Research map / Label-efficient and robust learning / Recognition and long-tail
Reconstruct & Crush Network
Erinc Merdivan, Mohammad Reza Loghmani, Matthieu Geist
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 |
|---|---|---|---|
| use RCN instead of a standard CNN | autoencoder classifier · CIFAR-10 unseen ship | true negative rate: 0.755 -> 0.793 | Improved |
| use RCN instead of a standard CNN | autoencoder classifier · CIFAR-10 automobile | true positive rate: 0.74 -> 0.81 | Improved |
| use RCN instead of class-balanced CNN-reduced | autoencoder classifier · CIFAR-10 automobile | true positive rate: 0.82 -> 0.81 | No clear effect |
| use RCN instead of CNN-reduced | autoencoder classifier · CIFAR-10 unseen ship | true negative rate: 0.638 -> 0.793 | Improved |
| balance positive and negative training sets (CNN-reduced vs CNN) | CNN · CIFAR-10 unseen ship | true negative rate: 0.755 -> 0.638 | Hurt |
| use RCN instead of CNN | autoencoder classifier · CIFAR-100 plates | true positive rate: 0.718 -> 0.861 | Improved |
| use RCN instead of CNN | autoencoder classifier · CIFAR-100 tulips | true negative rate: 0.81 -> 0.853 | Improved |
| use RCN instead of PU-specific methods with 5% positives in unlabeled set | RCN · Amazon reviews PU | F-measure: 0.92 -> 0.90 | Hurt |
| raise positive share of unlabeled set from 5% to 50% | RCN · Amazon reviews PU | F-measure: 0.90 -> 0.83 | Hurt |
| raise positive share of unlabeled set from 5% to 50% | NB-SVM · Amazon reviews PU | F-measure: 0.92 -> 0.86 | Hurt |
| raise positive share of unlabeled set from 5% to 50% | Roc-EM · Amazon reviews PU | F-measure: 0.91 -> 0.90 | No clear effect |
| use RCN energy as response scorer | GRU autoencoder · Facebook bAbI tasks 1 and 2 | accuracy: 100% accuracy | No clear effect |
Closest papers by meaning
- Deconstructing Data Reconstruction: Multiclass, Weight Decay and General Losses (NeurIPS 2023 · similarity 0.84)
- Rethinking Reconstruction Autoencoder-Based Out-of-Distribution Detection (CVPR 2022 · similarity 0.82)
- Adversarial generalization of unfolding (model-based) networks (NeurIPS 2025 · similarity 0.81)
- Compression of Structured Data with Autoencoders: Provable Benefit of Nonlinearities and Depth (ICML 2024 · similarity 0.81)
- Scaling and evaluating sparse autoencoders (ICLR 2025 · similarity 0.81)
- Network Inversion of Convolutional Neural Nets (Student Abstract) (AAAI 2025 · similarity 0.81)
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.
