PonderNet: Learning to Ponder (Machine Learning Research Paper Explained)

#pondernet #deepmind #machinelearning Humans don't spend the same amount of mental effort on all problems equally. Instead, we respond quickly to easy tasks, and we take our time to deliberate hard tasks. DeepMind's PonderNet attempts to achieve the same by dynamically deciding how many computation steps to allocate to any single input sample. This is done via a recurrent architecture and a trainable function that computes a halting probability. The resulting model performs well in dynamic computation tasks and is surprisingly robust to different hyperparameter settings. OUTLINE: 0:00 - Intro & Overview 2:30 - Problem Statement 8:00 - Probabilistic formulation of dynamic halting 14:40 - Training via unrolling 22:30 - Loss function and regularization of the halting distribution 27:35 - Experimental Results 37:10 - Sensitivity to hyperparameter choice 41:15 - Discussion, Conclusion, Broader Impact Paper: https://arxiv.org/abs/2107.05407 Abstract: In standard neural networks the amount of computation used grows with the size of the inputs, but not with the complexity of the problem being learnt. To overcome this limitation we introduce PonderNet, a new algorithm that learns to adapt the amount of computation based on the complexity of the problem at hand. PonderNet learns end-to-end the number of computational steps to achieve an effective compromise between training prediction accuracy, computational cost and generalization. On a complex synthetic problem, PonderNet dramatically improves performance over previous adaptive computation methods and additionally succeeds at extrapolation tests where traditional neural networks fail. Also, our method matched the current state of the art results on a real world question and answering dataset, but using less compute. Finally, PonderNet reached state of the art results on a complex task designed to test the reasoning capabilities of neural networks.1 Authors: Andrea Banino, Jan Balaguer, Charles Blundell Links: TabNine