Dice Question Streamline Icon: https://streamlinehq.com

Applicability of LPU drift-robustness theorems to low-rank leaky current RNNs

Determine whether the drift-robustness results for linearly encoded latent processing units extend to low-rank leaky current recurrent neural network architectures that enforce linear maps from latent variables to neural activities (r = M κ), given their constrained tuning relationships.

Information Square Streamline Icon: https://streamlinehq.com

Background

The authors compare their framework to low-rank leaky current RNNs, which impose linear relationships between latent variables and neural activities, yielding flat manifolds and linear tuning. They note the absence of results on representational drift in these models and question whether their own theorems apply under such constraints.

Establishing or refuting the applicability of drift-robustness theorems to low-rank leaky current RNNs would clarify the generality of latent computation framework predictions across modeling paradigms.

References

No known result on representational drift, and unclear whether our theorems cover these since tuning relationships are overly constrained.

— Latent computing by biological neural networks: A dynamical systems framework (2502.14337 - Dinc et al., 20 Feb 2025) in Table 1 (tabs1), “An overview of existing research and relevance to this work”