Research map / Learning theory and algorithms / Prediction and decision losses
Automatic Parameter Tying: A New Approach for Regularized Parameter Learning in Markov Networks
Li Chou, Pracheta Sahoo, Somdeb Sarkhel, Nicholas Ruozzi, Vibhav Gogate
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 |
|---|---|---|---|
| replace L2 with automatic parameter tying (APT) | Markov network pairwise models · bbc, d = 50 | test neg. PLL (lower better): 331.67 -> 260.95 | Improved |
| replace L2 with APT | Markov network models · dna, d = 50 | test neg. PLL (lower better): 69.26 -> 62.84 | Improved |
| replace L2 with APT | Markov network models · bbc, d = 5 | test neg. PLL (lower better): 259.96 -> 259.18 | Improved |
| replace L2 with APT | Markov network models · reuters, d = 5 | test neg. PLL (lower better): 91.33 -> 91.26 | No clear effect |
| replace L2 with APT | Markov network models · 20ng, d = 15 | test neg. PLL (lower better): 160.82 -> 162.29 | Hurt |
| replace LTR with APT | Markov network models · bbc, d = 50 | test neg. PLL (lower better): 343.90 -> 260.95 | Improved |
| replace LTR with APT | Markov network models · webkb, d = 15 | test neg. PLL (lower better): 169.17 -> 155.51 | Improved |
| replace L2 with LTR (hard tying L2 solution) | Markov network models · dna, d = 5 | test neg. PLL (lower better): 59.16 -> 58.54 | Improved |
| replace L2 with LTR | Markov network models · bbc, d = 50 | test neg. PLL (lower better): 331.67 -> 343.90 | Hurt |
| increase model density d from 5 to 50 for L2 vs APT | L2 and APT · accidents | test neg. PLL (lower better): L2 14.71 -> 13.85; APT 14.47 -> 12.69 | Mixed |
| vary number of clusters k | APT · dna | test neg. PLL (lower better): approximately 20 clusters | Improved |
| use hard tying with small k instead of soft tying | APT · general | test neg. PLL (lower better) | Hurt |
Closest papers by meaning
- On Parameter Tying by Quantization (AAAI 2016 · similarity 0.85)
- Infinity Learning: Learning Markov Chains from Aggregate Steady-State Observations (AAAI 2020 · similarity 0.84)
- Structure Learning of Partitioned Markov Networks (ICML 2016 · similarity 0.84)
- MetaOptimize: A Framework for Optimizing Step Sizes and Other Meta-parameters (ICML 2025 · similarity 0.84)
- Improving Deep Learning Optimization through Constrained Parameter Regularization (NeurIPS 2024 · similarity 0.82)
- Generalization Bounds via Meta-Learned Model Representations: PAC-Bayes and Sample Compression Hypernetworks (ICML 2025 · similarity 0.82)
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.
