Why Synaptic Intelligence?
Synaptic intelligence is the heart of Nara Logics’ AI platform. It’s what makes continual learning and extremely accurate results possible.
The synaptic intelligence technology we developed is based on recent discoveries about the way brain circuits work. We apply the logic about how biological synaptic networks associate and process information to enterprise-scale, “Cognitive AI.”
What makes synaptic intelligence ideal for next-gen AI:
Explicitly represents information
Unlike traditional AI systems, which are like black boxes that process information using hidden nodes, every node in a synaptic network represents an explicit object or feature. Because everything is transparent, the reasons that lead to every AI recommendation are clear.
The network’s recurrent structure and the way information is cross-connected allow it to evaluate possibilities across multiple dimensions. Instead of being serial, decisions can be expressed in a loopy belief network where weights are set through biological learning rules. The resulting truth is much more accurate than what piecemeal processing, sequential binary decisions or matrix operations can deliver.
Accurately infers relationships
The weights assigned to connections between nodes are key to inferring additional relationships. Unlike traditional neural networks, which use supervised learning to set those weights, we use synaptic learning rules at the cellular and local network levels to determine when and how strongly information should be connected, and automatically organization that information.
Just like our brains, which activate a small number of neurons at any given time, rather than the massive numbers previously assumed, Nara Logics’ synaptic network eliminates the need to compute over large matrices that are mostly empty. This enables efficient storage capacity and a natural, unambiguous way to directly read encoded information.
Learns in place
Synaptic networks don’t need to be retrained to keep up with evolving knowledge. The network format allows new “facts” to be added simply by updating connections between two nodes, and then re-balancing a small number of associated connections. This is in stark contrast to traditional neural networks, which need to be burned down and retrained to adapt to new data or requirements.