Neural network compression techniques are able to reduce the parameter counts of trained networks by over 90%--decreasing storage requirements and improving inference performance--without compromising accuracy. However, contemporary experience is that it is difficult to train small architectures from scratch, which would similarly improve training performance. We articulate a new conjecture to explain why it is easier to train large networks: the "lottery ticket hypothesis." It states that large networks that train successfully contain subnetworks that--when trained in isolation--converge in a comparable number of iterations to comparable accuracy. These subnetworks, which we term "winning tickets," have won the initialization lottery: their connections have initial weights that make training particularly effective. We find that a standard technique for pruning unnecessary network weights naturally uncovers a subnetwork which, at the start of training, comprised a winning ticket. We present an algorithm to identify winning tickets and a series of experiments that support the lottery ticket hypothesis. We consistently find winning tickets that are less than 20% of the size of several fully-connected, convolutional, and residual architectures for MNIST and CIFAR10. Furthermore, winning tickets at moderate levels of pruning (20-50% of the original network size) converge up to 6.7x faster than the original network and exhibit higher test accuracy. Click to Read Paper
Recurrent neural network grammars (RNNGs) are generative models of (tree,string) pairs that rely on neural networks to evaluate derivational choices. Parsing with them using beam search yields a variety of incremental complexity metrics such as word surprisal and parser action count. When used as regressors against human electrophysiological responses to naturalistic text, they derive two amplitude effects: an early peak and a P600-like later peak. By contrast, a non-syntactic neural language model yields no reliable effects. Model comparisons attribute the early peak to syntactic composition within the RNNG. This pattern of results recommends the RNNG+beam search combination as a mechanistic model of the syntactic processing that occurs during normal human language comprehension. Click to Read Paper
We explore building generative neural network models of popular reinforcement learning environments. Our world model can be trained quickly in an unsupervised manner to learn a compressed spatial and temporal representation of the environment. By using features extracted from the world model as inputs to an agent, we can train a very compact and simple policy that can solve the required task. We can even train our agent entirely inside of its own hallucinated dream generated by its world model, and transfer this policy back into the actual environment. An interactive version of this paper is available at Click to Read Paper
What can we learn from a connectome? We constructed a simplified model of the first two stages of the fly visual system, the lamina and medulla. The resulting hexagonal lattice convolutional network was trained using backpropagation through time to perform object tracking in natural scene videos. Networks initialized with weights from connectome reconstructions automatically discovered well-known orientation and direction selectivity properties in T4 neurons and their inputs, while networks initialized at random did not. Our work is the first demonstration, that knowledge of the connectome can enable in silico predictions of the functional properties of individual neurons in a circuit, leading to an understanding of circuit function from structure alone. Click to Read Paper
This paper describes the submissions of the "Marian" team to the WNMT 2018 shared task. We investigate combinations of teacher-student training, low-precision matrix products, auto-tuning and other methods to optimize the Transformer model on GPU and CPU. By further integrating these methods with the new averaging attention networks, a recently introduced faster Transformer variant, we create a number of high-quality, high-performance models on the GPU and CPU, dominating the Pareto frontier for this shared task. Click to Read Paper
We introduce autoregressive implicit quantile networks (AIQN), a fundamentally different approach to generative modeling than those commonly used, that implicitly captures the distribution using quantile regression. AIQN is able to achieve superior perceptual quality and improvements in evaluation metrics, without incurring a loss of sample diversity. The method can be applied to many existing models and architectures. In this work we extend the PixelCNN model with AIQN and demonstrate results on CIFAR-10 and ImageNet using Inception score, FID, non-cherry-picked samples, and inpainting results. We consistently observe that AIQN yields a highly stable algorithm that improves perceptual quality while maintaining a highly diverse distribution. Click to Read Paper
Recurrent neural networks are a powerful tool for modeling sequential data, but the dependence of each timestep's computation on the previous timestep's output limits parallelism and makes RNNs unwieldy for very long sequences. We introduce quasi-recurrent neural networks (QRNNs), an approach to neural sequence modeling that alternates convolutional layers, which apply in parallel across timesteps, and a minimalist recurrent pooling function that applies in parallel across channels. Despite lacking trainable recurrent layers, stacked QRNNs have better predictive accuracy than stacked LSTMs of the same hidden size. Due to their increased parallelism, they are up to 16 times faster at train and test time. Experiments on language modeling, sentiment classification, and character-level neural machine translation demonstrate these advantages and underline the viability of QRNNs as a basic building block for a variety of sequence tasks. Click to Read Paper
We propose a memory augmented neural network to perform text normalization i.e. the transformation of words from the written to the spoken form. With the addition of dynamic memory access and storage mechanism, we present an architecture that will serve as a language agnostic text normalization system while avoiding the kind of silly errors made by the LSTM based recurrent neural architectures. By reducing the number of unacceptable mistakes, we show that such a novel architecture is indeed a better alternative. Our proposed system requires significantly lesser amounts of data, training time and compute resources. However, some occurrences of errors still remain in certain semiotic classes. Nevertheless, we demonstrate that memory augmented networks with meta-learning capabilities can open many doors to a superior text normalization system. Click to Read Paper
Semantic parsing is the task of transducing natural language (NL) utterances into formal meaning representations (MRs), commonly represented as tree structures. Annotating NL utterances with their corresponding MRs is expensive and time-consuming, and thus the limited availability of labeled data often becomes the bottleneck of data-driven, supervised models. We introduce StructVAE, a variational auto-encoding model for semisupervised semantic parsing, which learns both from limited amounts of parallel data, and readily-available unlabeled NL utterances. StructVAE models latent MRs not observed in the unlabeled data as tree-structured latent variables. Experiments on semantic parsing on the ATIS domain and Python code generation show that with extra unlabeled data, StructVAE outperforms strong supervised models. Click to Read Paper