• Corpus ID: 249097794

Efficient and robust multi-task learning in the brain with modular latent primitives

@inproceedings{Marton2021EfficientAR,
  title={Efficient and robust multi-task learning in the brain with modular latent primitives},
  author={Christian David M'arton and L'eo Gagnon and Guillaume Lajoie and Kanaka Rajan},
  year={2021}
}

• Christian David M'arton, L'eo Gagnon, +1 author Kanaka Rajan
• Published 28 May 2021
• Computer Science

Biological agents do not have infinite resources to learn new things. For this reason, a central aspect of human learning is the ability to recycle previously acquired knowledge in a way that allows for faster, less resource-intensive acquisition of new skills. In spite of that, how neural networks in the brain leverage existing knowledge to learn new computations is not well understood. In this work, we study this question in artificial recurrent neural networks (RNNs) trained on a corpus of commonly used neuroscience tasks. Combining brain-inspired inductive biases we call functional and structural, we propose a system that learns new tasks by building on top of pre-trained latent dynamics organised into separate recurrent modules. These modules, acting as prior knowledge acquired previously through evolution or development, are pre-trained on the statistics of the full corpus of tasks so as to be independent and maximally informative. The resulting model, we call a Modular Latent Primitives (MoLaP) network, allows for learning multiple tasks while keeping parameter counts, and updates, low. We also show that the skills acquired with our approach are more robust to a broad range of perturbations compared to those acquired with other multi-task learning strategies, and that generalisation to new tasks is facilitated. This work offers a new perspective on achieving efficient multi-task learning in the brain, illustrating the benefits of leveraging pre-trained latent dynamical primitives. models of biologically plausible multi-task learning. Finally, we show that generalization to new tasks that share basic structure with the corpus is also facilitated. This draws a path towards cheaper multitask solutions, and offers new hypotheses for how continual learning might be implemented in the biological brain. which are used to pre-train separate network modules. B. During task training, recurrent weights remain frozen and only input and readout weights are adjusted. Task-specific responses can be obtained via separate readout weights. C. Parameter overview for the various multi-task training methods. Parameters trained (or updated) and total network size is compared across the three methods. MoLaP makes ∼ 2 orders of magnitude fewer parameter updates than DNI, at a relatively moderate increase of total parameter count.