Videos of actions are complex spatio-temporal signals, containing rich compositional structures. Current generative models are limited in their ability to generate examples of object configurations outside the range they were trained on. Towards this end, we introduce a generative model (AG2Vid) based on Action Graphs, a natural and convenient structure that represents the dynamics of actions between objects over time. Our AG2Vid model disentangles appearance and position features, allowing for more accurate generation. AG2Vid is evaluated on the CATER and Something-Something datasets and outperforms other baselines. Finally, we show how Action Graphs can be used for generating novel compositions of unseen actions.
Object Permanence allows people to reason about the location of non-visible objects, by understanding that they continue to exist even when not perceived directly. Object Permanence is critical for building a model of the world, since objects in natural visual scenes dynamically occlude and contain each-other. Intensive studies in developmental psychology suggest that object permanence is a challenging task that is learned through extensive experience. Here we introduce the setup of learning Object Permanence from data. We explain why this learning problem should be dissected into four components, where objects are (1) visible, (2) occluded, (3) contained by another object and (4) carried by a containing object. The fourth subtask, where a target object is carried by a containing object, is particularly challenging because it requires a system to reason about a moving location of an invisible object. We then present a unified deep architecture that learns to predict object location under these four scenarios. We evaluate the architecture and system on a new dataset based on CATER, and find that it outperforms previous localization methods and various baselines.
Training overparameterized convolutional neural networks with gradient based methods is the most successful learning method for image classification. However, its theoretical properties are far from understood even for very simple learning tasks. In this work, we consider a simplified image classification task where images contain orthogonal patches and are learned with a 3-layer overparameterized convolutional network and stochastic gradient descent. We empirically identify a novel phenomenon where the dot-product between the learned pattern detectors and their detected patterns are governed by the pattern statistics in the training set. We call this phenomenon Pattern Statistics Inductive Bias (PSI) and prove that PSI holds for a simple setup with two points in the training set. Furthermore, we prove that if PSI holds, stochastic gradient descent has sample complexity $O(d^2\log(d))$ where $d$ is the filter dimension. In contrast, we show a VC dimension lower bound in our setting which is exponential in $d$. Taken together, our results provide strong evidence that PSI is a unique inductive bias of stochastic gradient descent, that guarantees good generalization properties.
Generating realistic images of complex visual scenes becomes very challenging when one wishes to control the structure of the generated images. Previous approaches showed that scenes with few entities can be controlled using scene graphs, but this approach struggles as the complexity of the graph (number of objects and edges) increases. Moreover, current approaches fail to generalize conditioned on the number of objects or when given different input graphs which are semantic equivalent. In this work, we propose a novel approach to mitigate these issues. We present a novel model which can inherently learn canonical graph representations, thus ensuring that semantically similar scene graphs will result in similar predictions. In addition, the proposed model can better capture object representation independently of the number of objects in the graph. We show improved performance of the model on three different benchmarks: Visual Genome, COCO and CLEVR.
We introduce a novel method for multilingual transfer that utilizes deep contextual embeddings, pretrained in an unsupervised fashion. While contextual embeddings have been shown to yield richer representations of meaning compared to their static counterparts, aligning them poses a challenge due to their dynamic nature. To this end, we construct context-independent variants of the original monolingual spaces and utilize their mapping to derive an alignment for the context-dependent spaces. This mapping readily supports processing of a target language, improving transfer by context-aware embeddings. Our experimental results demonstrate the effectiveness of this approach for zero-shot and few-shot learning of dependency parsing. Specifically, our method consistently outperforms the previous state-of-the-art on 6 tested languages, yielding an improvement of 6.8 LAS points on average.
