AUTHOR=Daram Anurag , Yanguas-Gil Angel , Kudithipudi Dhireesha 

TITLE=Exploring Neuromodulation for Dynamic Learning

JOURNAL=Frontiers in Neuroscience

VOLUME=Volume 14 - 2020

YEAR=2020

URL=https://www.frontiersin.org/journals/neuroscience/articles/10.3389/fnins.2020.00928

DOI=10.3389/fnins.2020.00928

ISSN=1662-453X

ABSTRACT=A continual learning system requires the ability to dynamically adopt and generalize to new tasks while being exposed to few samples. In the central nervous system across species, it is observed that continual and dynamic behavior demonstrated by brain is an active result of a mechanism known as neuromodulation Therefore, in this work, neuromodulatory plasticity is embedded with dynamic learning architectures as a step towards realizing power and area efficient few shot learning systems. This is enabled by having an inbuilt modulatory unit that regulates learning based on the context and internal state of the system. This renders the system with the ability to self modify its weights. In one of the proposed architectures, ModNet, a modulatory layer is introduced in a random projection framework. This layer modulates the weights of the output layer neurons in tandem with hebbian learning. ModNet architecture's learning capabilities are enhanced by integrating inbuilt attention along with compartmentalization based mechanisms. Moreover, to explore modulatory mechanisms in conjunction with backpropagation in deeper networks, a modulatory trace learning rule is introduced. The proposed learning rule, uses a time dependent trace to automatically modify the synaptic connections as a function of ongoing states and activations. The trace itself is updated via simple plasticity rules thus reducing the demand on resources.The proposed modulatory learning architecture and learning rules demonstrate the ability to learn from few samples, train quickly, and perform one shot image classification in a computationally efficient manner. The simple ModNet and the compartmentalized ModNet architecture perform well on image classification tasks, while training for just 2 epochs. The network with modulatory trace achieves an average accuracy of 98.8%±1.16 on the omniglot dataset for five-way one-shot image classification task while requiring ~20x fewer trainable parameters in comparison to other state of the art models. In general, incorporating neuromodulation in deep neural networks shows promise for energy and resource efficient lifelong learning systems.