Replica Symmetry Breaking in Dense Hebbian Neural Networks
نویسندگان
چکیده
Understanding the glassy nature of neural networks is pivotal both for theoretical and computational advances in Machine Learning Theoretical Artificial Intelligence. Keeping focus on dense associative Hebbian networks, purpose this paper two-fold: at first we develop rigorous mathematical approaches to address properly a statistical mechanical picture phenomenon {\em replica symmetry breaking} (RSB) these then -- deepening results stemmed via routes aim inspect glassiness} that they hide. In particular, regarding methodology, provide two techniques: former an adaptation transport PDE case, while latter extension Guerra's interpolation breakthrough. Beyond coherence among results, either symmetric one-step breaking level description, prove Gardner's identify maximal storage capacity by ground-state analysis Baldi-Venkatesh high-storage regime. second part investigate structure networks: contrast with scenario (RS), RSB actually stabilizes spin-glass phase. We report huge differences w.r.t. standard pairwise Hopfield limit: it known possible express free energy network as linear combination energies hard spin glass (i.e. Sherrington-Kirkpatrick model) soft (the Gaussian or "spherical" model). This no longer true when interactions are more than (whatever RS RSB): solely survives, proving diversity underlying glassiness networks.
منابع مشابه
Replica Symmetry Breaking in Bipartite Spin Glasses and Neural Networks
Some interesting recent advances in the theoretical understanding of neural networks have been informed by results from the physics of disordered many-body systems. Motivated by these findings, this work uses the replica technique to study the mathematically tractable bipartite Sherrington-Kirkpatrick (SK) spin glass model, which is formally similar to a Restricted Boltzmann Machine (RBM) neura...
متن کاملReplica Symmetry Breaking, Scaling Theory, and Sensor Networks
Device and sensor networks are shaping many activities in our society. These networks are being deployed in a growing number of applications as diverse as agricultural management, industrial controls, crime watch, and military applications. Indeed, sensor networks can be considered as a promising technology with a wide range of potential future markets [1]. Still, for all the promise, it is oft...
متن کاملReplica symmetry breaking in the minority game
We extend and complete recent work concerning the analytic solution of the minority game. Nash equilibria (NE) of the game have been found to be related to the ground states of a disordered hamiltonian with replica symmetry breaking (RSB), signalling the presence of a large number of them. Here we study the number of NE both analytically and numerically. We then analyze the stability of the rec...
متن کاملSpontaneous Symmetry Breaking in Neural Networks
We propose a framework to understand the unprecedented performance and robustness of deep neural networks using field theory. Correlations between the weights within the same layer can be described by symmetries in that layer, and networks generalize better if such symmetries are broken to reduce the redundancies of the weights. Using a two parameter field theory, we find that the network can b...
متن کاملذخیره در منابع من
با ذخیره ی این منبع در منابع من، دسترسی به آن را برای استفاده های بعدی آسان تر کنید
ژورنال
عنوان ژورنال: Journal of Statistical Physics
سال: 2022
ISSN: ['0022-4715', '1572-9613']
DOI: https://doi.org/10.1007/s10955-022-02966-8