A note on the capacity of the binary perceptron
Altschuler, Dylan J., Tikhomirov, Konstantin
–arXiv.org Artificial Intelligence
Determining the capacity $\alpha_c$ of the Binary Perceptron is a long-standing problem. Krauth and Mezard (1989) conjectured an explicit value of $\alpha_c$, approximately equal to .833, and a rigorous lower bound matching this prediction was recently established by Ding and Sun (2019). Regarding the upper bound, Kim and Roche (1998) and Talagrand (1999) independently showed that $\alpha_c$ < .996, while Krauth and Mezard outlined an argument which can be used to show that $\alpha_c$ < .847. The purpose of this expository note is to record a complete proof of the bound $\alpha_c$ < .847. The proof is a conditional first moment method combined with known results on the spherical perceptron
arXiv.org Artificial Intelligence
Jan-22-2024
- Country:
- North America > United States
- New Jersey > Mercer County
- Princeton (0.05)
- New York (0.05)
- Pennsylvania > Philadelphia County
- Philadelphia (0.05)
- New Jersey > Mercer County
- North America > United States
- Genre:
- Research Report (0.40)
- Technology: