Summary of Stable Training Of Normalizing Flows For High-dimensional Variational Inference, by Daniel Andrade
Stable Training of Normalizing Flows for High-dimensional Variational Inference
by Daniel Andrade
First submitted to arxiv on: 26 Feb 2024
Categories
- Main: Machine Learning (stat.ML)
- Secondary: Machine Learning (cs.LG)
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary This paper proposes a novel approach to stabilize the training of normalizing flows (NFs) based on coupling layers (Real NVPs) when approximating high-dimensional posterior distributions. The authors identify that previous methods for stabilizing stochastic gradient descent can be insufficient, as samples often exhibit unusual high values during training. They introduce two modifications: soft-thresholding of the scale in Real NVPs and a bijective soft log transformation of the samples. These modifications enable stable training of Real NVPs for posteriors with several thousand dimensions, allowing for accurate marginal likelihood estimation via importance sampling. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper helps us understand how to make machines learn better by making adjustments to a special kind of machine learning model called normalizing flows. Normalizing flows are used to find the most likely outcome from some data. The problem is that when we try to train these models, they can get stuck and not work well. To solve this issue, the authors suggest two new techniques: one adjusts the scale of the model, and the other changes how the model looks at the data. By using these techniques, machines can learn better from big datasets. |
Keywords
* Artificial intelligence * Likelihood * Machine learning * Stochastic gradient descent
