Learning to store information over extended time intervals by recurrent backpropagation takes a very long time, mostly because of insufficient, decaying error backflow. We briefly review Hochreiter's (1991) analysis of this problem, then address it by introducing a novel, efficient, gradient based method called long short-term memory (LSTM). Truncating the gradient where this does not do harm, LSTM can learn to bridge minimal time lags in excess of 1000 discrete-time steps by enforcing constant error flow through constant error carousels within special units. Multiplicative gate units learn to open and close access to the constant error flow. LSTM is local in space and time; its computational complexity per time step and weight is O. 1. Our experiments with artificial data involve local, distributed, real-valued, and noisy pattern representations. In comparisons with real-time recurrent learning, back propagation through time, recurrent cascade correlation, Elman nets, and neural sequence chunking, LSTM leads to many more successful runs, and learns much faster. LSTM also solves complex, artificial long-time-lag tasks that have never been solved by previous recurrent network algorithms.

Long short-term memory

https://www2.econ.iastate.edu/tesfatsi/DeepLearningInNeuralNetworksOverview.JSchmidhuber2015.pdf

Deep learning in neural networks

Probability Distributions.- Linear Models for Regression.- Linear Models for Classification.- Neural Networks.- Kernel Methods.- Sparse Kernel Machines.- Graphical Models.- Mixture Models and EM.- Approximate Inference.- Sampling Methods.- Continuous Latent Variables.- Sequential Data.- Combining Models.

Pattern Recognition and Machine Learning

Long short-term memory (LSTM; Hochreiter & Schmidhuber, 1997) can solve numerous tasks not solvable by previous learning algorithms for recurrent neural networks (RNNs). We identify a weakness of LSTM networks processing continual input streams that are not a priori segmented into subsequences with explicitly marked ends at which the network's internal state could be reset. Without resets, the state may grow indefinitely and eventually cause the network to break down. Our remedy is a novel, adaptive "forget gate" that enables an LSTM cell to learn to reset itself at appropriate times, thus releasing internal resources. We review illustrative benchmark problems on which standard LSTM outperforms other RNN algorithms. All algorithms (including LSTM) fail to solve continual versions of these problems. LSTM with forget gates, however, easily solves them, and in an elegant way.

Learning to Forget: Continual Prediction with LSTM

An overview of the self-organizing map algorithm, on which the papers in this issue are based, is presented in this article.

The Self-Organizing Map

It has previously been shown that gradient-descent learning algorithms for recurrent neural networks can perform poorly on tasks that involve long-term dependencies, i.e. those problems for which the desired output depends on inputs presented at times far in the past. We show that the long-term dependencies problem is lessened for a class of architectures called nonlinear autoregressive models with exogenous (NARX) recurrent neural networks, which have powerful representational capabilities. We have previously reported that gradient descent learning can be more effective in NARX networks than in recurrent neural network architectures that have "hidden states" on problems including grammatical inference and nonlinear system identification. Typically, the network converges much faster and generalizes better than other networks. The results in this paper are consistent with this phenomenon. We present some experimental results which show that NARX networks can often retain information for two to three times as long as conventional recurrent neural networks. We show that although NARX networks do not circumvent the problem of long-term dependencies, they can greatly improve performance on long-term dependency problems. We also describe in detail some of the assumptions regarding what it means to latch information robustly and suggest possible ways to loosen these assumptions.

Learning long-term dependencies in NARX recurrent neural networks

Reservoir computing (RC) refers to a new class of state-space models with a fixed state transition structure (the reservoir) and an adaptable readout form the state space. The reservoir is supposed to be sufficiently complex so as to capture a large number of features of the input stream that can be exploited by the reservoir-to-output readout mapping. The field of RC has been growing rapidly with many successful applications. However, RC has been criticized for not being principled enough. Reservoir construction is largely driven by a series of randomized model-building stages, with both researchers and practitioners having to rely on a series of trials and errors. To initialize a systematic study of the field, we concentrate on one of the most popular classes of RC methods, namely echo state network, and ask: What is the minimal complexity of reservoir construction for obtaining competitive models and what is the memory capacity (MC) of such simplified reservoirs? On a number of widely used time series benchmarks of different origin and characteristics, as well as by conducting a theoretical analysis we show that a simple deterministically constructed cycle reservoir is comparable to the standard echo state network methodology. The (short-term) of linear cyclic reservoirs can be made arbitrarily close to the proved optimal value.

