Research map / Probabilistic and scientific modeling / Variational and generative sampling
Combining Generative and Discriminative Models for Hybrid Inference
Victor Garcia Satorras, Zeynep Akata, Max Welling
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 learned GNN correction to graphical-model messages (Hybrid vs GM-messages) (ablation) | Kalman/HMM graphical inference · Michigan NCLT | MSE: 3.0048 -> 1.4109 | Improved |
| Add graphical-model messages to GNN (Hybrid vs GNN-messages) (ablation) | Gated GNN · Michigan NCLT | MSE: 1.7929 -> 1.4109 | Improved |
| Replace Kalman Smoother with Hybrid model | Kalman Smoother · Michigan NCLT | MSE: 3.0099 -> 1.4109 | Improved |
| Kalman Smoother vs raw GPS observations | Kalman Smoother · Michigan NCLT | MSE: 3.4974 -> 3.0099 | Improved |
| Use Taylor expansion J=1 prior knowledge in hybrid instead of fully learned GNN (J=0) (ablation) | Hybrid model · Lorenz attractor | samples needed for MSE=0.1: 5.000 -> 500 | Improved |
| Use Taylor expansion J=1 prior knowledge in hybrid instead of GNN (J=0) (ablation) | Hybrid model · Lorenz attractor | samples needed for MSE=0.05: 90.000 -> 5.000 | Improved |
| Increase Taylor terms from J=1 to J=2 or 5 (ablation) | Hybrid model · Lorenz attractor | samples needed for MSE=0.1: 500 -> 400 | Improved |
| Increase Taylor terms from J=1 to J=2 or 5 (ablation) | Hybrid model · Lorenz attractor | samples needed for MSE=0.05: 5.000 -> 4.000 | Improved |
| Increase Taylor terms J beyond 5 (ablation) | Hybrid model · Lorenz attractor | MSE | No clear effect |
| Add graphical-model messages to GNN (Hybrid vs GNN) on linear dynamics (ablation) | GNN-messages · Linear dynamics synthetic | MSE | Improved |
| Hybrid vs Kalman Smoother with more than 1K training samples | Kalman Smoother · Linear dynamics synthetic | MSE | Improved |
| Use Kalman Smoother vs GNN at low data (<10K) vs high data (>10K) | Kalman Smoother vs GNN · Linear dynamics synthetic | MSE | Mixed |
Closest papers by meaning
- Self-Supervised Inference in State-Space Models (ICLR 2022 · similarity 0.87)
- Gated Inference Network: Inference and Learning State-Space Models (NeurIPS 2024 · similarity 0.85)
- Inference via Interpolation: Contrastive Representations Provably Enable Planning and Inference (NeurIPS 2024 · similarity 0.84)
- On Neural Networks as Infinite Tree-Structured Probabilistic Graphical Models (NeurIPS 2024 · similarity 0.84)
- A Graph Dynamics Prior for Relational Inference (AAAI 2024 · similarity 0.84)
- Inferring stochastic low-rank recurrent neural networks from neural data (NeurIPS 2024 · 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.
