Synergies

This page collects some of the functional ⇔ biological synergies present in the Axon framework, where properties at the different levels of analysis involved in computational cognitive neuroscience converge together. These synergies support the idea that working across these different levels of analysis can help solve the complex problem of understanding how the brain supports the range of human cognitive functions, ultimately including full general intelligence.

• Bidirectional connectivity among excitatory neurons, which is a prominent and well-established property of the neocortex, plays a critical functional role in both constraint satisfaction and error-driven learning, and a wide range of cognitive functionality including top-down attention, imagery, and many aspects of perception.

• Extensive inhibition implemented by specialized types of interneurons in the neocortex is necessary for controlling the runaway positive feedback loops that arise from bidirectional connectivity, and it also produces significant benefits for attention learning by imposing competition that helps decorrelate neural responses and maximize the robustness of distgributed representations.

• Furthermore the specific form of inhibition that works best for learning across a wide range of models depends critically on having a mix of fast and slow components, which is nicely consistent with the properties of the two dominant interneuron types that drive this inhibition in the neocortex (PV and SST, respectively).

• The complementary and synergistic properties of the NMDA and GABA-B channels are critical for the stable activation needed for learning and processing in spiking networks, which in turn is essential for supporting conscious awareness.

• The detailed biology of the hippocampus, including high levels of inhibition and broad diffuse connectivity, fit together with its unique role in rapidly learning new episodic memories, and the remarkable data from patient HM who had his hippocampus resected to prevent intractable epilepsy. The computational models of the hippocampus (hippocampus simulation)) show how these biological details produce high levels of pattern separation which keep memories highly distinct, and thus enable rapid learning without creating catastrophic levels of interference.

• The detailed biology of the connections between dopamine, basal ganglia, and prefrontal cortex fit together with the computational requirements for making decisions based on prior reward history, and learning what information is important to hold on to, versus what can be ignored. These principles are captured in the Rubicon framework, which show how the dopamine system can exhibit a kind of time travel needed to translate later utility into an earlier decision of what information to maintain. And the interaction between the basal ganglia and the prefrontal cortex enables basal ganglia decisions to influence what is maintained and acted upon in the prefrontal cortex. There are a lot of pieces here, but the fact that they all fit together so well into a functional model — and that many aspects of them have withstood the test of direct experimentation — makes it that much more likely that this is really what is going on.

• The detailed biology of the basal ganglia, supporting graded, parallel processing and associated credit assignment mechanisms allows the system to efficiently search through high-dimensional state and action sensorimotor space to solve the reinforcement learning problem.