Summary of Learned Feature Representations Are Biased by Complexity, Learning Order, Position, and More, By Andrew Kyle Lampinen et al.
Learned feature representations are biased by complexity, learning order, position, and more
by Andrew Kyle Lampinen, Stephanie C. Y. Chan, Katherine Hermann
First submitted to arxiv on: 9 May 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Computer Vision and Pattern Recognition (cs.CV)
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 proposed research investigates the relationship between representation learning and computation in machine learning and neuroscience. Surprisingly, it finds that learned feature representations are systematically biased towards representing certain features more strongly than others, depending on factors like feature complexity, order of learning, and input distribution. For instance, simpler or earlier-learned features tend to be represented more strongly, even if all features are learned equally well. The study explores how these biases are affected by architectures, optimizers, and training regimes and highlights the challenge for interpretability and comparing model and brain representations. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary In this research, scientists discovered that when computers learn new skills, they often focus on certain parts of what they’re learning more than others. This is important because it can affect how well we understand what the computer has learned. The study shows that some features are easier for the computer to learn or are learned earlier in the process, which makes them more prominent in the final representation. This bias affects different architectures and training methods, making it harder to compare representations from different models or brains. |
Keywords
» Artificial intelligence » Machine learning » Representation learning
