Research map / Architectures and efficiency / RNNs and normalization
No One-Size-Fits-All Neurons: Task-Based Neurons for Artificial Neural Networks
Fenglei Fan, Meng Wang, Hangcheng Dong, Jianwei Ma, Tieyong Zeng
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 a task-based neuron learned by vectorized symbolic regression instead of a linear neuron | fully connected network · synthetic p1 (S1) | MSE: 0.0215 -> 0.0009 | Improved |
| use task-based neurons instead of linear neurons | fully connected network · fifa (S1) | MSE: 0.0895 -> 0.0637 | Improved |
| use task-based neurons instead of linear neurons | fully connected network · phoneme (S2) | accuracy: 0.8020 -> 0.8510 | Improved |
| replace a random polynomial neuron with a symbolic-regression-learned neuron (ablation) | task-based network · vehicle | accuracy: 0.6894 -> 0.8176 | Improved |
| replace a random polynomial neuron with a symbolic-regression-learned neuron (ablation) | task-based network · Bike Sharing Demand | MSE: 0.0747 -> 0.0184 | Improved |
| replace a random polynomial neuron with a symbolic-regression-learned neuron (ablation) | task-based network · credit | accuracy: 0.7427 -> 0.7441 | No clear effect |
| use VSR instead of conventional SR to fit low-dimensional tabular data | symbolic regression · concrete strength | MSE: 0.4617 -> 0.7914 | Hurt |
| build a neural network from the VSR neuron instead of using SR expression directly | task-based network · concrete strength | MSE: 0.4617 -> 0.1144 | Improved |
| use VSR instead of SR on high-dimensional synthetic data | symbolic regression · GMM synthetic D=400 | validation RMSE: 2775.011 -> 1273.311 | Improved |
| use Sigmoid instead of ReLU activation in task-based network (ablation) | TN · Bike Sharing Demand | MSE: 0.0189 -> 0.0110 | Improved |
| use Sigmoid instead of ReLU activation in linear network (ablation) | LN · Bike Sharing Demand | MSE: 0.0152 -> 0.0496 | Hurt |
| use a sine/cosine function space for the neuron instead of linear neuron | task-based network · california housing | MSE: 0.0564 -> 0.0512 | Improved |
Closest papers by meaning
- Structured flexibility in recurrent neural networks via neuromodulation (NeurIPS 2024 · similarity 0.85)
- Scaling Laws and Tradeoffs in Recurrent Networks of Expressive Neurons (NeurIPS 2026 · similarity 0.84)
- Evaluating Neuron Explanations: A Unified Framework with Sanity Checks (ICML 2025 · similarity 0.84)
- Flexible task abstractions emerge in linear networks with fast and bounded units (NeurIPS 2024 · similarity 0.84)
- What should a neuron aim for? Designing local objective functions based on information theory (ICLR 2025 · similarity 0.84)
- Measure gradients, not activations! Enhancing neuronal activity in deep reinforcement learning (NeurIPS 2025 · similarity 0.84)
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.
