Finally, in bipartite models of the CA3-DG circuit in the hippocampus, a wiring strategy called "quasi-indexing" can boost memory capacity and protect stored information even when some neurons are lost, offering new insights into how the brain encodes and safeguards memories.

This prediction holds across various biologically relevant scenarios, including the storage of independent patterns in both classical and dendritic networks, as well as the storage of correlated patterns clustered around prototypes that represent concepts.

Although heterogeneity in neuron coding levels and inward connection counts (in-degrees) generally reduces capacity, maximal capacity is retained when these two parameters are correlated.

We derived an analytical formula for the maximal memory capacity of networks with heterogeneous neuron activation rates (coding levels) and arbitrary connectivity architectures. Using this result, we made normative predictions about the properties that maximize capacity in brain-inspired networks.

Congrats Trevor! We are excited to welcome you to WashU :)

Here, we began to bridge computational and algorithmic perspectives on sensory coding, integrating multimodal function. In an upcoming study, we will further connect computational and implementational perspectives, moving toward a full integration of Marr's triad. Stay tuned!

(3) Finally, it provides a normative explanation for a class of observed 𝘮𝘶𝘭𝘵𝘪𝘮𝘰𝘥𝘢𝘭 receptive fields while integrating previous knowledge of 𝘶𝘯𝘪𝘮𝘰𝘥𝘢𝘭 processing as a special case within a broader, unitary framework.

(2) The study demonstrates how efficient and predictive 𝘤𝘰𝘮𝘱𝘶𝘵𝘢𝘵𝘪𝘰𝘯𝘴 are concurrently supported by a shared neural substrate and how different network components can implement these computations at the 𝘢𝘭𝘨𝘰𝘳𝘪𝘵𝘩𝘮𝘪𝘤 level.

(1) The study shows that in feedback-modulated canonical networks, efficient coding automatically generates predictive codes, where stimuli from one modality can serve as predictors for stimuli in another modality. Cross-modal 𝘱𝘳𝘦𝘥𝘪𝘤𝘵𝘪𝘷𝘦 computations are an emergent property of 𝘦𝘧𝘧𝘪𝘤𝘪𝘦𝘯𝘵 codes.