Summary of The Propensity For Density in Feed-forward Models, by Nandi Schoots et al.
The Propensity for Density in Feed-forward Models
by Nandi Schoots, Alex Jackson, Ali Kholmovaia, Peter McBurney, Murray Shanahan
First submitted to arxiv on: 18 Oct 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Artificial Intelligence (cs.AI)
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 study investigates whether neural networks tend to utilize all available weights during training, even when the task can be solved with fewer weights. The authors examine the effects of pruning fully connected, convolutional, and residual models on their performance, varying model widths in the process. They find that the proportion of prunable weights is largely invariant to model size, with increasing width having little effect on the density of the pruned network relative to its absolute size. This substantial prunability is observed across a wide range of model sizes, from small to large. The authors present three hypotheses to explain these findings. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Do neural networks really need all those weights? Scientists studied how well they work when some of those weights are “pruned” away. They looked at different types of neural networks and made them smaller or bigger. Surprisingly, it didn’t matter much whether the network was big or small – most of its weights could be cut without hurting performance. This is important for making computers smarter because it means we can use less power and memory to get the same results. |
Keywords
» Artificial intelligence » Pruning
