← All problems

Self-duality conjecture for neural-collapse solutions

3 years old · traced to

Consider the unconstrained feature model with classifier matrix W\boldsymbol{W} and feature matrix H\boldsymbol{H}. Let (W∗,H∗)(\boldsymbol{W}^*,\boldsymbol{H}^*) be a global solution and let hk∗\boldsymbol{h}_k^* denote the class feature aggregate.

Self-duality conjecture. The class feature aggregate is proportional to its corresponding classifier weight: for each k=1,…,Kk=1,\ldots,K, hk∗=αwk∗\boldsymbol{h}_k^*=\alpha\boldsymbol{w}_k^* for some α>0\alpha>0 independent of kk.

This conjecture expresses the self-duality expected between classifier weights and class features in neural collapse. Its resolution is not specified in the source.

References
  1. Jiachen Jiang, Jinxin Zhou, Peng Wang, Qing Qu, Dustin Mixon, Chong You and Zhihui Zhu, “Generalized Neural Collapse for a Large Number of Classes”, arXiv:2310.05351 (2023).

Progress summary

Never refreshed

Nothing recorded yet. Refresh searches the literature and the public web for attempts on this problem, and writes the first summary here.

Solutions 0

No solutions have been posted yet.