Search papers, labs, and topics across Lattice.
This paper introduces JEPAWG, a novel architecture that maps coupling constants directly to flow weights in lattice quantum field theories, enabling the extraction of physical observables from neural network parameters. By leveraging a well-understood synthetic dataset, the authors demonstrate that JEPAWG can accurately recover the intrinsic dimensionality of the underlying manifold, identify phase transitions, and align with known physical exponents. The findings suggest that network weights can serve as interpretable physical observables, offering a new perspective on neural network interpretability in physics.
JEPAWG reveals that neural network weights can be treated as new physical observables, effectively bridging the gap between machine learning and lattice quantum field theory.
Lattice field theory is the workhorse of non-perturbative physics, used to simulate phenomena from the strong nuclear force to critical phenomena in materials. Its Boltzmann distributions are parametrized analytically by coupling constants, but these bare parameters are weak predictors of observables -- extracting physics typically requires extensive simulation. While normalizing flows have emerged as effective samplers at fixed couplings, it remains difficult to interpret what these networks have learned. This raises a natural question: can the physics be read off directly from the flow network parameters themselves, and can those parameters be generated for unseen theories? We propose lattice field theory as a testbed for neural network interpretability: because the target physics is qualitatively well-understood and smoothly varying, it provides ideal synthetic data with known ground truth. To this end, we introduce JEPAWG, a Joint-Embedding Predictive Architecture-based Weight Generator that maps couplings directly to flow weights via a learned latent space. On a scalar theory at lattices of size $6^2$ to $11^2$, the JEPAWG latent space recovers the correct intrinsic dimension of the underlying manifold, locates the phase transition, and encodes a finite-size shift aligned with the 2D Ising exponent $\nu \approx 1$, allowing us to uncover physical structure by studying the network weights alone. This suggests the fascinating idea of treating the network weights as a new type of physical observable. As a generator, JEPAWG also interpolates and extrapolates to unseen couplings effectively and remains robust to weight-space discontinuities introduced by multi-seed training data, outperforming PCA, AE, and VAE baselines.
