THERMODYNAMIC COMPUTING
Methods and systems for thermodynamic computing based on the attractor dynamics of volatile dissipative electronics attempting to maximize circuit power consumption. A general model of memristive devices based on collections of metastable switches, adaptive synaptic weights can be formed from a differential pair of memristors and modified according to anti-hebbian and hebbian plasticity. The arrays of synaptic weights can be employed to build a neural node circuit with attractor states that are shown to be logic functions forming a computationally complete set. By configuring the attractor states of the computational building block in different ways, high-level machine learning functions can be demonstrated for real-world applications.
1 . A method for thermodynamic computing, comprising:
modifying adaptive synaptic weights according to anti-hebbian and hebbian plasticity, said adaptive synaptic weights configured from a differential pair of memristors;
configuring at least one neural node circuit with attractor states via an array of said adaptive synaptic weights;
configuring a computational building block from at least one neural node circuit with said attractor states; and
obtaining at least one high-level machine learning function from said computational building block for use in machine learning applications.
2 . The method of claim 1 wherein said attractor states comprise logic functions that form a computationally complete set.
3 . The method of claim 1 wherein said at least one high-level machine learning functions comprises unsupervised clustering.
4 . The method of claim 1 wherein said at least one high-level machine learning functions comprises supervised classification.
5 . The method of claim 1 wherein said at least one high-level machine learning functions comprises unsupervised classification.
6 . The method of claim 1 wherein said at least one high-level machine learning functions comprises complex signal prediction.
7 . The method of claim 1 wherein said at least one high-level machine learning functions comprises unsupervised robotic actuation.
8 . The method of claim 1 wherein said at least one high-level machine learning functions comprises combinatorial optimization of procedures.