Efficient softmax approximation for GPUs

We propose an approximate strategy to efficiently train neural network based language models over very large vocabularies. Our approach, called adaptive softmax, circumvents the linear dependency on the vocabulary size by exploiting the unbalanced word distribution to form clusters that explicitly minimize the expectation of computation time. Our approach further reduces the computational time by exploiting the specificities of modern architectures and matrix-matrix vector operations, making it particularly suited for graphical processing units. Our experiments carried out on standard benchmarks, such as EuroParl and One Billion Word, show that our approach brings a large gain in efficiency over standard approximations while achieving an accuracy close to that of the full softmax. The code of our method is available at https://github.com/facebookresearch/adaptive-softmax.

Vocabulary Selection Strategies for Neural Machine Translation

Classical translation models constrain the space of possible outputs by selecting a subset of translation rules based on the input sentence. Recent work on improving the efficiency of neural translation models adopted a similar strategy by restricting the output vocabulary to a subset of likely candidates given the source. In this paper we experiment with context and embedding-based selection methods and extend previous work by examining speed and accuracy trade-offs in more detail. We show that decoding time on CPUs can be reduced by up to 90% and training time by 25% on the WMT15 English-German and WMT16 English-Romanian tasks at the same or only negligible change in accuracy. This brings the time to decode with a state of the art neural translation system to just over 140 msec per sentence on a single CPU core for English-German.

Neural Text Generation from Structured Data with Application to the Biography Domain

This paper introduces a neural model for concept-to-text generation that scales to large, rich domains. We experiment with a new dataset of biographies from Wikipedia that is an order of magnitude larger than existing resources with over 700k samples. The dataset is also vastly more diverse with a 400k vocabulary, compared to a few hundred words for Weathergov or Robocup. Our model builds upon recent work on conditional neural language model for text generation. To deal with the large vocabulary, we extend these models to mix a fixed vocabulary with copy actions that transfer sample-specific words from the input database to the generated output sentence. Our neural model significantly out-performs a classical Kneser-Ney language model adapted to this task by nearly 15 BLEU.

Interactive Semantic Featuring for Text Classification

In text classification, dictionaries can be used to define human-comprehensible features. We propose an improvement to dictionary features called smoothed dictionary features. These features recognize document contexts instead of n-grams. We describe a principled methodology to solicit dictionary features from a teacher, and present results showing that models built using these human-comprehensible features are competitive with models trained with Bag of Words features.

Predicting distributions with Linearizing Belief Networks

Conditional belief networks introduce stochastic binary variables in neural networks. Contrary to a classical neural network, a belief network can predict more than the expected value of the output $Y$ given the input $X$. It can predict a distribution of outputs $Y$ which is useful when an input can admit multiple outputs whose average is not necessarily a valid answer. Such networks are particularly relevant to inverse problems such as image prediction for denoising, or text to speech. However, traditional sigmoid belief networks are hard to train and are not suited to continuous problems. This work introduces a new family of networks called linearizing belief nets or LBNs. A LBN decomposes into a deep linear network where each linear unit can be turned on or off by non-deterministic binary latent units. It is a universal approximator of real-valued conditional distributions and can be trained using gradient descent. Moreover, the linear pathways efficiently propagate continuous information and they act as multiplicative skip-connections that help optimization by removing gradient diffusion. This yields a model which trains efficiently and improves the state-of-the-art on image denoising and facial expression generation with the Toronto faces dataset.

Strategies for Training Large Vocabulary Neural Language Models

Training neural network language models over large vocabularies is still computationally very costly compared to count-based models such as Kneser-Ney. At the same time, neural language models are gaining popularity for many applications such as speech recognition and machine translation whose success depends on scalability. We present a systematic comparison of strategies to represent and train large vocabularies, including softmax, hierarchical softmax, target sampling, noise contrastive estimation and self normalization. We further extend self normalization to be a proper estimator of likelihood and introduce an efficient variant of softmax. We evaluate each method on three popular benchmarks, examining performance on rare words, the speed/accuracy trade-off and complementarity to Kneser-Ney.

ICE: Enabling Non-Experts to Build Models Interactively for Large-Scale Lopsided Problems

Quick interaction between a human teacher and a learning machine presents numerous benefits and challenges when working with web-scale data. The human teacher guides the machine towards accomplishing the task of interest. The learning machine leverages big data to find examples that maximize the training value of its interaction with the teacher. When the teacher is restricted to labeling examples selected by the machine, this problem is an instance of active learning. When the teacher can provide additional information to the machine (e.g., suggestions on what examples or predictive features should be used) as the learning task progresses, then the problem becomes one of interactive learning. To accommodate the two-way communication channel needed for efficient interactive learning, the teacher and the machine need an environment that supports an interaction language. The machine can access, process, and summarize more examples than the teacher can see in a lifetime. Based on the machine's output, the teacher can revise the definition of the task or make it more precise. Both the teacher and the machine continuously learn and benefit from the interaction. We have built a platform to (1) produce valuable and deployable models and (2) support research on both the machine learning and user interface challenges of the interactive learning problem. The platform relies on a dedicated, low-latency, distributed, in-memory architecture that allows us to construct web-scale learning machines with quick interaction speed. The purpose of this paper is to describe this architecture and demonstrate how it supports our research efforts. Preliminary results are presented as illustrations of the architecture but are not the primary focus of the paper.