Minimum Complexity Echo State Network

Along with the great success of deep neural networks, there is also growing concern about their black-box nature. The interpretability issue affects people's trust on deep learning systems. It is also related to many ethical problems, e.g., algorithmic discrimination. Moreover, interpretability is a desired property for deep networks to become powerful tools in other research fields, e.g., drug discovery and genomics. In this survey, we conduct a comprehensive review of the neural network interpretability research. We first clarify the definition of interpretability as it has been used in many different contexts. Then we elaborate on the importance of interpretability and propose a novel taxonomy organized along three dimensions: type of engagement (passive vs. active interpretation approaches), the type of explanation, and the focus (from local to global interpretability). This taxonomy provides a meaningful 3D view of distribution of papers from the relevant literature as two of the dimensions are not simply categorical but allow ordinal subcategories. Finally, we summarize the existing interpretability evaluation methods and suggest possible research directions inspired by our new taxonomy.

A Survey on Neural Network Interpretability

In this paper, we elaborate upon the claim that clustering in the recurrent layer of recurrent neural networks (RNNs) reflects meaningful information processing states even prior to training. By concentrating on activation clusters in RNNs, while not throwing away the continuous state space network dynamics, we extract predictive models that we call neural prediction machines (NPMs). When RNNs with sigmoid activation functions are initialized with small weights (a common technique in the RNN community), the clusters of recurrent activations emerging prior to training are indeed meaningful and correspond to Markov prediction contexts. In this case, the extracted NPMs correspond to a class of Markov models, called variable memory length Markov models (VLMMs). In order to appreciate how much information has really been induced during the training, the RNN performance should always be compared with that of VLMMs and NPMs extracted before training as the "null" base models. Our arguments are supported by experiments on a chaotic symbolic sequence and a context-free language with a deep recursive structure.

Markovian architectural bias of recurrent neural networks

In this paper, a sparse learning algorithm, probabilistic classification vector machines (PCVMs), is proposed. We analyze relevance vector machines (RVMs) for classification problems and observe that adopting the same prior for different classes may lead to unstable solutions. In order to tackle this problem, a signed and truncated Gaussian prior is adopted over every weight in PCVMs, where the sign of prior is determined by the class label, i.e., +1 or -1. The truncated Gaussian prior not only restricts the sign of weights but also leads to a sparse estimation of weight vectors, and thus controls the complexity of the model. In PCVMs, the kernel parameters can be optimized simultaneously within the training algorithm. The performance of PCVMs is extensively evaluated on four synthetic data sets and 13 benchmark data sets using three performance metrics, error rate (ERR), area under the curve of receiver operating characteristic (AUC), and root mean squared error (RMSE). We compare PCVMs with soft-margin support vector machines (SVMSoft), hard-margin support vector machines (SVMHard), SVM with the kernel parameters optimized by PCVMs (SVMPCVM), relevance vector machines (RVMs), and some other baseline classifiers. Through five replications of twofold cross-validation F test, i.e., 5 times 2 cross-validation F test, over single data sets and Friedman test with the corresponding post-hoc test to compare these algorithms over multiple data sets, we notice that PCVMs outperform other algorithms, including SVMSoft, SVMHard, RVM, and SVMPCVM, on most of the data sets under the three metrics, especially under AUC. Our results also reveal that the performance of SVMPCVM is slightly better than SVMSoft, implying that the parameter optimization algorithm in PCVMs is better than cross validation in terms of performance and computational complexity. In this paper, we also discuss the superiority of PCVMs' formulation using maximum a posteriori (MAP) analysis and margin analysis, which explain the empirical success of PCVMs.

/pdf/probabilistic-classification-vector-machines-3b2ev23rhk.pdf

Peter Tino

Papers

Learning long-term dependencies in NARX recurrent neural networks

Minimum Complexity Echo State Network

A Survey on Neural Network Interpretability

Markovian architectural bias of recurrent neural networks

Probabilistic Classification Vector Machines