Paper Reading AI Learner

Network mechanisms of working memory: the role of neuronal nonlinearities

2021-12-09 18:32:44
Alex Suarez-Perez, Omri Harish, David Hansel

Abstract

The oculomotor delayed-response (ODR) task is a common experimental paradigm of working memory (WM) study, in which a monkey must fixate its gaze on the center of a screen and, following a brief cue that flashes on the screen, keep fixating for several more seconds before shifting its gaze to the location where the cue flashed. Consequently, in the delay period between the cue and the response the monkey must maintain a memory of cue location. Electrophysiological recordings from the prefrontal area of the cortex (PFC) revealed neurons that display selective persistent activity: their firing rate change induced by the cue persists through delay period, but only in response to a confined range of cue locations. This suggests that the representation of the cue is maintained in the network by a change in network activity profile. In this work, we study a network of rate-model neurons that is capable of preserving information about a past input, owing to structured connectivity and nonlinearities in the neuronal transfer function (TFs). Particularly, we focus on the acceleration of the TF close to firing threshold and the concavity around TF saturation. Any memory mechanism which exploits TF saturation means that some neurons must fire close to their saturation rates; with our model, however, we show that a certain relation between the excitatory and inhibitory neurons' TFs can cause an effective saturation in the network without forcing the neurons into the saturating parts of their TFs. In addition, this mechanism enables the erasure of memory at the end of the delay by a global excitatory signal. Finally, we demonstrate the mechanism in a model network of spiking neurons which describes with more detail the oscillatory dynamics in the state transition due to the interaction of membrane and synaptic time constants which is neglected in the rate model.

Abstract (translated)

URL

https://arxiv.org/abs/2112.05091

PDF

https://arxiv.org/pdf/2112.05091.pdf


Tags
3D Action Action_Localization Action_Recognition Activity Adversarial Agent Attention Autonomous Bert Boundary_Detection Caption Chat Classification CNN Compressive_Sensing Contour Contrastive_Learning Deep_Learning Denoising Detection Dialog Diffusion Drone Dynamic_Memory_Network Edge_Detection Embedding Embodied Emotion Enhancement Face Face_Detection Face_Recognition Facial_Landmark Few-Shot Gait_Recognition GAN Gaze_Estimation Gesture Gradient_Descent Handwriting Human_Parsing Image_Caption Image_Classification Image_Compression Image_Enhancement Image_Generation Image_Matting Image_Retrieval Inference Inpainting Intelligent_Chip Knowledge Knowledge_Graph Language_Model Matching Medical Memory_Networks Multi_Modal Multi_Task NAS NMT Object_Detection Object_Tracking OCR Ontology Optical_Character Optical_Flow Optimization Person_Re-identification Point_Cloud Portrait_Generation Pose Pose_Estimation Prediction QA Quantitative Quantitative_Finance Quantization Re-identification Recognition Recommendation Reconstruction Regularization Reinforcement_Learning Relation Relation_Extraction Represenation Represenation_Learning Restoration Review RNN Salient Scene_Classification Scene_Generation Scene_Parsing Scene_Text Segmentation Self-Supervised Semantic_Instance_Segmentation Semantic_Segmentation Semi_Global Semi_Supervised Sence_graph Sentiment Sentiment_Classification Sketch SLAM Sparse Speech Speech_Recognition Style_Transfer Summarization Super_Resolution Surveillance Survey Text_Classification Text_Generation Tracking Transfer_Learning Transformer Unsupervised Video_Caption Video_Classification Video_Indexing Video_Prediction Video_Retrieval Visual_Relation VQA Weakly_Supervised Zero-Shot