End-to-end models in NLP rarely encode external world knowledge about length of time. We introduce two effective models for duration prediction, which incorporate external knowledge by reading temporal-related news sentences (time-aware pre-training). Specifically, one model predicts the range/unit where the duration value falls in (R-pred); and the other predicts the exact duration value E-pred. Our best model -- E-pred, substantially outperforms previous work, and captures duration information more accurately than R-pred. We also demonstrate our models are capable of duration prediction in the unsupervised setting, outperforming the baselines.

Neural sequence models can generate highly fluent sentences but recent studies have also shown that they are also prone to hallucinate additional content not supported by the input, which can cause a lack of trust in the model. To better assess the faithfulness of the machine outputs, we propose a new task to predict whether each token in the output sequence is hallucinated conditioned on the source input, and collect new manually annotated evaluation sets for this task. We also introduce a novel method for learning to model hallucination detection, based on pretrained language models fine tuned on synthetic data that includes automatically inserted hallucinations. Experiments on machine translation and abstract text summarization demonstrate the effectiveness of our proposed approach -- we obtain an average F1 of around 0.6 across all the benchmark datasets and achieve significant improvements in sentence-level hallucination scoring compared to baseline methods. We also release our annotated data and code for future research at https://github.com/violet-zct/fairseq-detect-hallucination.

When neural networks are used to solve differential equations, they usually produce solutions in the form of black-box functions that are not directly mathematically interpretable. We introduce a method for generating symbolic expressions to solve differential equations while leveraging deep learning training methods. Unlike existing methods, our system does not require learning a language model over symbolic mathematics, making it scalable, compact, and easily adaptable for a variety of tasks and configurations. As part of the method, we propose a novel neural architecture for learning mathematical expressions to optimize a customizable objective. The system is designed to always return a valid symbolic formula, generating a useful approximation when an exact analytic solution to a differential equation is not or cannot be found. We demonstrate through examples how our method can be applied on a number of differential equations, often obtaining symbolic approximations that are useful or insightful. Furthermore, we show how the system can be effortlessly generalized to find symbolic solutions to other mathematical tasks, including integration and functional equations.

Autonomous vehicles (AVs) need to share the road with multiple, heterogeneous road users in a variety of driving scenarios. It is overwhelming and unnecessary to carefully interact with all observed agents, and AVs need to determine whether and when to interact with each surrounding agent. In order to facilitate the design and testing of prediction and planning modules of AVs, in-depth understanding of interactive behavior is expected with proper representation, and events in behavior data need to be extracted and categorized automatically. Answers to what are the essential patterns of interactions are also crucial for these motivations in addition to answering whether and when. Thus, learning to extract interactive driving events and patterns from human data for tackling the whether-when-what tasks is of critical importance for AVs. There is, however, no clear definition and taxonomy of interactive behavior, and most of the existing works are based on either manual labelling or hand-crafted rules and features. In this paper, we propose the Interactive Driving event and pattern Extraction Network (IDE-Net), which is a deep learning framework to automatically extract interaction events and patterns directly from vehicle trajectories. In IDE-Net, we leverage the power of multi-task learning and proposed three auxiliary tasks to assist the pattern extraction in an unsupervised fashion. We also design a unique spatial-temporal block to encode the trajectory data. Experimental results on the INTERACTION dataset verified the effectiveness of such designs in terms of better generalizability and effective pattern extraction. We find three interpretable patterns of interactions, bringing insights for driver behavior representation, modeling and comprehension. Both objective and subjective evaluation metrics are adopted in our analysis of the learned patterns.

Deep Recurrent Neural Networks (RNN) continues to find success in predictive decision-making with temporal event sequences. Recent studies have shown the importance and practicality of visual analytics in interpreting deep learning models for real-world applications. However, very limited work enables interactions with deep learning models and guides practitioners to form hypotheticals towards the desired prediction outcomes, especially for sequence prediction. Specifically, no existing work has addressed the what-if analysis and value perturbation along different time-steps for sequence outcome prediction. We present a model-agnostic visual analytics tool, HypperSteer, that steers hypothetical testing and allows users to perturb data for sequence predictions interactively. We showcase how HypperSteer helps in steering patient data to achieve desired treatment outcomes and discuss how HypperSteer can serve as a comprehensive solution for other practical scenarios.

Recent years have seen big advances in the field of sentence-level quality estimation (QE), largely as a result of using neural-based architectures. However, the majority of these methods work only on the language pair they are trained on and need retraining for new language pairs. This process can prove difficult from a technical point of view and is usually computationally expensive. In this paper we propose a simple QE framework based on cross-lingual transformers, and we use it to implement and evaluate two different neural architectures. Our evaluation shows that the proposed methods achieve state-of-the-art results outperforming current open-source quality estimation frameworks when trained on datasets from WMT. In addition, the framework proves very useful in transfer learning settings, especially when dealing with low-resourced languages, allowing us to obtain very competitive results.

Performing fact verification based on structured data is important for many real-life applications and is a challenging research problem, particularly when it involves both symbolic operations and informal inference based on language understanding. In this paper, we present a Program-enhanced Verbalization and Graph Attention Network (ProgVGAT) to integrate programs and execution into textual inference models. Specifically, a verbalization with program execution model is proposed to accumulate evidences that are embedded in operations over the tables. Built on that, we construct the graph attention verification networks, which are designed to fuse different sources of evidences from verbalized program execution, program structures, and the original statements and tables, to make the final verification decision. To support the above framework, we propose a program selection module optimized with a new training strategy based on margin loss, to produce more accurate programs, which is shown to be effective in enhancing the final verification results. Experimental results show that the proposed framework achieves the new state-of-the-art performance, a 74.4% accuracy, on the benchmark dataset TABFACT.

Artificial intelligence researchers have not been successful in giving intelligent agents the common-sense knowledge they need to reason about the world. Without this knowledge, it is impossible for intelligent agents to truly interact with the world. Traditionally, there have been two unsuccessful approaches to getting computers to reason about the world--symbolic logic and deep learning. A new project, called COMET, tries to bring these two approaches together. Although it has not yet succeeded, it offers the possibility of progress.

OpenAI has come up with a language generator GPT-3, which is a successor of GPT-2. This newly developed AI was put forward to a few selected outside software developers for testing. GPT-2 released a year prior, and it let out convincing streams in regards to message in the extent of different styles when induced with an underlying sentence. The differentiating factor of GPT-3 is having 175 billion parameters(the qualities that a neural system attempts to upgrade during preparing), whereas GPT-2 had only 1.5 billion. GPT-3 is the most significant language model ever.

Artificial intelligence has arrived in our everyday lives -- from search engines to self-driving cars. This has to do with the enormous computing power that has become available in recent years. But new results from AI research now show that simpler, smaller neural networks can be used to solve certain tasks even better, more efficiently, and more reliably than ever before. An international research team from TU Wien (Vienna), IST Austria and MIT (USA) has developed a new artificial intelligence system based on the brains of tiny animals, such as threadworms. This novel AI-system can control a vehicle with just a few artificial neurons.