Deeper neural networks are more difficult to train. We present a residual learning framework to ease the training of networks that are substantially deeper than those used previously. We explicitly reformulate the layers as learning residual functions with reference to the layer inputs, instead of learning unreferenced functions. We provide comprehensive empirical evidence showing that these residual networks are easier to optimize, and can gain accuracy from considerably increased depth. On the ImageNet dataset we evaluate residual nets with a depth of up to 152 layers—8× deeper than VGG nets [40] but still having lower complexity. An ensemble of these residual nets achieves 3.57% error on the ImageNet test set. This result won the 1st place on the ILSVRC 2015 classification task. We also present analysis on CIFAR-10 with 100 and 1000 layers. The depth of representations is of central importance for many visual recognition tasks. Solely due to our extremely deep representations, we obtain a 28% relative improvement on the COCO object detection dataset. Deep residual nets are foundations of our submissions to ILSVRC & COCO 2015 competitions1, where we also won the 1st places on the tasks of ImageNet detection, ImageNet localization, COCO detection, and COCO segmentation.

/pdf/deep-residual-learning-for-image-recognition-b75jg61qrq.pdf

Deep Residual Learning for Image Recognition

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

Deep learning allows computational models that are composed of multiple processing layers to learn representations of data with multiple levels of abstraction. These methods have dramatically improved the state-of-the-art in speech recognition, visual object recognition, object detection and many other domains such as drug discovery and genomics. Deep learning discovers intricate structure in large data sets by using the backpropagation algorithm to indicate how a machine should change its internal parameters that are used to compute the representation in each layer from the representation in the previous layer. Deep convolutional nets have brought about breakthroughs in processing images, video, speech and audio, whereas recurrent nets have shone light on sequential data such as text and speech.

Deep learning

Deeper neural networks are more difficult to train. We present a residual learning framework to ease the training of networks that are substantially deeper than those used previously. We explicitly reformulate the layers as learning residual functions with reference to the layer inputs, instead of learning unreferenced functions. We provide comprehensive empirical evidence showing that these residual networks are easier to optimize, and can gain accuracy from considerably increased depth. On the ImageNet dataset we evaluate residual nets with a depth of up to 152 layers---8x deeper than VGG nets but still having lower complexity. An ensemble of these residual nets achieves 3.57% error on the ImageNet test set. This result won the 1st place on the ILSVRC 2015 classification task. We also present analysis on CIFAR-10 with 100 and 1000 layers. 
The depth of representations is of central importance for many visual recognition tasks. Solely due to our extremely deep representations, we obtain a 28% relative improvement on the COCO object detection dataset. Deep residual nets are foundations of our submissions to ILSVRC & COCO 2015 competitions, where we also won the 1st places on the tasks of ImageNet detection, ImageNet localization, COCO detection, and COCO segmentation.

Multilayer neural networks trained with the back-propagation algorithm constitute the best example of a successful gradient based learning technique. Given an appropriate network architecture, gradient-based learning algorithms can be used to synthesize a complex decision surface that can classify high-dimensional patterns, such as handwritten characters, with minimal preprocessing. This paper reviews various methods applied to handwritten character recognition and compares them on a standard handwritten digit recognition task. Convolutional neural networks, which are specifically designed to deal with the variability of 2D shapes, are shown to outperform all other techniques. Real-life document recognition systems are composed of multiple modules including field extraction, segmentation recognition, and language modeling. A new learning paradigm, called graph transformer networks (GTN), allows such multimodule systems to be trained globally using gradient-based methods so as to minimize an overall performance measure. Two systems for online handwriting recognition are described. Experiments demonstrate the advantage of global training, and the flexibility of graph transformer networks. A graph transformer network for reading a bank cheque is also described. It uses convolutional neural network character recognizers combined with global training techniques to provide record accuracy on business and personal cheques. It is deployed commercially and reads several million cheques per day.

Gradient-based learning applied to document recognition

Recurrent neural networks can be used to map input sequences to output sequences, such as for recognition, production or prediction problems. However, practical difficulties have been reported in training recurrent neural networks to perform tasks in which the temporal contingencies present in the input/output sequences span long intervals. We show why gradient based learning algorithms face an increasingly difficult problem as the duration of the dependencies to be captured increases. These results expose a trade-off between efficient learning by gradient descent and latching on information for long periods. Based on an understanding of this problem, alternatives to standard gradient descent are considered. >

http://cslt.riit.tsinghua.edu.cn/mediawiki/images/0/03/Learning_Long-Term_Dependencies_with_Gradient_Descent_is_Difficult.pdf

Learning long-term dependencies with gradient descent is difficult

The literature on short-term traffic flow forecasting has undergone great development recently. Many works, describing a wide variety of different approaches, which very often share similar features and ideas, have been published. However, publications presenting new prediction algorithms usually employ different settings, data sets, and performance measurements, making it difficult to infer a clear picture of the advantages and limitations of each model. The aim of this paper is twofold. First, we review existing approaches to short-term traffic flow forecasting methods under the common view of probabilistic graphical models, presenting an extensive experimental comparison, which proposes a common baseline for their performance analysis and provides the infrastructure to operate on a publicly available data set. Second, we present two new support vector regression models, which are specifically devised to benefit from typical traffic flow seasonality and are shown to represent an interesting compromise between prediction accuracy and computational efficiency. The SARIMA model coupled with a Kalman filter is the most accurate model; however, the proposed seasonal support vector regressor turns out to be highly competitive when performing forecasts during the most congested periods.

Short-Term Traffic Flow Forecasting: An Experimental Comparison of Time-Series Analysis and Supervised Learning

Motivation: Predicting the secondary structure of a protein (alpha-helix, beta-sheet, coil) is an important step towards elucidating its three-dimensional structure, as well as its function. Presently, the best predictors are based on machine learning approaches, in particular neural network architectures with a fixed, and relatively short, input window of amino acids, centered at the prediction site. Although a fixed small window avoids overfitting problems, it does not permit capturing variable long-rang information. Results: We introduce a family of novel architectures which can learn to make predictions based on variable ranges of dependencies. These architectures extend recurrent neural networks, introducing non-causal bidirectional dynamics to capture both upstream and downstream information. The prediction algorithm is completed by the use of mixtures of estimators that leverage evolutionary information, expressed in terms of multiple alignments, both at the input and output levels. While our system currently achieves an overall performance close to 76% correct prediction ‐ at least comparable to the best existing systems ‐ the main emphasis here is on the development of new algorithmic

/pdf/exploiting-the-past-and-the-future-in-protein-secondary-31c6uts69i.pdf

Exploiting the past and the future in protein secondary structure prediction.

A structured organization of information is typically required by symbolic processing. On the other hand, most connectionist models assume that data are organized according to relatively poor structures, like arrays or sequences. The framework described in this paper is an attempt to unify adaptive models like artificial neural nets and belief nets for the problem of processing structured information. In particular, relations between data variables are expressed by directed acyclic graphs, where both numerical and categorical values coexist. The general framework proposed in this paper can be regarded as an extension of both recurrent neural networks and hidden Markov models to the case of acyclic graphs. In particular we study the supervised learning problem as the problem of learning transductions from an input structured space to an output structured space, where transductions are assumed to admit a recursive hidden state-space representation. We introduce a graphical formalism for representing this class of adaptive transductions by means of recursive networks, i.e., cyclic graphs where nodes are labeled by variables and edges are labeled by generalized delay elements. This representation makes it possible to incorporate the symbolic and subsymbolic nature of data. Structures are processed by unfolding the recursive network into an acyclic graph called encoding network. In so doing, inference and learning algorithms can be easily inherited from the corresponding algorithms for artificial neural networks or probabilistic graphical model.

/pdf/a-general-framework-for-adaptive-processing-of-data-4rfior7cjo.pdf

A general framework for adaptive processing of data structures

We introduce a framework based on bilevel programming that unifies gradient-based hyperparameter optimization and meta-learning. We show that an approximate version of the bilevel problem can be solved by taking into explicit account the optimization dynamics for the inner objective. Depending on the specific setting, the outer variables take either the meaning of hyperparameters in a supervised learning problem or parameters of a meta-learner. We provide sufficient conditions under which solutions of the approximate problem converge to those of the exact problem. We instantiate our approach for meta-learning in the case of deep learning where representation layers are treated as hyperparameters shared across a set of training episodes. In experiments, we confirm our theoretical findings, present encouraging results for few-shot learning and contrast the bilevel approach against classical approaches for learning-to-learn.

Paolo Frasconi

Papers

Learning long-term dependencies with gradient descent is difficult

Short-Term Traffic Flow Forecasting: An Experimental Comparison of Time-Series Analysis and Supervised Learning

Exploiting the past and the future in protein secondary structure prediction.

A general framework for adaptive processing of data structures

Bilevel Programming for Hyperparameter Optimization and Meta-Learning