The use of explicit object detectors as an intermediate step to image captioning - which used to constitute an essential stage in early work - is often bypassed in the currently dominant end-to-end approaches, where the language model is conditioned directly on a mid-level image embedding. We argue that explicit detections provide rich semantic information, and can thus be used as an interpretable representation to better understand why end-to-end image captioning systems work well. We provide an in-depth analysis of end-to-end image captioning by exploring a variety of cues that can be derived from such object detections. Our study reveals that end-to-end image captioning systems rely on matching image representations to generate captions, and that encoding the frequency, size and position of objects are complementary and all play a role in forming a good image representation. It also reveals that different object categories contribute in different ways towards image captioning.

We present a learning-based system for rapid mass-scale material synthesis that is useful for novice and expert users alike. The user preferences are learned via Gaussian Process Regression and can be easily sampled for new recommendations. Typically, each recommendation takes 40-60 seconds to render with global illumination, which makes this process impracticable for real-world workflows. Our neural network eliminates this bottleneck by providing high-quality image predictions in real time, after which it is possible to pick the desired materials from a gallery and assign them to a scene in an intuitive manner. Workflow timings against Disney's "principled" shader reveal that our system scales well with the number of sought materials, thus empowering even novice users to generate hundreds of high-quality material models without any expertise in material modeling. Similarly, expert users experience a significant decrease in the total modeling time when populating a scene with materials. Furthermore, our proposed solution also offers controllable recommendations and a novel latent space variant generation step to enable the real-time fine-tuning of materials without requiring any domain expertise.

We introduce a dynamical spatio-temporal model formalized as a recurrent neural network for forecasting time series of spatial processes, i.e. series of observations sharing temporal and spatial dependencies. The model learns these dependencies through a structured latent dynamical component, while a decoder predicts the observations from the latent representations. We consider several variants of this model, corresponding to different prior hypothesis about the spatial relations between the series. The model is evaluated and compared to state-of-the-art baselines, on a variety of forecasting problems representative of different application areas: epidemiology, geo-spatial statistics and car-traffic prediction. Besides these evaluations, we also describe experiments showing the ability of this approach to extract relevant spatial relations.

Current neural network-based classifiers are susceptible to adversarial examples even in the black-box setting, where the attacker only has query access to the model. In practice, the threat model for real-world systems is often more restrictive than the typical black-box model of full query access. We define three realistic threat models that more accurately characterize many real-world classifiers: the query-limited setting, the partial-information setting, and the label-only setting. We develop new attacks that fool classifiers under these more restrictive threat models, where previous methods would be impractical or ineffective. We demonstrate that our methods are effective against an ImageNet classifier under our proposed threat models. We also demonstrate a targeted black-box attack against a commercial classifier, overcoming the challenges of limited query access, partial information, and other practical issues to attack the Google Cloud Vision API.

Current end-to-end machine reading and question answering (Q\&A) models are primarily based on recurrent neural networks (RNNs) with attention. Despite their success, these models are often slow for both training and inference due to the sequential nature of RNNs. We propose a new Q\&A architecture called QANet, which does not require recurrent networks: Its encoder consists exclusively of convolution and self-attention, where convolution models local interactions and self-attention models global interactions. On the SQuAD dataset, our model is 3x to 13x faster in training and 4x to 9x faster in inference, while achieving equivalent accuracy to recurrent models. The speed-up gain allows us to train the model with much more data. We hence combine our model with data generated by backtranslation from a neural machine translation model. On the SQuAD dataset, our single model, trained with augmented data, achieves 84.6 F1 score on the test set, which is significantly better than the best published F1 score of 81.8.

The SimpleQuestions dataset is one of the most commonly used benchmarks for studying single-relation factoid questions. In this paper, we present new evidence that this benchmark can be nearly solved by standard methods. First we show that ambiguity in the data bounds performance on this benchmark at 83.4%; there are often multiple answers that cannot be disambiguated from the linguistic signal alone. Second we introduce a baseline that sets a new state-of-the-art performance level at 78.1% accuracy, despite using standard methods. Finally, we report an empirical analysis showing that the upperbound is loose; roughly a third of the remaining errors are also not resolvable from the linguistic signal. Together, these results suggest that the SimpleQuestions dataset is nearly solved.

