Connectionists: Paper available: Phase transition analysis for shallow neural networks with arbitrary activation functions

[Michael Biehl]

Dear All.
Our publication in Physica A (Statistical Mechanics and its Applications)
*Phase transition analysis for shallow neural networks with arbitrary
activation functions*
Otavio Citton, Frederieke Richert, Michael Biehl
is available (open access) at
https://doi.org/10.1016/j.physa.2025.130356
<https://kwnsfk27.r.eu-west-1.awstrack.me/L0/https:%2F%2Fdoi.org%2F10.1016%2Fj.physa.2025.130356/1/010201946fb3ac5b-08eb8e1e-e3e2-4948-9f7f-d0fb6ddaef50-000000/LipkKfH2JDwvLfGBsKLrX4Q1fZI=409>

Highlights
• We present a statistical physics analysis of layered neural networks with
arbitrary activation functions.
• The method employs a Hermite polynomial series representation of the
activation function.
• The analysis enables the characterization of phase transitions in the
learning curve for any activation function.
• The approach allows the analytical study of student–teacher scenario with
mismatched activation functions.

Best regards,
Michael Biehl

---------------------------------------------------
Prof. Dr. Michael Biehl
Bernoulli Institute for Mathematics,
Computer Science & Artificial Intelligence
P.O. Box 407, 9700 AK Groningen, NL
https://www.cs.rug.nl/~biehl     m.biehl at rug.nl
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.srv.cs.cmu.edu/pipermail/connectionists/attachments/20250116/560e992f/attachment.html>