Randomized tensor network reservoir computing: validity and learnability phase transitions
Reservoir computing (RC) systems, traditionally based on echo state networks (ESN) or liquid state machines, have shown significant potential in dynamic temporal data modeling, such as weather forecasting and astronomical predictions. However, these frameworks are known not being applicable to quant...
Saved in:
| Main Authors: | , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
IOP Publishing
2025-01-01
|
| Series: | Machine Learning: Science and Technology |
| Subjects: | |
| Online Access: | https://doi.org/10.1088/2632-2153/aded56 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Reservoir computing (RC) systems, traditionally based on echo state networks (ESN) or liquid state machines, have shown significant potential in dynamic temporal data modeling, such as weather forecasting and astronomical predictions. However, these frameworks are known not being applicable to quantum dynamics-based RC. Tensor networks (TNs), with their efficient representation of high-dimensional quantum information and entanglement, are powerful tools for modeling correlated quantum dynamics. Introducing randomized effects into TNs, akin to randomization of recurrent connections in traditional RC, is expected to generate diverse quantum correlation patterns and provide robust, generalizable quantum-inspired dynamic models through randomization. In this work, we propose a novel randomized TN-based RC scheme, experimentally demonstrating its validity. A theoretical model selection criterion is constructed to find the optimal TNRC hyperparameters. Critical phenomena along with the phase transitions of learnability near the edge of chaos in TN RC are clearly identified. The distribution-independent universality in phase transitions observed in TN RC is captured by the newly developed learning theory and a self-consistent mean-field theory of the spin-glass type. The performance advantage of TNRC over ESN is demonstrated in several forecasting experiments. Our findings lay the groundwork for future explorations into randomized TN quantum machine learning, phase transitions in quantum RC, and the manipulation of critical phenomena in complex systems. |
|---|---|
| ISSN: | 2632-2153 |