Research map / Probabilistic and scientific modeling / Variational and generative sampling
Morpho-MNIST: Quantitative Assessment and Diagnostics for Representation Learning
Daniel C. Castro, Jeremy Tan, Bernhard Kainz, Ender Konukoglu, Ben Glocker
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 |
|---|---|---|---|
| evaluate on locally perturbed digits (swelling, fractures) instead of plain test digits | kNN · MNIST digit recognition | accuracy (%): 96.25 -> 95.22 | Hurt |
| evaluate on locally perturbed digits instead of plain test digits | SVM · MNIST digit recognition | accuracy (%): 95.71 -> 92.47 | Hurt |
| evaluate on locally perturbed digits instead of plain test digits | MLP · MNIST digit recognition | accuracy (%): 97.97 -> 93.15 | Hurt |
| evaluate on locally perturbed digits instead of plain test digits | CNN (LeNet-5) · MNIST digit recognition | accuracy (%): 98.95 -> 95.33 | Hurt |
| train InfoGAN on data with random thinning and thickening plus an extra continuous code (InfoGAN-B vs InfoGAN-A) | InfoGAN · Overall MIG on inferred codes | mutual information gap: 0.0582 -> 0.1748 | Improved |
| add two binary latent codes and train on swollen and fractured digits (InfoGAN-C vs InfoGAN-A) | InfoGAN · Overall MIG on inferred codes | mutual information gap: 0.0582 -> 0.0507 | Hurt |
| reduce GAN latent dimension from 64 to 2 | GAN · Morphometric MMD test vs test data | MMD^2_l (x10^-3): 1.458 -> 8.876 | Hurt |
| use a CNN instead of kNN for abnormality detection | abnormality detection · Morpho-MNIST local | accuracy (%): 65.10 -> 97.53 | Improved |
| use a CNN instead of kNN for thickness regression | thickness regression · Morpho-MNIST global | RMSE: 0.467 -> 0.279 | Improved |
Closest papers by meaning
- Mapping the Multiverse of Latent Representations (ICML 2024 · similarity 0.84)
- Observable Propagation: Uncovering Feature Vectors in Transformers (ICML 2024 · similarity 0.84)
- Forte : Finding Outliers with Representation Typicality Estimation (ICLR 2025 · similarity 0.84)
- Objective drives the consistency of representational similarity across datasets (ICML 2025 · similarity 0.84)
- Identifying Metric Structures of Deep Latent Variable Models (ICML 2025 · similarity 0.83)
- Nonparametric Identification of Latent Concepts (ICML 2025 · 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.
