• Corpus ID: 240419711

Recurrent neural network models for working memory of continuous variables: activity manifolds, connectivity patterns, and dynamic codes

  title={Recurrent neural network models for working memory of continuous variables: activity manifolds, connectivity patterns, and dynamic codes},
  author={Christopher J. Cueva and A. Ardalan and Misha Tsodyks and Ning Qian},
Many daily activities and psychophysical experiments involve keeping multiple items in working memory. When the items take continuous values (e.g., orientation, direction, contrast, length, weight, loudness) they must be stored in a continuous structure of appropriate dimensions. We investigate how such a structure might be represented in neural circuits by training recurrent networks to report two previously flashed stimulus orientations. We find that the activity manifold for the two… 

Figures from this paper


Synaptic mechanisms and network dynamics underlying spatial working memory in a cortical network model.
A network model endowed with a columnar architecture and based on the physiological properties of cortical neurons and synapses finds that recurrent synaptic excitation should be primarily mediated by NMDA receptors, and that overall recurrent synaptic interactions should be dominated by inhibition.
Stimuli Reduce the Dimensionality of Cortical Activity
It is found that dimensionality grows linearly with ensemble size, and grows significantly faster during ongoing activity compared to evoked activity, and a simple theory is presented that predicts the existence of an upper bound on dimensionality.
Short-Term Facilitation may Stabilize Parametric Working Memory Trace
It is proposed that short-term synaptic facilitation in recurrent connections significantly improves the robustness of the model by slowing down the drift of activity bump, rendering the network suitable as a model of WM.
Context-dependent computation by recurrent dynamics in prefrontal cortex
This work studies prefrontal cortex activity in macaque monkeys trained to flexibly select and integrate noisy sensory inputs towards a choice, and finds that the observed complexity and functional roles of single neurons are readily understood in the framework of a dynamical process unfolding at the level of the population.
A theory of multineuronal dimensionality, dynamics and measurement
A theory is presented that reveals conceptual insights into how task complexity governs both neural dimensionality and accurate recovery of dynamic portraits, thereby providing quantitative guidelines for future large-scale experimental design.
Flexible Control of Mutual Inhibition: A Neural Model of Two-Interval Discrimination
A simple mutual-inhibition network model is presented that captures all three task phases within a single framework and integrates both working memory and decision making because its dynamical properties are easily controlled without changing its connectivity.
Rotational Dynamics Reduce Interference Between Sensory and Memory Representations
It is shown that sensory representations rotate in neural space over time, to form an independent memory representation, thus reducing interference with future sensory inputs, and may be a general principle, by which the cortex protects memories of prior events from interference by incoming stimuli.
A neural network that finds a naturalistic solution for the production of muscle activity
This work explored the hypothesis that motor cortex reflects dynamics appropriate for generating temporally patterned outgoing commands and trained recurrent neural networks to reproduce the muscle activity of reaching monkeys to formalize this hypothesis.
Emergence of functional and structural properties of the head direction system by optimization of recurrent neural networks
The results show that optimization of RNNs in a goal-driven task can recapitulate the structure and function of biological circuits, suggesting that artificial neural networks can be used to study the brain at the level of both neural activity and anatomical organization.
Inferring Stimulus Selectivity from the Spatial Structure of Neural Network Dynamics
The absence of a detailed spatial map of afferent inputs and cortical connectivity does not limit the ability to design spatially extended stimuli that evoke strong responses, and a complementary approach to principal component analysis is developed.