No, its not another Westworld theory. A chip named Loihi, designed in Intel’s labs was recently presented in a study outlining its features and the potential of its architecture. It doesn’t work exactly like an artificial neural network (ANN) but more like a spiking neural network (SNN). SNN is defined by the researchers as: “a model of computation with neurons as the basic processing elements.”
Compared to previous approaches, this 60mm2 chip made in Intel’s 14nm labs “integrates a wide range of novel features for the field, such as hierarchical connectivity, dendritic compartments, synaptic delays, and, most importantly, programmable synaptic learning rules.
Such systems tend to have enhanced energy characteristics. Compared to more conventional architectures, this chip has enhanced performance in many tasks, making this architecture a valid option for many applications.
The following section from the article was really interesting to me: “Loihi’s flexible learning engine allows one to explore and experiment with various learning methods. We have developed and validated the following networks in pre-silicon FPGA emulation with all learning taking place on chip:
• A single-layer classifier using a supervised variant of STDP, similar to F. Ponulak and A. Kasinski, 4 as the learning method. This network, when trained with local-intensity change-based temporally spike-coded image samples, can achieve 96 percent accuracy on the MNIST dataset using ten neurons, in line with a reference ANN of the same structure.
• Solving the shortest path problem of a weighted graph. Vertices and edges are represented as neurons and synapses, respectively. The algorithm is based on the effects of STDP on a propagating wave-front of spikes.13
• Solving a one-dimensional, non-Markovian sequential decision-making problem. The network learns the decision-making policy in response to delayed reward and punishment feedback similar to R.V. Florian.14
The algorithmic development and characterization of Loihi is just beginning. These proof-of concept examples use only a fraction of the resources and features available in the chip. With Loihi now in hand, our focus turns to scaling and further evaluating these networks.
I really like the concept of neuromorphic chips because they can help us understand the way our brain work and even show us how to deal with certain aspects of cognition. Additionally, it is very important to look at nature for inspiration as it had millions of years to perfect its creations. As a result the brain is a very efficient and powerful processor. As said in the study though, there may be limitations with silicon or the fabrication process until we reach the point at which we will be able to build a processor that is comparable to a biological brain.