Supervised and unsupervised learning using a fully-plastic all-optical unit of artificial intelligence based on solitonic waveguides

Abstract

The software implementations of neuronal systems have shown great effectiveness, even if the natural hardware separation between the processing and memory areas in computers slows down the analysis capacity. To overcome these limitations, new hardware configurations are moving towards neuromorphic models, capable of unifying the processing/memory dichotomy. Recently, integrated photonic X-junctions formed by waveguides written by spatial solitons have shown the ability to perform supervised learning. The solitonic technology, compared to the traditional one, offers the advantage of realizing plastic circuitry, a typical characteristic of biological neural networks. This work extensively studies both supervised and unsupervised learning of photonic soliton X-junctions. By exploiting the plasticity of the nonlinear refractive index at the base of the soliton formation, X-junctions can readdress their behaviours forwarding data to different outputs. In this article, we will extend the state-of-the-art: starting from supervised learning, for which all possible cases are now investigated, a material sensitive to the transported signals will be introduced to allow the junction to carry out unsupervised learning. In this way, the junction autonomously recognises the transported signals without the external intervention of the operator. Learning and memory now physically coincide in fact, learning means that the junction slowly switches based on the information sent; any further unknown information sent will find the junction in the modified state which corresponds to the learned information and will be recognised as well (reasoning based on comparison with stored information).

Stigmergic electronic gates and networks

“…innovative technology based on junctions with an arbitrary number of tributary channels. A practical example is represented by the neurons [30][31][32][33], which act as logic gates with N synaptic inputs and outputs. The development of complex reasoning is therefore ensured by the ability to manage single switching ports with many inputs and many outputs, with a well-controlled energy management plan, uniformly distributed across the entire processing network.…”

Optical Soliton Neural Networks

“…In 2018, a collaboration between Sapienza and Nanyang Technological University in Singapore demonstrated that X-junctions formed by soliton waveguides learn information [55]. Recently, it has been shown that X-junctions can perform both supervised and unsupervised learning, behaving as if they were neurons that fully exploit the plasticity of the substrate both to write the circuit and to post-modification based on the evolution of the system [74]. By exploiting the X junctions as elementary units, it is possible to create complex neural networks capable of storing information as specific trajectories within the circuit network [75].…”

“…For different angles, the node is characterized by an area that is too limited which determines a low coupling between the waveguides. The soliton neuron can perform supervised and unsupervised learning tasks [55,74]. From a theoretical point of view, supervised learning is performed using a fundamental truth, or in other words, there is prior knowledge of what the output values to learn should be [81][82][83].…”

“…The X junction is modified using a feedback system that locally alters the refractive index contrast, depending on the information received, through successive cycles (Figure 7). This mechanism is fully explained in [74], where a numerical code FDTD solving the nonlinear equation (Eq. ( 1)) reported below, shows the morphological evolution of the neuron (see Figure 7), index of learning what is happening.…”

Novel Model Based on Artificial Neural Networks to Predict Short-Term Temperature Evolution in Museum Environment

“…This is mainly due to the limited availability of long-term series of key environmental variables at the sites where the artifacts are exhibited or stored. Although machine learning algorithms allowed to achieve important results in the processing automation and in the improvement of forecasting techniques, their complexity results into long training periods and into the demand for large available datasets [2][3][4]. When only short time series are available, complex networks tend to memorize processes rather than learn them, then failing to identify sudden variations or trends.…”