Vince D. Calhoun The Mind Research Network Albuquerque, NM 87106 vcalhoun@mrn.org Deep learning methods have recently made notable advances in the tasks of classification and representation learning. These tasks are important for brain imaging and neuroscience discovery, making the methods attractive for porting to a neuroimager's toolbox. Success of these methods is, in part, explained by the flexibility of deep learning models. However, this flexibility makes the process of porting to new areas a difficult parameter optimization problem. In this work we demonstrate our results (and feasible parameter ranges) in application of deep learning methods to structural and functional brain imaging data. We also describe a novel constraint-based approach to visualizing high dimensional data. We use it to analyze the effect of parameter choices on data transformations. Our results show that deep learning methods are able to learn physiologically important representations and detect latent relations in neuroimaging data.

Motivated by an abstract notion of low-level edge detector filters, we propose a simple method of unsupervised feature construction based on pairwise statistics of features. In the first step, we construct neighborhoods of features by regrouping features that correlate. Then we use these subsets as filters to produce new neighborhood features. Next, we connect neighborhood features that correlate, and construct edge features by subtracting the correlated neighborhood features of each other. To validate the usefulness of the constructed features, we ran AdaBoost.MH on four multi-class classification problems. Our most significant result is a test error of 0.94% on MNIST with an algorithm which is essentially free of any image-specific priors. On CIFAR-10 our method is suboptimal compared to today's best deep learning techniques, nevertheless, we show that the proposed method outperforms not only boosting on the raw pixels, but also boosting on Haar filters.

We present a comprehensive study on the use of autoencoders for modelling text data, in which (differently from previous studies) we focus our attention on the following issues: i) we explore the suitability of two different models bDA and rsDA for constructing deep autoencoders for text data at the sentence level; ii) we propose and evaluate two novel metrics for better assessing the text-reconstruction capabilities of autoencoders; and iii) we propose an automatic method to find the critical bottleneck dimensionality for text language representations (below which structural information is lost).

We consider the task of learning to extract motion from videos. To this end, we show that the detection of spatial transformations can be viewed as the detection of synchrony between the image sequence and a sequence of features undergoing the motion we wish to detect. We show that learning about synchrony is possible using very fast, local learning rules, by introducing multiplicative "gating" interactions between hidden units across frames. This makes it possible to achieve competitive performance in a wide variety of motion estimation tasks, using a small fraction of the time required to learn features, and to outperform hand-crafted spatio-temporal features by a large margin. We also show how learning about synchrony can be viewed as performing greedy parameter estimation in the well-known motion energy model.

Bi-linear feature learning models, like the gated autoencoder, were proposed as a way to model relationships between frames in a video. By minimizing reconstruction error of one frame, given the previous frame, these models learn "mapping units" that encode the transformations inherent in a sequence, and thereby learn to encode motion. In this work we extend bi-linear models by introducing "higher-order mapping units" that allow us to encode transformations between frames and transformations between transformations. We show that this makes it possible to encode temporal structure that is more complex and longer-range than the structure captured within standard bi-linear models. We also show that a natural way to train the model is by replacing the commonly used reconstruction objective with a prediction objective which forces the model to correctly predict the evolution of the input multiple steps into the future. Learning can be achieved by back-propagating the multi-step prediction through time. We test the model on various temporal prediction tasks, and show that higher-order mappings and predictive training both yield a significant improvement over bi-linear models in terms of prediction accuracy.

Long Short-Term Memory (LSTM) is a recurrent neural network (RNN) architecture that has been designed to address the vanishing and exploding gradient problems of conventional RNNs. Unlike feedforward neural networks, RNNs have cyclic connections making them powerful for modeling sequences. They have been successfully used for sequence labeling and sequence prediction tasks, such as handwriting recognition, language modeling, phonetic labeling of acoustic frames. However, in contrast to the deep neural networks, the use of RNNs in speech recognition has been limited to phone recognition in small scale tasks. In this paper, we present novel LSTM based RNN architectures which make more effective use of model parameters to train acoustic models for large vocabulary speech recognition. We train and compare LSTM, RNN and DNN models at various numbers of parameters and configurations. We show that LSTM models converge quickly and give state of the art speech recognition performance for relatively small sized models.

In this paper, we study ordered representations of data in which different dimensions have different degrees of importance. To learn these representations we introduce nested dropout, a procedure for stochastically removing coherent nested sets of hidden units in a neural network. We first present a sequence of theoretical results in the simple case of a semi-linear autoencoder. We rigorously show that the application of nested dropout enforces identifiability of the units, which leads to an exact equivalence with PCA. We then extend the algorithm to deep models and demonstrate the relevance of ordered representations to a number of applications. Specifically, we use the ordered property of the learned codes to construct hash-based data structures that permit very fast retrieval, achieving retrieval in time logarithmic in the database size and independent of the dimensionality of the representation. This allows codes that are hundreds of times longer than currently feasible for retrieval. We therefore avoid the diminished quality associated with short codes, while still performing retrieval that is competitive in speed with existing methods. We also show that ordered representations are a promising way to learn adaptive compression for efficient online data reconstruction.

A deep belief network (DBN) (Hinton et al., 2006) is a layered stochastic network with undirected bipartite interactions between the units in the top two layers, and directed bipartite interactions between the units in all other subsequent pairs of layers, directed towards the bottom layer. The top two layers form a restricted Boltzmann machine (RBM) (Smolensky, 1986). The entire network defines a model of probability distributions on the states of the units in the bottom layer, the visible layer. When the number of units in every layer has the same order of magnitude, the network is called narrow . The depth refers to the number of layers. Deep network architectures are believed to play a key role in information processing of intelligent agents, see (Bengio, 2009) for an overview on this exciting topic. DBNs were the first deep architectures to be envisaged together with an efficient unsupervised training algorithm (Hinton et al., 2006).

Generative Stochastic Networks (GSNs) have been recently introduced as an alternative to traditional probabilistic modeling: instead of parametrizing the data distribution directly, one parametrizes a transition operator for a Markov chain whose stationary distribution is an estimator of the data generating distribution. The result of training is therefore a machine that generates samples through this Markov chain. However, the previously introduced GSN consistency theorems suggest that in order to capture a wide class of distributions, the transition operator in general should be multimodal, something that has not been done before this paper. We introduce for the first time multimodal transition distributions for GSNs, in particular using models in the NADE family (Neural Autoregressive Density Estimator) as output distributions of the transition operator. A NADE model is related to an RBM (and can thus model multimodal distributions) but its likelihood (and likelihood gradient) can be computed easily. The parameters of the NADE are obtained as a learned function of the previous state of the learned Markov chain. Experiments clearly illustrate the advantage of such multimodal transition distributions over unimodal GSNs.

We present a theoretical analysis of Gaussian-binary restricted Boltzmann machines (GRBMs) from the perspective of density models. The key aspect of this analysis is to show that GRBMs can be formulated as a constrained mixture of Gaussians, which gives a much better insight into the model's capabilities and limitations. We show that GRBMs are capable of learning meaningful features both in a two-dimensional blind source separation task and in modeling natural images. Further, we show that reported difficulties in training GRBMs are due to the failure of the training algorithm rather than the model itself. Based on our analysis we are able to propose several training recipes, which allowed successful and fast training in our experiments. Finally, we discuss the relationship of GRBMs to several modifications that have been proposed to improve the model.