ویراساینس.
سند علمی

Opening the Black Box: Low-Dimensional Dynamics in High-Dimensional Recurrent Neural Networks

David Sussillo, Omri Barak

چکیده

Recurrent neural networks (RNNs) are useful tools for learning nonlinear relationships between time-varying inputs and outputs with complex temporal dependencies. Recently developed algorithms have been successful at training RNNs to perform a wide variety of tasks, but the resulting networks have been treated as black boxes: their mechanism of operation remains unknown. Here we explore the hypothesis that fixed points, both stable and unstable, and the linearized dynamics around them, can reveal crucial aspects of how RNNs implement their computations. Further, we explore the utility of linearization in areas of phase space that are not true fixed points but merely points of very slow movement. We present a simple optimization technique that is applied to trained RNNs to find the fixed and slow points of their dynamics. Linearization around these slow regions can be used to explore, or reverse-engineer, the behavior of the RNN. We describe the technique, illustrate it using simple examples, and finally showcase it on three high-dimensional RNN examples: a 3-bit flip-flop device, an input-dependent sine wave generator, and a two-point moving average. In all cases, the mechanisms of trained networks could be inferred from the sets of fixed and slow points and the linearized dynamics around them.

اسناد مشابه

  • Dynamical Control by Recurrent Neural Networks through Genetic Algorithm

    سند علمی۲۰۰۳Toru KUMAGAI, Mitsuo WADA, Sadayoshi MIKAMI, Ryoichi HASHIMOTO

    1. Introduction Dynamic control problems often require a set of control rules. For example, an inverted pendulum system requires two different control rules for swinging up and stabilization of a pendulum. Recurrent neural networks (RNNs) are potential candidates for service as c

  • Prediction of Dynamical Systems by Recurrent Neural Networks

    سند علمی۲۰۰۹Peter Trebatický

    Recurrent neural networks in general achieve better results in prediction of time series then feedforward networks. Echo state neural networks seem to be one alternative to them. I have shown on the task of text correction, that they achieve slightly better results compared to al

  • Local Dynamics in Trained Recurrent Neural Networks.

    سند علمی۲۰۱۷Physical review lettersAlexander Rivkind, Omri Barak

    Learning a task induces connectivity changes in neural circuits, thereby changing their dynamics. To elucidate task-related neural dynamics, we study trained recurrent neural networks. We develop a mean field theory for reservoir computing networks trained to have multiple fixed

  • Inferring low-dimensional microstructure representations using convolutional neural networks

    سند علمی۲۰۱۷Physical review. ENicholas Lubbers, Turab Lookman, Kipton Barros

    We apply recent advances in machine learning and computer vision to a central problem in materials informatics: the statistical representation of microstructural images. We use activations in a pretrained convolutional neural network to provide a high-dimensional characterization

  • High Order Recurrent Neural Networks for Acoustic Modelling

    سند علمی۲۰۱۸CoRRChao Zhang, Philip C. Woodland

    Vanishing long-term gradients are a major issue in training standard recurrent neural networks (RNNs), which can be alleviated by long short-term memory (LSTM) models with memory cells. However, the extra parameters associated with the memory cells mean an LSTM layer has four tim

  • Dynamics of a Recurrent Neural Network Acquired through the Learning of a Context-based Attention Task

    سند علمی۲۰۰۲Katsunari Shibata, Masanori Sugisaka

    A context-based attention task is employed in this paper. An Elman-type recurrent neural network is utilized to extract and keep the context information, and only the reinforcement signal that indicates whether the answer is correct or not is given. Through this learning, the fun