Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Information encoding and decoding in in-vitro neural networks on micro electrode arrays through stimulation timing

Trym A. E. Lindell, Ola H. Ramstad, Ionna Sandvig, Axel Sandvig, Stefano Nichele

TL;DR

This work investigates how to encode and decode information in in-vitro neural networks on microelectrode arrays using stimulation timing as the encoding channel. It systematically characterizes the upper and lower bounds and acuity of timing-based encoding, and identifies optimal readout parameter regimes via a grid search over epoch length, time-bin size, and epoch offset, using logistic regression to test linear separability of different probe delays. The study shows that meaningful separability exists for stimulation delays from near 0 ms up to several seconds, with best acuity in the 36-436 ms range, and reveals significant day-to-day variability likely due to metabolic state and media changes. These findings provide practical guidelines and highlight challenges for deploying biological reservoirs as low-energy computation substrates, including stability and phase dependence.

Abstract

A primary challenge in utilizing in-vitro biological neural networks for computations is finding good encoding and decoding schemes for inputting and decoding data to and from the networks. Furthermore, identifying the optimal parameter settings for a given combination of encoding and decoding schemes adds additional complexity to this challenge. In this study we explore stimulation timing as an encoding method, i.e. we encode information as the delay between stimulation pulses and identify the bounds and acuity of stimulation timings which produce linearly separable spike responses. We also examine the optimal readout parameters for a linear decoder in the form of epoch length, time bin size and epoch offset. Our results suggest that stimulation timings between 36 and 436ms may be optimal for encoding and that different combinations of readout parameters may be optimal at different parts of the evoked spike response.

Information encoding and decoding in in-vitro neural networks on micro electrode arrays through stimulation timing

TL;DR

This work investigates how to encode and decode information in in-vitro neural networks on microelectrode arrays using stimulation timing as the encoding channel. It systematically characterizes the upper and lower bounds and acuity of timing-based encoding, and identifies optimal readout parameter regimes via a grid search over epoch length, time-bin size, and epoch offset, using logistic regression to test linear separability of different probe delays. The study shows that meaningful separability exists for stimulation delays from near 0 ms up to several seconds, with best acuity in the 36-436 ms range, and reveals significant day-to-day variability likely due to metabolic state and media changes. These findings provide practical guidelines and highlight challenges for deploying biological reservoirs as low-energy computation substrates, including stability and phase dependence.

Abstract

A primary challenge in utilizing in-vitro biological neural networks for computations is finding good encoding and decoding schemes for inputting and decoding data to and from the networks. Furthermore, identifying the optimal parameter settings for a given combination of encoding and decoding schemes adds additional complexity to this challenge. In this study we explore stimulation timing as an encoding method, i.e. we encode information as the delay between stimulation pulses and identify the bounds and acuity of stimulation timings which produce linearly separable spike responses. We also examine the optimal readout parameters for a linear decoder in the form of epoch length, time bin size and epoch offset. Our results suggest that stimulation timings between 36 and 436ms may be optimal for encoding and that different combinations of readout parameters may be optimal at different parts of the evoked spike response.
Paper Structure (39 sections, 23 figures, 7 tables)

This paper contains 39 sections, 23 figures, 7 tables.

Table of Contents

  1. Introduction
  2. Background
  3. Reservoir Computing
  4. In-Vitro Neural Networks on Micro Electrode Arrays
  5. Reservoir Computing With in-Vitro Neural Networks
  6. Methods
  7. System Methods
  8. Neural Network Cultures
  9. MEA Recording System
  10. Spike and Network Burst Detection
  11. Machine learning algorithm
  12. Experimental Methods
  13. Experiment 1
  14. Experiment 2
  15. Analysis and Results
  16. ...and 24 more sections

Figures (23)

  • Figure 1: Components of minimal reservoir computing system. Input is given through the encoder into the reservoir, in this case a recurrent neural network. The input propagates through the reservoir network and the decoder reads out the reservoir states and translates them to the desired output after training.
  • Figure 2: Schematic figure of in-vitro neural network on Micro Electrode Array (MEA).
  • Figure 3: Fluorescent microscopy of an in-vitro neural network on an MEA interface after immunostaining : neurons (Neurofilament heavy; red), astrocytes (GFAP;green); cell nuclei (Hoechst; blue).
  • Figure 4: Schematic figure of in-vitro neural network on Micro Electrode Array (MEA) as a reservoir computing system
  • Figure 5: Components of the stimulation pulse. All stimulation pulses were identical and consisted of a bi-polar stimulation lasting 0.5 ms directly followed by a 36 ms artefact removal stimulation. Recorded time of stimulation was set to the end of the bi-polar pulse.
  • ...and 18 more figures