Research map / Image generation / GANs
Illuminating Mario Scenes in the Latent Space of a Generative Adversarial Network
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 quality-diversity search (MAP-Elites, ME (line), or CMA-ME) instead of CMA-ES or random search | DCGAN latent-space search · Representation-based BCs | QD-Score: 385.1 -> 692.5; 308.6 -> 692.5 | Helped |
| Use CMA-ME instead of MAP-Elites | DCGAN latent-space search · Representation-based BCs | QD-Score: 692.5 -> 776.8 | Helped |
| Use CMA-ME instead of ME (line) | DCGAN latent-space search · Representation-based BCs | QD-Score: 682.7 -> 776.8 | Helped |
| Use quality-diversity search (MAP-Elites, ME (line), or CMA-ME) instead of CMA-ES or random search | DCGAN latent-space search · KL-divergence | QD-Score: 331.5 -> 485.6; 210.6 -> 485.6 | Helped |
| Use CMA-ME instead of MAP-Elites | DCGAN latent-space search · KL-divergence | QD-Score: 485.6 -> 551.3 | Helped |
| Use CMA-ME instead of ME (line) | DCGAN latent-space search · KL-divergence | QD-Score: 488.0 -> 551.3 | Helped |
| Use any tested QD algorithm rather than random search | DCGAN latent-space search · Agent-based BCs | QD-Score: 20.2; 19.8; 20.4; 18.9; 21.6 | No clear effect |
| Add a directional line variation operator to MAP-Elites (ablation) | MAP-Elites · Representation-based BCs | QD-Score: 692.5 -> 682.7 | No clear effect |
| Add a directional line variation operator to MAP-Elites (ablation) | MAP-Elites · KL-divergence | QD-Score: 485.6 -> 488.0 | No clear effect |
| Increase the number of sky tiles | CMA-ME-generated scenes · Representation-based scenes rated by participants | Perceived difficulty: β = 4.35, t(817)= 11.56, p <0.001 | Helped |
| Increase the number of enemies | CMA-ME-generated scenes · Representation-based scenes rated by participants | Perceived difficulty: β = 0.94, t(817)= 2.48, p = 0.001 | Helped |
| Increase KL-divergence from a human-designed reference scene | CMA-ME-generated scenes · Scenes compared with groundtruth scene 1 | Perceived similarity: β_1 = -1.24, t(773) = -10.97, p < 0.001 | Hurt |
Closest papers by meaning
- Procedural Level Generation with Diffusion Models from a Single Example (AAAI 2024 · similarity 0.88)
- GAN Dissection: Visualizing and Understanding Generative Adversarial Networks (ICLR 2019 · similarity 0.87)
- A Closer Look at the Optimization Landscapes of Generative Adversarial Networks (ICLR 2020 · similarity 0.87)
- Discovering Interpretable Latent Space Directions of GANs Beyond Binary Attributes (CVPR 2021 · similarity 0.87)
- Geometry Score: A Method For Comparing Generative Adversarial Networks (ICML 2018 · similarity 0.87)
- A Domain Agnostic Measure for Monitoring and Evaluating GANs (NeurIPS 2019 · 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.