Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Efficient Optimisation of Physical Reservoir Computers using only a Delayed Input

Enrico Picco, Lina Jaurigue, Kathy Lüdge, Serge Massar

TL;DR

The paper tackles the challenge of hyperparameter tuning in reservoir computing by validating a delayed-input optimization method on an optoelectronic RC. By injecting a delayed version of the input signal and tuning only two parameters, $\beta_{2}$ and the delay $d$, the approach identifies an effective operating region without exhaustive search. Experimental results across NARMA10, Mackey-Glass, Spoken Digit, and Speaker Recognition tasks show consistent performance gains over standard hyperparameter tuning, including under suboptimal reservoir settings. This low-complexity, hardware-friendly strategy promises practical benefits for physical RC implementations where full hyperparameter optimization is impractical.

Abstract

We present an experimental validation of a recently proposed optimization technique for reservoir computing, using an optoelectronic setup. Reservoir computing is a robust framework for signal processing applications, and the development of efficient optimization approaches remains a key challenge. The technique we address leverages solely a delayed version of the input signal to identify the optimal operational region of the reservoir, simplifying the traditionally time-consuming task of hyperparameter tuning. We verify the effectiveness of this approach on different benchmark tasks and reservoir operating conditions.

Efficient Optimisation of Physical Reservoir Computers using only a Delayed Input

TL;DR

The paper tackles the challenge of hyperparameter tuning in reservoir computing by validating a delayed-input optimization method on an optoelectronic RC. By injecting a delayed version of the input signal and tuning only two parameters, and the delay , the approach identifies an effective operating region without exhaustive search. Experimental results across NARMA10, Mackey-Glass, Spoken Digit, and Speaker Recognition tasks show consistent performance gains over standard hyperparameter tuning, including under suboptimal reservoir settings. This low-complexity, hardware-friendly strategy promises practical benefits for physical RC implementations where full hyperparameter optimization is impractical.

Abstract

We present an experimental validation of a recently proposed optimization technique for reservoir computing, using an optoelectronic setup. Reservoir computing is a robust framework for signal processing applications, and the development of efficient optimization approaches remains a key challenge. The technique we address leverages solely a delayed version of the input signal to identify the optimal operational region of the reservoir, simplifying the traditionally time-consuming task of hyperparameter tuning. We verify the effectiveness of this approach on different benchmark tasks and reservoir operating conditions.
Paper Structure (13 sections, 7 equations, 6 figures, 1 table)

This paper contains 13 sections, 7 equations, 6 figures, 1 table.

Table of Contents

  1. Introduction
  2. Time-delay Reservoir Computing
  3. Reservoir Computing with Delayed Input
  4. Experimental Setup
  5. Tasks
  6. NARMA10
  7. Mackey-Glass system
  8. Spoken Digit Recognition
  9. Speaker Recognition
  10. Results
  11. Effectiveness of optimisation with delayed input
  12. Optimisation with delayed input vs standard hyperparameters optimisation
  13. Conclusions

Figures (6)

  • Figure 1: Optoelectronic reservoir computer. In orange the optic part and in blue the electronic part. SLD: superluminescent diode. MZ: Mach-Zender intensity modulator. OA: Optical Attenuator. PD: Photo-Detector. FPGA: Field Programmable Gate Array. Amp: amplifier.
  • Figure 2: Experimental results on the NARMA10 task as a function of $\beta_{2}$ and the delay $d$ of equation (\ref{['eq_dly_inp']}), using different optical feedback attenuation: 2 dB (a) and 15 dB (b).
  • Figure 3: Experimental results on the prediction of Mackey-Glass system 10 steps in the future, as a function of $\beta_{2}$ and the delay $d$ of equation (\ref{['eq_dly_inp']}), using different optical feedback attenuation: 2 dB (a) and 15 dB (b).
  • Figure 4: Experimental results on Spoken Digit Recognition with noise, as a function of $\beta_{2}$ and the delay $d$ of equation (\ref{['eq_dly_inp']}).
  • Figure 5: Experimental results on Speaker Recognition, as a function of $\beta_{2}$ and the delay $d$ of equation (\ref{['eq_dly_inp']}).
  • ...and 1 more figures