Research map / Probabilistic and scientific modeling / Variational and generative sampling
On Parameter Tying by Quantization
Li Chou, 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 |
|---|---|---|---|
| relearn parameters after quantization (tied) versus quantization only (ablation) | quantized MLE · bn2o networks | max parameter error: 0.339 -> 0.299 | Improved |
| relearn parameters after quantization versus quantization only (ablation) | quantized MLE · students network | max parameter error: 0.141 -> 0.140 | Improved |
| relearn parameters after quantization versus quantization only (ablation) | quantized MLE · grid network | max parameter error: 0.0186 -> 0.0181 | Improved |
| relearn parameters after quantization versus quantization only (ablation) | quantized MLE · friends network | max parameter error: 0.008 -> 0.007 | Improved |
| quantize MLE weights without relearning | MLE · bn2o, students, grid, friends | max parameter error: 0.339 -> 0.339 | No clear effect |
| quantization alone at low k | MLE · noisy-or networks | test average log-likelihood | Hurt |
| relearning after quantization | quantized model · noisy-or networks | test average log-likelihood | Improved |
| increase k toward true number of parameters | quantized/relearned model · noisy-or networks | log-likelihood gap to MLE | Improved |
| tied-weight slice importance sampling (TW) instead of MC-SAT and Gibbs on relearned parameters | inference · students and grid networks | Hellinger distance of marginals | Improved |
| TW versus MC-SAT on untied MLE parameters | inference · noisy-or, relational, grid networks | Hellinger distance of marginals | No clear effect |
Closest papers by meaning
- Fisher Efficient Inference of Intractable Models (NeurIPS 2019 · similarity 0.85)
- Automatic Parameter Tying: A New Approach for Regularized Parameter Learning in Markov Networks (AAAI 2018 · similarity 0.85)
- Differentiable Annealed Importance Sampling and the Perils of Gradient Noise (NeurIPS 2021 · similarity 0.84)
- Forward $\chi^2$ Divergence Based Variational Importance Sampling (ICLR 2024 · similarity 0.84)
- Conformal Inference under High-Dimensional Covariate Shifts via Likelihood-Ratio Regularization (NeurIPS 2025 · similarity 0.84)
- GSQ: Highly-Accurate Low-Precision Scalar Quantization for LLMs via Gumbel-Softmax Sampling (NeurIPS 2026 · 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.
