David Saad

Rishona Daniels, Duna Wattad, Ronny Ronen et al.

Reservoir computing (RC) is an emerging recurrent neural network architecture that has attracted growing attention for its low training cost and modest hardware requirements. Memristor-based circuits are particularly promising for RC, as their intrinsic dynamics can reduce network size and parameter overhead in tasks such as time-series prediction and image recognition. Although RC has been demonstrated with several memristive devices, a comprehensive evaluation of device-level requirements remains limited. In this paper, we analyze and explain the operation of a parallel delayed feedback network (PDFN) RC architecture with volatile memristors, focusing on how device characteristics -- such as decay rate, quantization, and variability -- affect reservoir performance. We further discuss strategies to improve data representation in the reservoir using preprocessing methods and suggest potential improvements. The proposed approach achieves 95.89% classification accuracy on MNIST, comparable with the best reported memristor-based RC implementations. Furthermore, the method maintains high robustness under 20% device variability, achieving an accuracy of up to 94.2%. These results demonstrate that volatile memristors can support reliable spatio-temporal information processing and reinforce their potential as key building blocks for compact, high-speed, and energy-efficient neuromorphic computing systems.

Alexander Mozeika, Bo Li, David Saad

We study the space of functions computed by random-layered machines, including deep neural networks and Boolean circuits. Investigating the distribution of Boolean functions computed on the recurrent and layer-dependent architectures, we find that it is the same in both models. Depending on the initial conditions and computing elements used, we characterize the space of functions computed at the large depth limit and show that the macroscopic entropy of Boolean functions is either monotonically increasing or decreasing with the growing depth.

Andrey Y. Lokhov, David Saad

In a diffusion process on a network, how many nodes are expected to be influenced by a set of initial spreaders? This natural problem, often referred to as influence estimation, boils down to computing the marginal probability that a given node is active at a given time when the process starts from specified initial condition. Among many other applications, this task is crucial for a well-studied problem of influence maximization: finding optimal spreaders in a social network that maximize the influence spread by a certain time horizon. Indeed, influence estimation needs to be called multiple times for comparing candidate seed sets. Unfortunately, in many models of interest an exact computation of marginals is #P-hard. In practice, influence is often estimated using Monte-Carlo sampling methods that require a large number of runs for obtaining a high-fidelity prediction, especially at large times. It is thus desirable to develop analytic techniques as an alternative to sampling methods. Here, we suggest an algorithm for estimating the influence function in popular independent cascade model based on a scalable dynamic message-passing approach. This method has a computational complexity of a single Monte-Carlo simulation and provides an upper bound on the expected spread on a general graph, yielding exact answer for treelike networks. We also provide dynamic message-passing equations for a stochastic version of the linear threshold model. The resulting saving of a potentially large sampling factor in the running time compared to simulation-based techniques hence makes it possible to address large-scale problem instances.

Bo Li, David Saad

Mean field theory has been successfully used to analyze deep neural networks (DNN) in the infinite size limit. Given the finite size of realistic DNN, we utilize the large deviation theory and path integral analysis to study the deviation of functions represented by DNN from their typical mean field solutions. The parameter perturbations investigated include weight sparsification (dilution) and binarization, which are commonly used in model simplification, for both ReLU and sign activation functions. We find that random networks with ReLU activation are more robust to parameter perturbations with respect to their counterparts with sign activation, which arguably is reflected in the simplicity of the functions they generate.

Bo Li, David Saad

The function space of deep-learning machines is investigated by studying growth in the entropy of functions of a given error with respect to a reference function, realized by a deep-learning machine. Using physics-inspired methods we study both sparsely and densely-connected architectures to discover a layer-wise convergence of candidate functions, marked by a corresponding reduction in entropy when approaching the reference function, gain insight into the importance of having a large number of layers, and observe phase transitions as the error increases.