Neural timescales from a computational perspective

Timescales of neural activity are diverse across and within brain areas, and experimental observations suggest that neural timescales reflect information in dynamic environments. However, these observations do not specify how neural timescales are shaped, nor whether particular timescales are necessary for neural computations and brain function. Here, we take a complementary perspective and synthesize three directions where computational methods can distill the broad set of empirical observations into quantitative and testable theories: We review (i) how data analysis methods allow us to capture different timescales of neural dynamics across different recording modalities, (ii) how computational models provide a mechanistic explanation for the emergence of diverse timescales, and (iii) how task-optimized models in machine learning uncover the functional relevance of neural timescales. This integrative computational approach, combined with empirical findings, would provide a more holistic understanding of how neural timescales capture the relationship between brain structure, dynamics, and behavior.

Emergent mechanisms for long timescales depend on training curriculum and affect performance in memory tasks

Recurrent neural networks (RNNs) in the brain and in silico excel at solving tasks with intricate temporal dependencies. Long timescales required for solving such tasks can arise from properties of individual neurons (single-neuron timescale, $\tau$, e.g., membrane time constant in biological neurons) or recurrent interactions among them (network-mediated timescale). However, the contribution of each mechanism for optimally solving memory-dependent tasks remains poorly understood. Here, we train RNNs to solve $N$-parity and $N$-delayed match-to-sample tasks with increasing memory requirements controlled by $N$ by simultaneously optimizing recurrent weights and $\tau$s. We find that for both tasks RNNs develop longer timescales with increasing $N$, but depending on the learning objective, they use different mechanisms. Two distinct curricula define learning objectives: sequential learning of a single-$N$ (single-head) or simultaneous learning of multiple $N$s (multi-head). Single-head networks increase their $\tau$ with $N$ and are able to solve tasks for large $N$, but they suffer from catastrophic forgetting. However, multi-head networks, which are explicitly required to hold multiple concurrent memories, keep $\tau$ constant and develop longer timescales through recurrent connectivity. Moreover, we show that the multi-head curriculum increases training speed and network stability to ablations and perturbations, and allows RNNs to generalize better to tasks beyond their training regime. This curriculum also significantly improves training GRUs and LSTMs for large-$N$ tasks. Our results suggest that adapting timescales to task requirements via recurrent interactions allows learning more complex objectives and improves the RNN's performance.