Research map / Learning theory and algorithms / Prediction and decision losses
Training Decision Trees as Replacement for Convolution Layers
Wolfgang Fuhl, Gjergji Kasneci, Wolfgang Rosenstiel, Enkeljda Kasneci
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 |
|---|---|---|---|
| Replace convolution layers with TI2 decision-tree layers (ablation) | LeNet-5 · MNIST test | classification accuracy: 99.37 -> 99.23 | Hurt |
| Use TI3 index pattern with inception instead of convolution layers (ablation) | LeNet-5 · MNIST test | classification accuracy: 99.37 -> 99.48 | Improved |
| Replace convolution layers with TI1 decision-tree layers | ResNet-34 · CIFAR10 test | classification accuracy: 91.12 -> 92.20 | Improved |
| Replace convolution layers with TI1 decision-tree layers | ResNet-16 · 300W common | NME: 3.69 -> 3.64 | Improved |
| Replace convolution layers with TI1 decision-tree layers | ResNet-16 · 300W challenging | NME: 6.76 -> 6.81 | Hurt |
| Replace convolution layers with TI1 decision-tree layers | ResNet-16 · 300W full | NME: 4.29 -> 4.26 | Improved |
| Replace convolution layers with TI2 decision-tree layers (ablation) | LeNet-5 · MNIST test inference, 1 CPU core | runtime: 0.83ms -> 0.18ms | Improved |
| Use TI3 rather than TI2 index pattern (ablation) | LeNet-5 · MNIST test inference, 1 CPU core | runtime: 0.18ms -> 0.36ms | Hurt |
| Replace convolution layers with TI1 decision-tree layers | ResNet-34 · CIFAR10 test inference, GPU 1050ti | runtime: 18.1ms -> 6.77ms | Improved |
| Replace convolution layers with TI1 decision-tree layers | ResNet-16 · 300W test inference, GPU 1050ti | runtime: 10.76ms -> 4.64ms | Improved |
Closest papers by meaning
- Interpreting CNNs via Decision Trees (CVPR 2019 · similarity 0.89)
- Learning Decision Trees Recurrently Through Communication (CVPR 2021 · similarity 0.88)
- Learning Binary Decision Trees by Argmin Differentiation (ICML 2021 · similarity 0.88)
- Training binary neural networks with real-to-binary convolutions (ICLR 2020 · similarity 0.87)
- PICNN: A Pathway towards Interpretable Convolutional Neural Networks (AAAI 2024 · similarity 0.86)
- Convolutions and More as Einsum: A Tensor Network Perspective with Advances for Second-Order Methods (NeurIPS 2024 · 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.
