تقرير
A spin-glass model for the loss surfaces of generative adversarial networks
العنوان: | A spin-glass model for the loss surfaces of generative adversarial networks |
---|---|
المؤلفون: | Baskerville, Nicholas P, Keating, Jonathan P, Mezzadri, Francesco, Najnudel, Joseph |
سنة النشر: | 2021 |
المجموعة: | Computer Science Mathematics Condensed Matter Mathematical Physics |
مصطلحات موضوعية: | Mathematical Physics, Condensed Matter - Disordered Systems and Neural Networks, Condensed Matter - Statistical Mechanics, Computer Science - Machine Learning, Mathematics - Probability |
الوصف: | We present a novel mathematical model that seeks to capture the key design feature of generative adversarial networks (GANs). Our model consists of two interacting spin glasses, and we conduct an extensive theoretical analysis of the complexity of the model's critical points using techniques from Random Matrix Theory. The result is insights into the loss surfaces of large GANs that build upon prior insights for simpler networks, but also reveal new structure unique to this setting. Comment: 26 pages, 9 figures |
نوع الوثيقة: | Working Paper |
DOI: | 10.1007/s10955-022-02875-w |
URL الوصول: | http://arxiv.org/abs/2101.02524 |
رقم الأكسشن: | edsarx.2101.02524 |
قاعدة البيانات: | arXiv |
DOI: | 10.1007/s10955-022-02875-w |
---|