Research map / Continual learning / Online and replay-based
Firefly Neural Architecture Descent: a General Approach for Growing Neural Networks
Lemeng Wu, Bo Liu, Peter Stone, Qiang Liu
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 |
|---|---|---|---|
| Add brand-new neurons alongside neuron splitting (ablation) | Firefly growing a toy RBF network · Toy RBF network | training loss | Improved |
| Use only neuron splitting, without brand-new neurons (ablation) | Firefly growing a toy RBF network · Toy RBF network | training loss | Hurt |
| Use Firefly's greedy selection instead of random search | Growing a toy RBF network · Toy RBF network | training loss | Improved |
| Increase the number of brand-new neuron candidates (ablation) | Firefly growing a toy RBF network · Toy RBF network | training loss | Improved |
| Grow depth using Firefly rather than NASH | Growing neural networks · MNIST | Accuracy: 99.50 -> 99.59 | Improved |
| Grow depth using Firefly rather than AutoGrow | Growing neural networks · MNIST | Param (M): 2.3 -> 1.9 | Improved |
| Grow depth using Firefly rather than NASH | Growing neural networks · FashionMNIST | Accuracy: 94.34 -> 94.48 | Improved |
| Grow depth using Firefly rather than AutoGrow | Growing neural networks · FashionMNIST | Param (M): 2.3 -> 2.1 | Improved |
| Use Firefly for cell-based architecture search instead of DARTS (second order) | Cell-based NAS · CIFAR-10 | Search Time (GPU Days): 4 -> 1.5 | Improved |
| Use Firefly for continual learning instead of CPG | Continual image classification · CIFAR-100, 20-way split | Avg. Accuracy (20 tasks): 90.75 -> 91.03 | Improved |
| Use Firefly for continual learning instead of CPG | Continual image classification · CIFAR-100, 20-way split | Extra Growing Epochs: 420 -> 80 | Improved |
Closest papers by meaning
- On the Theory of Continual Learning with Gradient Descent for Neural Networks (ICML 2026 · similarity 0.86)
- Growing Efficient Deep Networks by Structured Continuous Sparsification (ICLR 2021 · similarity 0.85)
- Learned Optimizers that Scale and Generalize (ICML 2017 · similarity 0.85)
- Splitting Steepest Descent for Growing Neural Architectures (NeurIPS 2019 · similarity 0.85)
- Rethinking the Stability-Plasticity Trade-off in Continual Learning from an Architectural Perspective (ICML 2025 · similarity 0.84)
- Convergence and Implicit Bias of Gradient Descent on Continual Linear Classification (ICLR 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.
