Research map / Architectures and efficiency / RNNs and normalization
Deep Learning in Target Space
Michael Fairbank, Spyridon Samothrakis, Luca Citi
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 in target space instead of weight space | CNN · CIFAR-100 (no dropout) | test accuracy (%): 40.3 -> 50.5 | Improved |
| train in target space instead of weight space | CNN · CIFAR-10 (no dropout) | test accuracy (%): 77.9 -> 82.6 | Improved |
| train in target space instead of weight space | CNN · CIFAR-100 (dropout) | test accuracy (%): 52.22 -> 60.4 | Improved |
| train in target space instead of weight space | CNN · CIFAR-10 (dropout) | test accuracy (%): 82.90 -> 87.4 | Improved |
| train in target space instead of weight space | CNN · CIFAR-100 (dropout + ensemble) | test accuracy (%): 56.85 -> 63.24 | Improved |
| train in target space instead of weight space | CNN · MNIST (no dropout) | test accuracy (%): 99.26 -> 99.29 | No clear effect |
| train in target space instead of weight space plus batch normalisation | CNN · CIFAR-100 (no dropout) | test accuracy (%): 46.5 -> 50.5 | Improved |
| train in target space instead of weight space plus batch normalisation | CNN · MNIST-Fashion (dropout) | test accuracy (%): 94.07 -> 93.7 | Hurt |
| train in target space instead of weight space plus batch normalisation | CNN · MNIST (no dropout) | test accuracy (%): 99.41 -> 99.29 | Hurt |
| add batch normalisation to weight space | CNN · CIFAR-100 (no dropout) | test accuracy (%): 40.3 -> 46.5 | Improved |
| add dropout to target space (ablation) | CNN · CIFAR-100 | test accuracy (%): 50.5 -> 60.4 | Improved |
| add ensemble of two networks to target space with dropout (ablation) | CNN · CIFAR-100 | test accuracy (%): 60.4 -> 63.24 | Improved |
Closest papers by meaning
- Expand-and-Cluster: Parameter Recovery of Neural Networks (ICML 2024 · similarity 0.84)
- Feature Learning Dynamics in Infinite-Depth Neural Networks (NeurIPS 2026 · similarity 0.83)
- Local Loss Optimization in the Infinite Width: Stable Parameterization of Predictive Coding Networks and Target Propagation (ICLR 2025 · similarity 0.83)
- Function-space Parameterization of Neural Networks for Sequential Learning (ICLR 2024 · similarity 0.83)
- The Benefits of Reusing Batches for Gradient Descent in Two-Layer Networks: Breaking the Curse of Information and Leap Exponents (ICML 2024 · similarity 0.83)
- Towards Scalable and Versatile Weight Space Learning (ICML 2024 · 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.
