dynamic-sampling-non-stationary-spontaneous-activity

name	dynamic-sampling-non-stationary-spontaneous-activity
title	Dynamic sampling of non-stationary spontaneous activity in dissociated neuronal networks
arxiv_id	2607.24269
date	2026-07-27T00:00:00.000Z
authors	["Kazushi Takehana","Dai Akita","Hirokazu Takahashi"]
domain	neuroscience
description	Adaptive electrode-selection method using discounted Poisson-Gamma model with Thompson sampling for tracking non-stationary spontaneous activity during long-term HD-MEA recordings under fixed channel budget constraints.
tags	["hd-mea","adaptive electrode selection","thompson sampling","non-stationary neural activity","bayesian optimization"]

Dynamic Sampling of Non-Stationary Spontaneous Activity in Dissociated Neuronal Networks

Overview

This methodology develops and evaluates an adaptive electrode-selection method for tracking non-stationary spontaneous activity during long-term high-density microelectrode array (HD-MEA) recordings under a fixed channel budget constraint.

Core Methodology

Problem Formulation

• Objective: Track evolving neural activity patterns over extended recording periods with limited readout channels
• Challenge: Neural activity is non-stationary, with electrode activity levels changing substantially over time (47.8% turnover at 34 hours)
• Constraint: Fixed channel budget (e.g., 100 electrodes from 529 candidates)

Technical Approach

• Sequential subset-selection: Formulate electrode allocation as a sequential decision problem
• Discounted Poisson-Gamma model: Bayesian framework for modeling spike count dynamics with temporal discounting
• Thompson sampling: Uncertainty-aware exploration strategy for adaptive electrode selection
• Real-time updates: Continuously update electrode-specific activity estimates from observed spike counts

Implementation Details

• Offline evaluation: Tested on nine 34-hour HD-MEA recordings with 100/529 electrode selection
• Online validation: Demonstrated in real-time recording with 1,024 routed electrodes
• Performance metric: Fraction of spikes captured compared to oracle selector

Key Results

Performance Gains

• 17.2 percentage point improvement over static electrode selection at final time point
• Optimal spike capture: Bayesian method captured the largest fraction of available spikes among tested strategies
• Dynamic adaptation: Successfully tracked substantial changes in active electrode sets over time

Practical Applications

• Synchronized burst detection: Captured first synchronized burst in online recording
• Trajectory analysis: Supported center-of-activity trajectory analysis
• Long-term monitoring: Enables efficient recording over extended periods despite non-stationarity

Significance and Applications

Scientific Impact