[Paper Review] ORGaNICs: A Theory of Working Memory in Brains and Machines

David J. Heeger, Wayne E. Mackey|arXiv (Cornell University)|Mar 16, 2018

Neural dynamics and brain function106 references3 citations

TL;DR

ORGaNICs proposes a biophysically plausible recurrent neural circuit model—Oscillatory Recurrent Gated Neural Integrator Circuits—that explains working memory dynamics in the prefrontal and parietal cortices. By adapting Long Short-Term Memory (LSTM) units into a biologically grounded framework with shunting inhibition via thalamocortical loops, ORGaNICs enables analytically tractable, dynamically variable information encoding and readout, offering a canonical computational framework for brains and machines.

ABSTRACT

Working memory is a cognitive process that is responsible for temporarily holding and manipulating information. Most of the empirical neuroscience research on working memory has focused on measuring sustained activity in prefrontal cortex (PFC) and/or parietal cortex during simple delayed-response tasks, and most of the models of working memory have been based on neural integrators. But working memory means much more than just holding a piece of information online. We describe a new theory of working memory, based on a recurrent neural circuit that we call ORGaNICs (Oscillatory Recurrent GAted Neural Integrator Circuits). ORGaNICs are a variety of Long Short Term Memory units (LSTMs), imported from machine learning and artificial intelligence. ORGaNICs can be used to explain the complex dynamics of delay-period activity in prefrontal cortex (PFC) during a working memory task. The theory is analytically tractable so that we can characterize the dynamics, and the theory provides a means for reading out information from the dynamically varying responses at any point in time, in spite of the complex dynamics. ORGaNICs can be implemented with a biophysical (electrical circuit) model of pyramidal cells, combined with shunting inhibition via a thalamocortical loop. Although introduced as a computational theory of working memory, ORGaNICs are also applicable to models of sensory processing, motor preparation and motor control. ORGaNICs offer computational advantages compared to other varieties of LSTMs that are commonly used in AI applications. Consequently, ORGaNICs are a framework for canonical computation in brains and machines.

Motivation & Objective

• To develop a biologically plausible theory of working memory that extends beyond sustained activity models.
• To address the limitations of traditional neural integrator models in explaining complex delay-period dynamics in prefrontal cortex.
• To unify working memory, sensory processing, and motor control under a single computational framework.
• To provide an analytically tractable model capable of reading out information from time-varying neural responses.

Proposed method

• ORGaNICs is a variant of Long Short-Term Memory (LSTM) units adapted for biological plausibility.
• The model uses oscillatory recurrent dynamics to support transient and sustained activity patterns.
• It incorporates shunting inhibition via a thalamocortical loop to regulate excitation and enable dynamic gating.
• The circuit is implemented using biophysical models of pyramidal cells and synaptic integration.
• Theoretical analysis enables characterization of neural dynamics and information readout at any time point.
• The framework supports real-time decoding of information from non-stationary neural responses.

Experimental results

Research questions

• RQ1 How can working memory be modeled with biologically plausible neural circuits that support complex, time-varying activity patterns?
• RQ2 What mechanisms allow for dynamic information encoding and readout in prefrontal cortex during working memory tasks?
• RQ3 How can LSTM-like units be adapted to reflect known cortical microcircuitry and biophysics?
• RQ4 Can a single computational framework explain working memory, sensory processing, and motor control?
• RQ5 What advantages does ORGaNICs offer over standard LSTMs in both neuroscience and AI applications?

Key findings

• ORGaNICs successfully reproduces complex delay-period activity patterns observed in prefrontal cortex during working memory tasks.
• The model enables analytically tractable characterization of neural dynamics, including transient and oscillatory responses.
• Information can be read out from dynamically varying neural responses at any point in time, despite non-stationary activity.
• The framework supports multiple cognitive functions—working memory, sensory processing, and motor control—within a unified architecture.
• ORGaNICs provides a biophysically grounded alternative to standard LSTMs, with potential advantages in interpretability and computational efficiency.