Research map / Optimization / Optimizers
End-to-End Learning for Structured Prediction Energy Networks
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 |
|---|---|---|---|
| Replace the FOE prior with a deep convolutional prior (ablation) | End-to-end SPEN on 7-Scenes · 7-Scenes, test scenes | PSNR: 36.41 -> 40.3 | Helped |
| Add per-iterate average loss, momentum 0.25, and trainable learning rates (ablation) | FOE-20 · 7-Scenes, test scenes | PSNR: 36.41 -> 37.34 | Helped |
| Reduce unrolling from 20 to 3 iterations (ablation) | FOE-20+ · 7-Scenes, test scenes | PSNR: 37.34 -> 37.62 | Helped |
| Reduce unrolling from 20 to 3 iterations (ablation) | DP-20 · 7-Scenes, test scenes | PSNR: 40.3 -> 40.4 | Helped |
| Train the deep-prior SPEN with an SSVM loss instead of end-to-end learning (ablation) | Deep-prior SPEN on 7-Scenes · 7-Scenes, test scenes | PSNR: 40.3 -> 38.7 | Hurt |
| Use the SPEN global energy rather than the local-only model (ablation) | CoNLL-2005 SRL with AD³ · CoNLL-2005 WSJ test | F1: 80.0 -> 80.7 | Helped |
| Use the SPEN global energy rather than the local-only model (ablation) | CoNLL-2005 SRL with AD³ · CoNLL-2005 Brown test | F1: 69.9 -> 69.4 | Hurt |
| Replace heuristic constraint enforcement with AD³ (ablation) | SPEN on CoNLL-2005 SRL · CoNLL-2005 WSJ test | F1: 80.7 -> 80.7 | No clear effect |
| Constrain the SPEN energy to be convex with respect to the output (ablation) | SPEN on CoNLL-2005 SRL · CoNLL-2005 WSJ test | F1: 80.7 -> 80.3 | Hurt |
| Use end-to-end learning instead of SSVM training (ablation) | SPEN on CoNLL-2005 SRL · CoNLL-2005 WSJ test | F1 | Helped |
| Use end-to-end SPEN learning with a deeper image prior rather than the ProximalNet baseline | 7-Scenes denoising · 7-Scenes, test scenes | PSNR: 36.3 -> 40.4 | Helped |
Closest papers by meaning
- Structured Prediction Energy Networks (ICML 2016 · similarity 0.92)
- A Stable Whitening Optimizer for Efficient Neural Network Training (NeurIPS 2025 · similarity 0.84)
- SOInter: A Novel Deep Energy-Based Interpretation Method for Explaining Structured Output Models (ICLR 2024 · similarity 0.84)
- Learning normalized image densities via dual score matching (NeurIPS 2025 · similarity 0.84)
- Benchmarking Predictive Coding Networks -- Made Simple (ICLR 2025 · similarity 0.83)
- ASGO: Adaptive Structured Gradient Optimization (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.