phenomenon · first seen 5 Oct 2026
superposition in neural representations
Also: superposition
Neural networks encoding more features than they have dimensions by storing them as nearly orthogonal directions.
Mentions per week
PapersCodeDiscussion
Earliest sources
- Inference and learning in sparse autoencoders as natural gradient flowPapers · 5 Oct 2026 · Hadi Vafaii, Tejas Rao, David Chanin, Thomas Fel, Jacob L. Yates, Bruno Olshause…
- The Ball and the Box: Two Geometries of Computation in SuperpositionPapers · 8 Oct 2026 · Xiaoyu Li, Lequan Lin, Dai Shi, Jiaojiao Jiang, Junbin Gao, Andi Han
- Unifying and Beyond the Theory of Pervasive Capability Degradation in LLM Post-training SFT from Feature InterferencePapers · 3 Nov 2026 · Hang Chen, Jiaying Zhu, Wenya Wang
Latest sources
- Unifying and Beyond the Theory of Pervasive Capability Degradation in LLM Post-training SFT from Feature InterferencePapers · 3 Nov 2026 · Hang Chen, Jiaying Zhu, Wenya Wang
- The Ball and the Box: Two Geometries of Computation in SuperpositionPapers · 8 Oct 2026 · Xiaoyu Li, Lequan Lin, Dai Shi, Jiaojiao Jiang, Junbin Gao, Andi Han
- Inference and learning in sparse autoencoders as natural gradient flowPapers · 5 Oct 2026 · Hadi Vafaii, Tejas Rao, David Chanin, Thomas Fel, Jacob L. Yates, Bruno Olshause…