Summary of Advancing Deep Active Learning & Data Subset Selection: Unifying Principles with Information-theory Intuitions, by Andreas Kirsch
Advancing Deep Active Learning & Data Subset Selection: Unifying Principles with Information-Theory Intuitions
by Andreas Kirsch
First submitted to arxiv on: 9 Jan 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Information Theory (cs.IT)
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 PhD thesis aims to improve the efficiency of deep learning models by enhancing label and training efficiency. To achieve this, the research investigates data subset selection techniques grounded in information-theoretic principles. Active learning improves label efficiency, while active sampling enhances training efficiency. The goal is to develop a more principled approach inspired by information theory, rather than relying on heuristics or ad-hoc methods. The thesis begins by analyzing epistemic and aleatoric uncertainty in deep neural networks, which provides valuable insights into different forms of uncertainty and their relevance for data subset selection. It then proposes various approaches for active learning and data subset selection in Bayesian deep learning. Finally, the research relates these approaches to approximations of information quantities in weight or prediction space. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This PhD thesis tries to make it easier to use deep learning by making label and training more efficient. To do this, they look at ways to choose which parts of a big dataset to use for training. They find that using “active learning” makes it possible to get the same results with fewer labels, and “active sampling” makes it faster to train models. The goal is to make deep learning work better in real-world situations outside of academia. |
Keywords
* Artificial intelligence * Active learning * Deep learning
