Neural Stored-program Memory
This work addresses the challenge of creating more flexible and adaptive neural networks for tasks requiring program switching and context adaptation, representing a novel advancement rather than an incremental improvement.
The paper tackled the problem of enabling neural networks to switch programs and adapt to variable contexts by introducing a new memory to store weights for the neural controller, analogous to stored-program memory in computers. The result is a model that excels in classical algorithmic problems and shows potential for compositional, continual, few-shot learning, and question-answering tasks.
Neural networks powered with external memory simulate computer behaviors. These models, which use the memory to store data for a neural controller, can learn algorithms and other complex tasks. In this paper, we introduce a new memory to store weights for the controller, analogous to the stored-program memory in modern computer architectures. The proposed model, dubbed Neural Stored-program Memory, augments current memory-augmented neural networks, creating differentiable machines that can switch programs through time, adapt to variable contexts and thus resemble the Universal Turing Machine. A wide range of experiments demonstrate that the resulting machines not only excel in classical algorithmic problems, but also have potential for compositional, continual, few-shot learning and question-answering tasks.