0.0292 | Mixed |
| vary number of discretized phylogeny levels (2, 4, 6, 8) (ablation) | Phylo-Diffusion · GLIN fish | FID: 11.84 / 11.38 / 11.41 / 11.77 | No clear effect |
| vary HIER-Embed dimension from 16 to 1024 (ablation) | Phylo-Diffusion · GLIN fish | FID: 11.23 to 11.69 | No clear effect |
| use HIER-Embed instead of class conditional on birds | LDM · CUB-200-2011 (190 birds) | F1: 0.68 -> 0.64 | Hurt |
| use HIER-Embed instead of scientific name encoding on birds | LDM · CUB-200-2011 (190 birds) | FID: 8.5 -> 6.7 | Improved |
| leave three species out of training (ablation) | Phylo-Diffusion · Leave-three-out generalization | F1: F1 95.6 | Improved |
Closest papers by meaning
- Deep Generative Models for Phylogenetic Inference with Complex Evolutionary Processes (NeurIPS 2026 · similarity 0.88)
- What Do You See in Common? Learning Hierarchical Prototypes over Tree-of-Life to Discover Evolutionary Traits (ICLR 2025 · similarity 0.87)
- Diffusion Models are Evolutionary Algorithms (ICLR 2025 · similarity 0.87)
- PhyloGen: Language Model-Enhanced Phylogenetic Inference via Graph Structure Generation (NeurIPS 2024 · similarity 0.86)
- BioCLIP 2: Emergent Properties from Scaling Hierarchical Contrastive Learning (NeurIPS 2025 · similarity 0.86)
- Likelihood-free inference of phylogenetic tree posterior distributions (NeurIPS 2026 · similarity 0.86)
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.
