A New Mathematical Tool For Artificial Intelligence Borrowed From Physics

This research aims to increase our understanding¬† and our mathematical control of “natural” (i.e.”spontaneous/emergent”) information processing skills shown by Artificial Intelligence (AI), namely by neural networks and learning machines. Indeed AI is experiencing a “magic moment” as¬†finally theorists have been overwhelmed by “big data” that can be used to train these networks and¬† we can check their capabilities concretely.

Among a plethora of variations on theme, in particular a bulk of algorithms overall termed “Deep Learning” is showing impressive successes in several fields, ranging from scientific applications (e.g. statistical learning and feature extraction from high dimensional data for health care) to more applied ones (e.g. image and video processing and/or natural language processing).¬†As an immediate consequece of these recent¬†great triumphs¬†(see e.g. [1]), the quest for a¬†deeper (mathematical) control on these systems is continuously raising and we aim to contribute to construct a “rationale” for Deep Learning by taking advantage of methods and techniques typical of Theoretical Physics.
Indeed in the past decades Theoretical Physics has been heavily involved in¬†the mathematical formalization of the emergent/spontaneous¬†properties shown by neural networks as, for instance, distributed¬†memory, pattern classification, feature extraction, multitasking¬†capabilities and much more. The bulk of contributions came from¬†Statistical Mechanics and Stochastic Processes: the former¬†(Statistical Mechanics) has been used to paint the “phase diagrams”¬†of¬† crucial networks in statistical learning (e.g. restricted¬†Boltzmann machines) as well as in pattern recognition (e.g. Hopfield¬†neural networks) while the latter (Stochastic Processes) has been¬†naturally adapted to describe the dynamical evolution of the¬†(artificial) neurons and synapses building up the aforementioned¬†networks.

We have proved that the mathematical framework(s) stemming from Theoretical Physics can be enlarged in order to include both classical and relativistic mechanics too: in particular, for these models (i.e. Boltzmann machines and Hopfield networks), the variational principle usually underlying any minimization of a cost function (in their learning/retrieval algorithms) can be shown to coincide sharply with the Least Action Principle: as a natural consequence, the equations for the evolution of the order parameters (e.g. Mattis overlaps with the stored patterns, etc.) in the space of the tunable parameters (e.g. noise level, load of the net, etc..) do coincide naturally with the equations of motion as prescribed by Lagrangian Mechanics in Physics and this allows drawing a number of conclusions.

At first, this bridge between the mathematics involved in a rationale for AI and Lagrangian mechanics allows to import an arsenal of mathematical weapons ready to be used by researchers working in machine learning and neural networks: for instance dynamical instabilities of the network’s evolution can now be inspected by classical Hopf bifurcation theory and conserved quantities -if present- can be studied by inspecting symmetries √† la Noether.
Then, focusing on the priorities in AI research, that is Deep Learning, through the perspective we offer to tackle the problem, it shines clearly¬†¬†that Boltzmann machines and Hopfield nets play solely as the “classical¬†limit” (storing just pairwise correlation functions of the learnt/retrieved patterns of information) of a much¬†broader theory (i.e., the relativistic extension), where all the higher¬†order correlations functions are properly accounted. It is worth pointing out that the relativistic generalization shows several¬†“deep-learning-like” skills: beyond the development of the general¬†mechanic approach to neural networks, in the paper we have extensively shown (both¬†analytically and numerically) how the relativistic extension¬†outperforms w.r.t. the “classical limit”.
Next steps in this branch of research will be achieved by a  systematic exploration of the proposed relation among AI and Lagrangian Mechanics with the hope that this analogy can act as a little Pandora box: we plan to report soon our findings.
These findings are described in the article entitled A new mechanical approach to handle generalized Hopfield neural networks, recently published in the journal Neural Networks. This work was conducted by Adriano Barra from the Università del Salento, the INFN, Istituto Nazionale di Fisica Nucleare, and GNFM-INdAM, Gruppo Nazionale per la Fisica Matematica, and Matteo Beccaria and Alberto Fachechi from the Università del Salento and INFN, Istituto Nazionale di Fisica Nucleare.
Reference:
  1. LeCun, Yann, Yoshua Bengio, and Geoffrey Hinton. “Deep learning.”¬†nature¬†521.7553 (2015): 436.¬†

Speak Your Mind!

READ THIS NEXT

Airport Noise Modelling For Strategic Environmental Impact Assessment Of Aviation

Aircraft noise is usually the main environmental concern for communities living near airports.¬† Among the primary effects of aircraft noise, […]

Toddlers Prefer Those Who Win Fairly But Are Averse To Those Who Win By Force

It is a universal truth that humans are social animals. Humans necessarily exist in societies complete with norms of behavior, […]

Does Air Force One Really Need A New $4 Billion Upgrade?

In the fall of 2016, Donald Trump proved that with the help of his Twitter account, he could do anything, […]

The Religion Developed In Jamaica

The religion developed in Jamaica is Rastafari or the Rasta movement. This type of religion is relatively young, developed in […]

Implication Of Global Warming On Electricity Demand In Australia

The increase in temperatures as a result of global warming has serious implications for electricity demand. Global warming, which is […]

A New Approach To The Electrochemical Detection Of Organophosphorus Pesticides

The organophosphorus pesticides (OPs) are synthetic esters, amides, and thiol derivative of the phosphoric, phosphorotioic, and phosphonothioic acids (Fig. 1) […]

Towards A Feasible Deployment Of Solar Energy Technologies

The use of fossil fuels is being questioned for future energy requirements because of its environmentally hazardous substance emissions, price […]