**3.2 Smith's biologically plausible neural networks**

We leverage and build on another significant body of work by James E. Smith on reverse-architecting the brain with the goal of replicating it using off-the-shelf digital CMOS silicon [7, 8, 10, 55]. He proposes a new category of spiking neural networks called temporal neural networks (TNNs) [7] that are architected to mimic the key attributes of biological neocortex (see **Figure 3**). Unlike DNNs performing real-valued tensor-based computations supported by global back propagation for stochastic gradient descent, TNNs employ spiking neurons that encode and process inputs as timings of events or spikes, and can learn using local biologically plausible algorithm called spike timing dependent plasticity (STDP). TNNs also differ from most other spiking neural networks (SNNs) that encode values based on the rate of spikes as opposed to spike timing. **Figure 3** shows this taxonomy. Consequently TNNs, unlike most other ANNs, are more truly "neuromorphic" due to their strong adherence to biological plausibility.

Smith has also developed a TNN-based architecture for the cortical columns and demonstrated the capability for doing unsupervised learning with rapid convergence, and the capability to do online, concurrent, and continuous learning [10, 55]. Formulating the underlying mathematical basis for implementing TNNs, Smith has proposed a new algebra called space-time (temporal) algebra [8]. Based on this new temporal algebra (instead of Boolean algebra), temporal functions can be implemented very efficiently in hardware by re-purposing the current digital logic gates to make use of time as a "free" resource for both encoding and processing of information.

#### **Figure 3.**

*Neural network taxonomy contrasting neocortex-inspired temporal neural networks (TNNs) with other artificial neural networks (ANNs). In contrast to other ANNs including deep neural networks (DNNs),TNNs incorporate attributes with strong adherence to biological plausibility, including spiking neuron model, temporal coding of inputs, and local simple spike timing dependent plasticity (STDP) learning rules.*

*Cortical Columns Computing Systems: Microarchitecture Model, Functional Building Blocks… DOI: http://dx.doi.org/10.5772/intechopen.110252*
