Research map / Probabilistic and scientific modeling / Variational and generative sampling
Goal-directed Generation of Discrete Structures with Conditional Generative Models
Amina Mollaysa, Brooks Paige, Alexandros Kalousis
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 |
|---|---|---|---|
| Train with reward-based sampled targets instead of maximum likelihood | LSTM conditional generator · Python expressions | MAE: 13.917 -> 11.823 | Improved |
| Train with reward-based sampled targets instead of maximum likelihood | LSTM conditional generator · Python expressions | accuracy within +-3: 0.596 -> 0.682 | Improved |
| Train with reward-based sampled targets instead of maximum likelihood | LSTM conditional generator · Python expressions | exact accuracy: 0.166 -> 0.166 | No clear effect |
| Train with reward-based sampled targets instead of maximum likelihood | LSTM conditional generator · Python expressions | negative log-likelihood: 1.830 -> 1.986 | Hurt |
| Train with reward-based sampled targets instead of maximum likelihood | LSTM conditional generator · QM9 | total MSE: 10.7237 -> 7.6398 | Improved |
| Train with reward-based sampled targets instead of maximum likelihood | LSTM conditional generator · ChEMBL | total MSE: 204.4400 -> 302.5634 | Mixed |
| Train with reward-based sampled targets instead of maximum likelihood | LSTM conditional generator · QM9 | valid fraction: 0.962 -> 0.989 | Improved |
| Train with reward-based sampled targets instead of maximum likelihood | LSTM conditional generator · ChEMBL | valid fraction: 0.895 -> 0.945 | Improved |
| Train with reward-based sampled targets instead of maximum likelihood | LSTM conditional generator · QM9 | novel fraction: 0.366 -> 0.261 | Hurt |
| Replace data augmentation (RAML-like) with entropy-regularised reward training | LSTM conditional generator · QM9 | novel fraction: 0.747 -> 0.274 | Hurt |
| Replace classic data augmentation with entropy-regularised reward training | LSTM conditional generator · QM9 | validity: 0.937 -> 0.977 | Improved |
| Use warm-started REINFORCE instead of reward-based sampled training | LSTM conditional generator · QM9 | validity: 0.9886 -> 0.4013 | Hurt |
Closest papers by meaning
- Variational Search Distributions (ICLR 2025 · similarity 0.87)
- Unlocking Guidance for Discrete State-Space Diffusion and Flow Models (ICLR 2025 · similarity 0.87)
- Fine-Tuning Discrete Diffusion Models via Reward Optimization with Applications to DNA and Protein Design (ICLR 2025 · similarity 0.86)
- Bridging Model-Based Optimization and Generative Modeling via Conservative Fine-Tuning of Diffusion Models (NeurIPS 2024 · similarity 0.85)
- Molecule Design by Latent Prompt Transformer (NeurIPS 2024 · similarity 0.85)
- Discrete Markov Probabilistic Models: An Improved Discrete Score-Based Framework with sharp convergence bounds under minimal assumptions (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.
