Summary of Neural Normalized Compression Distance and the Disconnect Between Compression and Classification, by John Hurwitz et al.
Neural Normalized Compression Distance and the Disconnect Between Compression and Classification
by John Hurwitz, Charles Nicholas, Edward Raff
First submitted to arxiv on: 20 Oct 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Machine Learning (stat.ML)
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 The paper investigates the relationship between predictive classification and compression in information theory, particularly in deep learning methods. It introduces the Normalized Compression Distance (NCD) as a means of measuring similarity between sequences using compression. The authors convert large language models (LLMs) into lossless compressors to develop a Neural NCD, comparing them to classic algorithms like gzip. The study reveals that classification accuracy is not solely determined by compression rate, and other empirical anomalies are found, challenging current understanding. This research highlights the need for further investigation into what it means for a neural network to “compress” and what is required for effective classification. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper looks at how machines can learn to classify things correctly. It’s based on an idea that says good compression (making data smaller) leads to better learning. The researchers use a new way of measuring similarity between sequences called Normalized Compression Distance, or NCD. They take popular language models and turn them into special compressors to test their accuracy. Surprisingly, they found that how well something is compressed doesn’t always mean it will be classified correctly. This study shows that we don’t fully understand what makes a neural network “compress” data effectively. |
Keywords
» Artificial intelligence » Classification » Deep learning » Neural network
