Biology subjects

Rosen, M. C.

Publications and source records attributed to Rosen, M. C..

3 recordsLinked to original sources

bioRxiv2022

Rapid learning with highly localized synaptic plasticity

The brains of all animals are plastic, allowing us to form new memories, adapt to new environments, and to learn new tasks. What is less clear is how much plasticity is required to perform these cognitive functions: does learning require widespread plasticity across the brain, or can learning occur with more rigid networks, in which plasticity is highly localized? Here, we use biologically-inspired recurrent neural network (RNN) models to show that rapid multitask learning can be accomplished in reservoir-style networks, in which synaptic plasticity is sparse and highly localized. Crucially, only RNNs initialized with highly specific combinations of network properties, such as topology, normalization and reciprocal connection strength, are capable of such learning. Finally, we show that this rapid learning with localized plasticity can be accomplished with purely local error signals, without backpropagation, using a reinforcement learning setup. This work suggests that rapid learning in artificial (and potentially biological) agents can be accomplished with mostly-rigid networks, in which synaptic plasticity is highly constrained.

neuroscience↗

bioRxiv2020

Primate neuronal connections are sparse as compared to mouse

The mouse and macaque primary visual cortices are foundational models of cortical functioning, particularly at the level of single neurons. Therefore, detailing differences in how individual neurons connect across these species would inform models of cortical functioning and of how brains evolve. However, existing comparisons are limited, measuring synapse density without regard to where synapses are made or on what types of neurons. We use large volume electron microscopy to address this gap, reconstructing a total of 7735 synapses across 160 total neurons (146 excitatory, 14 inhibitory) from adult Rhesus macaque and mouse Layer 2/3 of primary visual cortex (V1). We find that primate connections are broadly sparse: primate excitatory and inhibitory neurons received 3-5 times fewer spine and somatic synapses with lower ratios of excitatory to inhibitory synapses than mouse equivalents. However, despite reductions in absolute synapse number, patterns of axonal innervation were preserved: inhibitory axons sparsely innervated neighboring excitatory neurons in macaque and mouse at similar rates and proportions. On the output side, most excitatory axons in mice myelinated close to the soma (81%) while most primate axons (68%) did not. Interestingly, primate axons, but not mouse axons, that myelinated had 3.3 fold more axon initial segment synapses than axons that did not myelinate, suggesting differential inhibitory control of long distance output in primate brains. Finally, we discover that when artificial recurrent neural networks (RNNs) are constrained by the metabolic cost of creating and maintaining synapses, increasing the number of nodes (e.g. neurons) as networks optimize for a cognitive task, reduces the number of connections per node, similar to primate neurons as compared to mice. One Sentence SummaryUsing large volume serial electron microscopy, we show that primate cortical neural networks are sparser than mouse and using recursive neural nets, we show that energetic costs of synaptic maintenance could underlie this difference.

neuroscience↗

bioRxiv2020

Distributed functions of prefrontal and parietal cortices during sequential categorical decisions

The ability to compare sequential sensory inputs is crucial for solving many behavioral tasks. To understand the neuronal mechanisms underlying sequential decisions, we compared neuronal responses in the prefrontal cortex (PFC) and the lateral and medial intra-parietal (LIP and MIP) areas in monkeys trained to decide whether sequentially presented stimuli were from matching (M) or nonmatching (NM) categories. We found that PFC leads the M/NM decision process relying on nonlinear neuronal integration of sensory and mnemonic information, whereas LIP and MIP are more involved in sensory evaluation and motor planning, respectively. Furthermore, multi-module recurrent neural networks trained on the same task exhibited the key features of PFC and LIP encoding, including nonlinear integrative encoding in the PFC-like module which was crucial for M/NM decisions. Together, our results illuminate the relative functions of LIP, PFC, and MIP in sensory, cognitive and motor functions, and suggest that nonlinear integration of task-related variables in PFC is important for mediating sequential decisions.

neuroscience↗