Subscribe to the gold package and get unlimited access to Shamra Academy
Register a new userReplica Symmetry Breaking and the Kuhn-Tucker Cavity Method in simple and multilayer Perceptrons
194
0
0.0
(
0
)
Authors
F. Gerl
Ask ChatGPT about the research
No Arabic abstract
Within a Kuhn-Tucker cavity method introduced in a former paper, we study optimal stability learning for situations, where in the replica formalism the replica symmetry may be broken, namely (i) the case of a simple perceptron above the critical loading, and (ii) the case of two-layer AND-perceptrons, if one learns with maximal stability. We find that the deviation of our cavity solution from the replica symmetric one in these cases is a clear indication of the necessity of replica symmetry breaking. In any case the cavity solution tends to underestimate the storage capabilities of the networks.
rate research
Read More
Directed polymers on 1+1 dimensional lattices coupled to a heat bath at temperature $T$ are studied numerically for three ensembles of the site disorder. In particular correlations of the disorder as well as fractal patterning are considered. Configurations are directly sampled in perfect thermal equilibrium for very large system sizes with up to $N=L^2= 32768 times 32768 approx 10^{9}$ sites. The phase-space structure is studied via the distribution of overlaps and hierarchical clustering of configurations. One ensemble shows a simple behavior like a ferromagnet. The other two ensembles exhibit indications for complex behavior reminiscent of multiple replica-symmetry breaking. Also results for the ultrametricity of the phase space and the phase transition behavior of $P(q)$ when varying the temperature $T$ are studied. In total, the present model ensembles offer convenient numerical accesses to comprehensively studying complex behavior.
We study the set of solutions of random k-satisfiability formulae through the cavity method. It is known that, for an interval of the clause-to-variables ratio, this decomposes into an exponential number of pure states (clusters). We refine substantially this picture by: (i) determining the precise location of the clustering transition; (ii) uncovering a second `condensation phase transition in the structure of the solution set for k larger or equal than 4. These results both follow from computing the large deviation rate of the internal entropy of pure states. From a technical point of view our main contributions are a simplified version of the cavity formalism for special values of the Parisi replica symmetry breaking parameter m (in particular for m=1 via a correspondence with the tree reconstruction problem) and new large-k expansions.
Simulational studies of spin glasses in the last decade have focussed on the so-called replicon exponent $alpha$ as a means of determining whether the low-temperature phase of spin glasses is described by the replica symmetry breaking picture of Parisi or by the droplet-scaling picture. On the latter picture, it should be zero, but we shall argue that it will only be zero for systems of linear dimension $L > L^*$. The crossover length $L^*$ may be of the order of hundreds of lattice spacings in three dimensions and approach infinity in 6 dimensions. We use the droplet-scaling picture to show that the apparent non-zero value of $alpha$ when $L < L^*$ should be $2 theta$, where $theta$ is the domain wall energy scaling exponent, This formula is in reasonable agreement with the reported values of $alpha$.
We prove the impossibility of recent attempts to decouple the Replica Symmetry Breaking (RSB) picture for finite-dimensional spin glasses from the existence of many thermodynamic (i.e., infinite-volume) pure states while preserving another signature RSB feature --- space filling relative domain walls between different finite-volume states. Thus revisions of the notion of pure states cannot shield the RSB picture from the internal contradictions that rule out its physical correctness in finite dimensions at low temperature in large finite volume.
The fully-connected Ising $p$-spin model has for $p >2$ a discontinuous phase transition from the paramagnetic phase to a stable state with one-step replica symmetry breaking (1RSB). However, simulations in three dimension do not look like these mean-field results and have features more like those which would arise with full replica symmetry breaking (FRSB). To help understand how this might come about we have studied in the fully connected $p$-spin model the state of two-step replica symmetry breaking (2RSB). It has a free energy degenerate with that of 1RSB, but the weight of the additional peak in $P(q)$ vanishes. We expect that the state with full replica symmetry breaking (FRSB) is also degenerate with that of 1RSB. We suggest that finite size effects will give a non-vanishing weight to the FRSB features, as also will fluctuations about the mean-field solution. Our conclusion is that outside the fully connected model in the thermodynamic limit, FRSB is to be expected rather than 1RSB.
Log in to be able to interact and post comments
comments
Fetching comments
Sign in to be able to follow your search criteria
