Polysemanticity and Capacity in Neural Networks

October 04, 2022 ยท Declared Dead ยท ๐Ÿ› arXiv.org

๐Ÿ‘ป CAUSE OF DEATH: Ghosted
No code link whatsoever

"No code URL or promise found in abstract"

Evidence collected by the PWNC Scanner

Authors Adam Scherlis, Kshitij Sachan, Adam S. Jermyn, Joe Benton, Buck Shlegeris arXiv ID 2210.01892 Category cs.NE: Neural & Evolutionary Cross-listed cs.AI, cs.LG Citations 52 Venue arXiv.org Last Checked 3 months ago
Abstract
Individual neurons in neural networks often represent a mixture of unrelated features. This phenomenon, called polysemanticity, can make interpreting neural networks more difficult and so we aim to understand its causes. We propose doing so through the lens of feature \emph{capacity}, which is the fractional dimension each feature consumes in the embedding space. We show that in a toy model the optimal capacity allocation tends to monosemantically represent the most important features, polysemantically represent less important features (in proportion to their impact on the loss), and entirely ignore the least important features. Polysemanticity is more prevalent when the inputs have higher kurtosis or sparsity and more prevalent in some architectures than others. Given an optimal allocation of capacity, we go on to study the geometry of the embedding space. We find a block-semi-orthogonal structure, with differing block sizes in different models, highlighting the impact of model architecture on the interpretability of its neurons.
Community shame:
Not yet rated
Community Contributions

Found the code? Know the venue? Think something is wrong? Let us know!

๐Ÿ“œ Similar Papers

In the same crypt โ€” Neural & Evolutionary

๐Ÿ”ฎ ๐Ÿ”ฎ The Ethereal

LSTM: A Search Space Odyssey

Klaus Greff, Rupesh Kumar Srivastava, ... (+3 more)

cs.NE ๐Ÿ› IEEE TNNLS ๐Ÿ“š 6.0K cites 11 years ago

Died the same way โ€” ๐Ÿ‘ป Ghosted