Research map / Learning theory and algorithms / Prediction and decision losses
Janossy Pooling: Learning Deep Permutation-Invariant Functions for Variable-Size Inputs
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| increase k from 1 to 2 or 3 in Janossy pooling (ablation) | MLP(30) with Linear rho · variance task | RMSE: 119.05 -> 4.37 | Helped |
| increase k from 1 to 3 with linear rho (ablation) | MLP(30) with Linear rho · range task | accuracy: 0.04 -> 0.21 | Helped |
| replace linear rho with MLP(100) at k=1 (ablation) | DeepSets MLP(30) · range task | accuracy: 0.04 -> 0.97 | Helped |
| increase k beyond 1 when rho is an MLP (ablation) | MLP(30) with MLP(100) rho · variance task | RMSE: 1.95 -> 6.90 | Hurt |
| use full-dependency GRU with pi-SGD instead of k-ary (ablation) | GRU(80) with MLP(100) · variance task | RMSE: 6.90 -> 0.42 | Helped |
| sample 20 permutations at inference instead of 1 (ablation) | GRU(80) with Linear rho · variance task | RMSE: 1.43 -> 1.20 | Helped |
| sample 20 permutations at inference instead of 1 (ablation) | GRU(80) with MLP(100) rho · variance task | RMSE: 0.42 -> 0.40 | No clear effect |
| increase neighborhood sample size k (ablation) | GraphSAGE LSTM pi-SGD · PPI | Micro-F1: 0.538 -> 0.757 | Helped |
| increase k1,k2 from 3 to 25 (ablation) | GraphSAGE LSTM pi-SGD · Cora and Pubmed | Micro-F1 | No clear effect |
| use mean-pooling over entire neighborhood instead of LSTM Janossy (ablation) | GraphSAGE · PPI | Micro-F1: 0.757 -> 0.767 | Helped |
| sample 7 permutations at test time instead of 1 (ablation) | GraphSAGE LSTM pi-SGD · PPI | accuracy | Helped |
Closest papers by meaning
- From Sorting Algorithms to Scalable Kernels: Bayesian Optimization in High-Dimensional Permutation Spaces (ICLR 2026 · similarity 0.87)
- Regularizing Towards Permutation Invariance In Recurrent Models (NeurIPS 2020 · similarity 0.87)
- Robust Sequence Submodular Maximization (NeurIPS 2020 · similarity 0.86)
- Scalable Normalizing Flows for Permutation Invariant Densities (ICML 2021 · similarity 0.86)
- Polynomial Time Learning Augmented Algorithms for NP-hard Permutation Problems (ICML 2025 · similarity 0.85)
- A Unified Analysis of Stochastic Gradient Descent with Arbitrary Data Permutations and Beyond (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.