A critical and challenging problem in reinforcement learning is how to learn the state-action value function from the experience replay buffer and simultaneously keep sample efficiency and faster convergence to a high quality solution. In prior works, transitions are uniformly sampled at random from the replay buffer or sampled based on their priority measured by temporal-difference (TD) error. However, these approaches do not fully take into consideration the intrinsic characteristics of transition distribution in the state space and could result in redundant and unnecessary TD updates, slowing down the convergence of the learning procedure. To overcome this problem, we propose a novel state distribution-aware sampling method to balance the replay times for transitions with skew distribution, which takes into account both the occurrence frequencies of transitions and the uncertainty of state-action values. Consequently, our approach could reduce the unnecessary TD updates and increase the TD updates for state-action value with more uncertainty, making the experience replay more effective and efficient. Extensive experiments are conducted on both classic control tasks and Atari 2600 games based on OpenAI gym platform and the experimental results demonstrate the effectiveness of our approach in comparison with the standard DQN approach.

Deep Reinforcement Learning (deep RL) has made several breakthroughs in recent years in applications ranging from complex control tasks in unmanned vehicles to game playing. Despite their success, deep RL still lacks several important capacities of human intelligence, such as transfer learning, abstraction and interpretability. Deep Symbolic Reinforcement Learning (DSRL) seeks to incorporate such capacities to deep Q-networks (DQN) by learning a relevant symbolic representation prior to using Q-learning. In this paper, we propose a novel extension of DSRL, which we call Symbolic Reinforcement Learning with Common Sense (SRL CS), offering a better balance between generalization and specialization, inspired by principles of common sense when assigning rewards and aggregating Q-values. Experiments reported in this paper show that SRL CS learns consistently faster than Q-learning and DSRL, achieving also a higher accuracy. In the hardest case, where agents were trained in a deterministic environment and tested in a random environment, SRL CS achieves nearly 100% average accuracy compared to DSRL's 70% and DQN's 50% accuracy. To the best of our knowledge, this is the first case of near perfect zero-shot transfer learning using Reinforcement Learning.

The process of aligning a pair of shapes is a fundamental operation in computer graphics. Traditional approaches rely heavily on matching corresponding points or features to guide the alignment, a paradigm that falters when significant shape portions are missing. These techniques generally do not incorporate prior knowledge about expected shape characteristics, which can help compensate for any misleading cues left by inaccuracies exhibited in the input shapes. We present an approach based on a deep neural network, leveraging shape datasets to learn a shape-aware prior for source-to-target alignment that is robust to shape incompleteness. In the absence of ground truth alignments for supervision, we train a network on the task of shape alignment using incomplete shapes generated from full shapes for self-supervision. Our network, called ALIGNet, is trained to warp complete source shapes to incomplete targets, as if the target shapes were complete, thus essentially rendering the alignment partial-shape agnostic. We aim for the network to develop specialized expertise over the common characteristics of the shapes in each dataset, thereby achieving a higher-level understanding of the expected shape space to which a local approach would be oblivious. We constrain ALIGNet through an anisotropic total variation identity regularization to promote piecewise smooth deformation fields, facilitating both partial-shape agnosticism and post-deformation applications. We demonstrate that ALIGNet learns to align geometrically distinct shapes, and is able to infer plausible mappings even when the target shape is significantly incomplete. We show that our network learns the common expected characteristics of shape collections, without over-fitting or memorization, enabling it to produce plausible deformations on unseen data during test time.

Neural network frameworks such as PyTorch and TensorFlow are the workhorses of numerous machine learning applications ranging from object recognition to machine translation. While these frameworks are versatile and straightforward to use, the training of and inference in deep neural networks is resource (energy, compute, and memory) intensive. In contrast to recent works focusing on algorithmic enhancements, we introduce BrainSlug, a framework that transparently accelerates neural network workloads by changing the default layer-by-layer processing to a depth-first approach, reducing the amount of data required by the computations and thus improving the performance of the available hardware caches. BrainSlug achieves performance improvements of up to 41.1% on CPUs and 35.7% on GPUs. These optimizations come at zero cost to the user as they do not require hardware changes and only need tiny adjustments to the software.