Memory of recurrent networks: Do we compute it right?

Giovanni Ballarin,Lyudmila Grigoryeva,Juan-Pablo Ortega

2024-09-10

Abstract:Numerical evaluations of the memory capacity (MC) of recurrent neural networks reported in the literature often contradict well-established theoretical bounds. In this paper, we study the case of linear echo state networks, for which the total memory capacity has been proven to be equal to the rank of the corresponding Kalman controllability matrix. We shed light on various reasons for the inaccurate numerical estimations of the memory, and we show that these issues, often overlooked in the recent literature, are of an exclusively numerical nature. More explicitly, we prove that when the Krylov structure of the linear MC is ignored, a gap between the theoretical MC and its empirical counterpart is introduced. As a solution, we develop robust numerical approaches by exploiting a result of MC neutrality with respect to the input mask matrix. Simulations show that the memory curves that are recovered using the proposed methods fully agree with the theory.

Machine Learning

What problem does this paper attempt to address?

The paper attempts to address the issue of discrepancies between theoretical and practical evaluations of memory capacity (MC) in recurrent neural networks (RNNs). Specifically, the numerical evaluations of the memory capacity of linear echo state networks (ESNs) reported in the literature often contradict established theoretical bounds. Using linear echo state networks as an example, this paper investigates the reasons for such inconsistencies and demonstrates that these issues are purely numerical in nature. The authors point out that when the Krylov structure of linear memory capacity is ignored, a gap arises between the theoretical memory capacity and its empirical estimates. To address this issue, the authors developed robust numerical methods that utilize input mask matrices to neutralize the results of memory capacity. Through simulation validation, the memory curves recovered using these new methods fully align with theoretical predictions.

In summary, the main objectives of the paper are:
1. **Reveal the problem**: Explain why existing numerical methods show discrepancies between theoretical and practical evaluations of the memory capacity of linear RNNs.
2. **Propose a solution**: Develop new numerical methods to ensure that the evaluation results of memory capacity are consistent with theory.
3. **Validate the methods**: Experimentally verify the effectiveness of the new methods, demonstrating their robustness and accuracy in practical applications.