Research map / Graph learning / GNN architectures
Snowflake: Scaling GNNs to high-dimensional continuous control via parameter freezing
Charlie Blake, Vitaly Kurin, Maximilian Igl, Shimon Whiteson
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 |
|---|---|---|---|
| Freeze encoder, decoder, and message function while training the update function | NerveNet · High-dimensional locomotion tasks, including Centipede agents | reward | Improved |
| Freeze encoder, decoder, and message function | NerveNet · Centipede-20 to Centipede-12 zero-shot transfer | transfer performance | Improved |
| Use L2 regularization on the message-function MLP (ablation) | NerveNet · Locomotion control | reward | Improved |
| Set encoder, message-function, and decoder learning rates to zero (ablation) | NerveNet · Locomotion control | reward | Improved |
| Freeze all functions except the update function (ablation) | NerveNet · Centipede-6 | reward | Improved |
| Freeze all functions except the encoder (ablation) | NerveNet · Centipede-6 | reward | Improved |
| Freeze all functions except the message function (ablation) | NerveNet · Centipede-6 | reward | Improved |
| Freeze all functions except the decoder (ablation) | NerveNet · Centipede-6 | reward | Improved |
| Replace the GRU update function with an MLP (ablation) | NerveNet · Locomotion control | reward | Hurt |
| Reduce training batch size (ablation) | NerveNet · Locomotion control | sample efficiency | Hurt |
| Reduce training batch size while freezing encoder, decoder, and message function (ablation) | Snowflake · Locomotion control | sample efficiency | Improved |
Closest papers by meaning
- On Vanishing Gradients, Over-Smoothing, and Over-Squashing in GNNs: Bridging Recurrent and Graph Learning (NeurIPS 2025 · similarity 0.85)
- A Transfer Approach Using Graph Neural Networks in Deep Reinforcement Learning (AAAI 2024 · similarity 0.84)
- Dynamic Rescaling for Training GNNs (NeurIPS 2024 · similarity 0.84)
- Diffusing to the Top: Boost Graph Neural Networks with Minimal Hyperparameter Tuning (ICLR 2025 · similarity 0.84)
- Graph Neural Networks Gone Hogwild (ICLR 2025 · similarity 0.84)
- Flattening Hierarchies with Policy Bootstrapping (NeurIPS 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.
