Making BREAD: Biomimetic Strategies for Artificial Intelligence Now and in the Future

Abstract

The Artificial Intelligence (AI) revolution foretold of during the 1960s is well underway in the second decade of the twenty first century. Its period of phenomenal growth likely lies ahead. AI-operated machines and technologies will extend the reach of Homo sapiens far beyond the biological constraints imposed by evolution: outwards further into deep space, as well as inwards into the nano-world of DNA sequences and relevant medical applications. And yet, we believe, there are crucial lessons that biology can offer that will enable a prosperous future for AI. For machines in general, and for AI's especially, operating over extended periods or in extreme environments will require energy usage orders of magnitudes more efficient than exists today. In many operational environments, energy sources will be constrained. The AI's design and function may be dependent upon the type of energy source, as well as its availability and accessibility. Any plans for AI devices operating in a challenging environment must begin with the question of how they are powered, where fuel is located, how energy is stored and made available to the machine, and how long the machine can operate on specific energy units. While one of the key advantages of AI use is to reduce the dimensionality of a complex problem, the fact remains that some energy is required for functionality. Hence, the materials and technologies that provide the needed energy represent a critical challenge toward future use scenarios of AI and should be integrated into their design. Here we look to the brain and other aspects of biology as inspiration for Biomimetic Research for Energy-efficient AI Designs (BREAD).

Multimodal Representation Learning for Place Recognition Using Deep Hebbian Predictive Coding

“…In the case of the JMVAE-kl network which learns unimodal representations in parallel to, and in support of, the joint modal distribution during training, the additional encoder-decoder network pairs are also disregarded during inference. In a purely software-based system this inefficiency is not an issue, however, as we look toward the future of embedded machine learning, particularly within robotics and edge based applications, we anticipate the increasing adoption of energy efficient hardware platforms, such as neuromorphic devices Krichmar et al (2019) . The immediate practical advantage in adopting the PCN approach, therefore, is that the algorithm is highly amenable to hardware optimisation through parallel distributed learning and processing.…”

Design Principles for Neurorobotics

“…Rather it frees up the system from expending energy and computational resources on some functions, while allowing it to concentrate on other functionality. Saving energy is a recurring theme in biology since biological organisms are under tight metabolic constraints (Beyeler et al, 2019 ; Krichmar et al, 2019 ). However, there is a trade-off that comes with efficiency.…”

“…Therefore, the nervous system must encode information as cheaply as possible. The brain operates on a mere 20 watts of power, approximately the same power required for a ceiling fan operating at low speed (Krichmar et al, 2019 ). Although being severely metabolically constrained is at one level a disadvantage, evolution has optimized brains in ways that lead to incredibly efficient representations of important environmental features that are distinctly different from those employed in current digital computers.…”

“…This architectural design also follows the Efficiency through Cheap Design principle described in Section 2.2. Neuromorphic hardware architectures use orders of magnitude less power than conventional computing by not relying on a synchronous clock and using spiking elements (Krichmar et al, 2019 ). A neuron uses most of its energy when it fires an action potential and when an action potential is processed at the synapse.…”

A Spiking Neural Network Model of Rodent Head Direction Calibrated With Landmark Free Learning

“…The primary motivation for this work is toward a deeper understanding of the complete head direction system of the rodent through the integration of components modeled at different levels of abstraction; namely, spiking neural attractor network models, deep learning based generative and discriminative models, and simulated robotic embodiment. However, the integration of models at multiple levels of abstraction also provides a framework for how energy efficient neuromorphic hardware components (Krichmar et al, 2019 ) can be usefully integrated into mobile robotic applications in the near future. We contend that to fully exploit this biologically inspired computing paradigm requires continued biomimetic study of fundamental neuroscience as epitomized in the field of neurorobotics.…”