Understanding the semantics of complex visual scenes involves perception of entities and reasoning about their relations. Scene graphs provide a natural representation for these tasks, by assigning labels to both entities (nodes) and relations (edges). However, scene graphs are not commonly used as intermediate components in visual reasoning systems, for two complementary reasons. First, training models to map images to scene graphs requires prohibitive manual annotation, and results in graphs that often do not match the needs of a downstream visual reasoning application. Second, using these discrete graphs as an intermediate latent representation results in a non-differentiable function that is difficult to optimize. Here we propose Differentiable Scene Graphs (DSGs), an image representation that is amenable to differentiable end-to-end optimization, and requires supervision only from the downstream tasks. DSGs provide a dense representation for all regions and pairs of regions, investing model capacity on the important aspects of the image. We describe a multi-task objective function that allows us to learn this representation from indirect supervision only, provided by the downstream task. We evaluate our model on the challenging task of identifying referring relationships, and show that training DSGs using our multi-task objective leads to new state-of-the-art performance.
Events defined by the interaction of objects in a scene often are of critical importance, yet such events are typically rare and available labeled examples insufficient to train a conventional deep model that performs well across expected object appearances. Most deep learning activity recognition models focus on global context aggregation and do not explicitly consider object interactions inside the video, potentially overlooking important cues relevant to interpreting activity in the scene. In this paper, we show that a new model for explicit representation of object interactions significantly improves deep video activity classification for driving collision detection. We propose a Spatio-Temporal Action Graph (STAG) network, which incorporates spatial and temporal relations of objects. The network is automatically learned from data, with a latent graph structure inferred for the task. As a benchmark to evaluate performance on collision detection tasks, we introduce a novel data set based on data obtained from real life driving collisions and near-collisions. This data set reflects the challenging task of detecting and classifying accidents in a richly varying but yet highly constrained setting, that is very relevant to the evaluation of autonomous driving and alerting systems. Our experiments confirm that our STAG model offers significantly improved results for collision activity classification.
Machine understanding of complex images is a key goal of artificial intelligence. One challenge underlying this task is that visual scenes contain multiple inter-related objects, and that global context plays an important role in interpreting the scene. A natural modeling framework for capturing such effects is structured prediction, which optimizes over complex labels, while modeling within-label interactions. However, it is unclear what principles should guide the design of a structured prediction model that utilizes the power of deep learning components. Here we propose a design principle for such architectures that follows from a natural requirement of permutation invariance. We prove a necessary and sufficient characterization for architectures that follow this invariance, and discuss its implication on model design. Finally, we show that the resulting model achieves new state of the art results on the Visual Genome scene graph labeling benchmark, outperforming all recent approaches.
Designing a reliable natural language (NL) interface for querying tables has been a longtime goal of researchers in both the data management and natural language processing (NLP) communities. Such an interface receives as input an NL question, translates it into a formal query, executes the query and returns the results. Errors in the translation process are not uncommon, and users typically struggle to understand whether their query has been mapped correctly. We address this problem by explaining the obtained formal queries to non-expert users. Two methods for query explanations are presented: the first translates queries into NL, while the second method provides a graphic representation of the query cell-based provenance (in its execution on a given table). Our solution augments a state-of-the-art NL interface over web tables, enhancing it in both its training and deployment phase. Experiments, including a user study conducted on Amazon Mechanical Turk, show our solution to improve both the correctness and reliability of an NL interface.
Training semantic parsers from weak supervision (denotations) rather than strong supervision (programs) complicates training in two ways. First, a large search space of potential programs needs to be explored at training time to find a correct program. Second, spurious programs that accidentally lead to a correct denotation add noise to training. In this work we propose that in closed worlds with clear semantic types, one can substantially alleviate these problems by utilizing an abstract representation, where tokens in both the language utterance and program are lifted to an abstract form. We show that these abstractions can be defined with a handful of lexical rules and that they result in sharing between different examples that alleviates the difficulties in training. To test our approach, we develop the first semantic parser for CNLVR, a challenging visual reasoning dataset, where the search space is large and overcoming spuriousness is critical, because denotations are either TRUE or FALSE, and thus random programs are likely to lead to a correct denotation. Our method substantially improves performance, and reaches 82.5% accuracy, a 14.7% absolute accuracy improvement compared to the best reported accuracy so far.