Switched neural networks

Claim 2

A neural network including a plurality of neurons, each comprising non-linear amplifying means including input means and an output terminal, the output terminal of each being connected to the input means of other amplifying means of the plurality for supplying its output as an input to said other amplifying means, each of said neurons being characterized by the inclusion of capacitor means and switching means in said input means for storing an input signal for sampling and for weighting.

Claim 3

A neural network in accordance with claim 1 further characterized in that the input means of each of the amplifying means comprises a plurality of input terminals each for receiving an input signal, an input bus, and a separate coupling means between each input signal terminal and the input bus, some of said coupling means being inverting and others of said coupling means being non-inverting.

Claim 4

A neural network in accordance with claim 2 further characterized in that each inverting coupling means comprise a pair of switches serially connected between an input terminal and an input bus, and capacitor means connected between a node between the two switches and a point of reference potential, the two switches being adapted to be closed sequentially, in non-overlapping intervals.

Claim 5

The neural network of claim 2 in which each non-inverting coupling means comprises a first switch, capacitor means, and a second switch serially connected between an input terminal and an input bus, a third switch connected between a node between the first switch and the capacitor means and a point of reference potential, and a fourth switch connected between a node between the capacitor means and the second switch and a point at said reference potential, the first and fourth switches being adapted to be closed synchronously, the second and third switches being adapted to be closed synchronously, subsequent to the closing of the first and fourth switches, in non-overlapping intervals.

Claim 6

The neural network of claim 2 in which each non-inverting coupling means comprises a first switch, capacitor means and a second switch serially connected between an input terminal and an input bus, a third switch connected between a node between the first switch and the capacitor means and a point of reference potential, and a fourth switch connected between a node between the capacitor means and the second switch and a point of said reference potential, and each inverting coupling means comprises fifth and sixth switches serially connected between an input terminal and the input bus and a second capacitor means connected between a node between the fifth and sixth switches and a point at said reference potential, the first, fourth, and fifth switches each being adapted to be closed for a first interval, and the second, third and sixth being adapted to be closed for a second interval, non-overlapping with the first interval.

Claim 7

The neural network of claim 5 in which the amplifying means include a third capacitance means in parallel with a seventh switch connected between the input and output of the amplifying means, the seventh switch being adapted to be closed at a third interval spaced between and non-overlapping said first and second intervals.

Claim 8

The neural network of claim 6 in which the first and second capacitor means are adapted for weighting input signals applied to their associated input terminals.