Generating a long, coherent text such as a paragraph requires a high-level control of different levels of relations between sentences (e.g., tense, coreference). We call such a logical connection between sentences as a (paragraph) flow. In order to produce a coherent flow of text, we explore two forms of intersentential relations in a paragraph: one is a human-created linguistical relation that forms a structure (e.g., discourse tree) and the other is a relation from latent representation learned from the sentences themselves. Our two proposed models incorporate each form of relations into document-level language models: the former is a supervised model that jointly learns a language model as well as discourse relation prediction, and the latter is an unsupervised model that is hierarchically conditioned by a recurrent neural network (RNN) over the latent information. Our proposed models with both forms of relations outperform the baselines in partially conditioned paragraph generation task. Our codes and data are publicly available.

The recent explosion of false claims in social media and on the Web in general has given rise to a lot of manual fact-checking initiatives. Unfortunately, the number of claims that need to be fact-checked is several orders of magnitude larger than what humans can handle manually. Thus, there has been a lot of research aiming at automating the process. Interestingly, previous work has largely ignored the growing number of claims about images. This is despite the fact that visual imagery is more influential than text and naturally appears alongside fake news. Here we aim at bridging this gap. In particular, we create a new dataset for this problem, and we explore a variety of features modeling the claim, the image, and the relationship between the claim and the image. The evaluation results show sizable improvements over the baseline. We release our dataset, hoping to enable further research on fact-checking claims about images.

Processing human affective behavior is important for developing intelligent agents that interact with humans in complex interaction scenarios. A large number of current approaches that address this problem focus on classifying emotion expressions by grouping them into known categories. Such strategies neglect, among other aspects, the impact of the affective responses from an individual on their interaction partner thus ignoring how people empathize towards each other. This is also reflected in the datasets used to train models for affective processing tasks. Most of the recent datasets, in particular, the ones which capture natural interactions ("in-the-wild" datasets), are designed, collected, and annotated based on the recognition of displayed affective reactions, ignoring how these displayed or expressed emotions are perceived. In this paper, we propose a novel dataset composed of dyadic interactions designed, collected and annotated with a focus on measuring the affective impact that eight different stories have on the listener. Each video of the dataset contains around 5 minutes of interaction where a speaker tells a story to a listener. After each interaction, the listener annotated, using a valence scale, how the story impacted their affective state, reflecting how they empathized with the speaker as well as the story. We also propose different evaluation protocols and a baseline that encourages participation in the advancement of the field of artificial empathy and emotion contagion.

We consider a crowdsourcing data acquisition scenario, such as federated learning, where a Center collects data points from a set of rational Agents, with the aim of training a model. For linear regression models, we show how a payment structure can be designed to incentivize the agents to provide high-quality data as early as possible, based on a characterization of the influence that data points have on the loss function of the model. Our contributions can be summarized as follows: (a) we prove theoretically that this scheme ensures truthful data reporting as a game-theoretic equilibrium and further demonstrate its robustness against mixtures of truthful and heuristic data reports, (b) we design a procedure according to which the influence computation can be efficiently approximated and processed sequentially in batches over time, (c) we develop a theory that allows correcting the difference between the influence and the overall change in loss and (d) we evaluate our approach on real datasets, confirming our theoretical findings.

For his next book, Joseph LeDoux knew he had to go deep. He had to go back in time, way back, 3.5 billion years ago. The author of the seminal The Emotional Brain, followed by Synaptic Self and Anxious, sensed a missing element in those books on how brain anatomy and function shape human behavior and emotions. In his new book, The Deep History of Ourselves: The Four-Billion-Year Story of How We Got Our Conscious Brains, LeDoux takes readers back to the emergence of life on Earth to show what our protean brains today owe to the canny survival of Protozoa. "I started asking, 'How far back in evolution does the ability to detect and respond to danger go?'" he said to me in a recent interview at his home in New York City. LeDoux directs the Emotional Brain Institute at New York University. In his research and previous books, he has shown the human brain processes that detect and respond to danger differ from the conscious experiences of fear itself. "I felt I needed to understand more about this process," he said.

Human "exceptionalism" is for many people an unquestioned assumption. For the religious, it is a God-given fact; for humanists, it is a celebration of our unique mental capacities.

Our sequential non-text information is best harnessed in a Bidirectional LSTM – a type of sequential model described in more detail here and here – that allows the model to learn end-of-sequence and beginning-of-sequence behavior. This maps to domain experts' knowledge that distinctive behavior at the end of the subscription period presages churn. It also captures the patterns in the progression of events over time that can be used to predict eventual churn. On the other hand our textual and categorical data need a separate model to learn from this differently structured data. We have several options here.

When offering customer assistance via a chatbot, do you let customers know they're not talking to an agent? Or do you try to make it seem like they're chatting with an agent? "Tell them right out of the gate. If you mask a chatbot as a human, I think I'm talking to a dumb human," said one contributor during Customer Contact Week's interactive discussion groups, focused on managing the transition between self-service and assisted-service. Lively discussions ensued, centred primarily around chatbots.

As a former Silicon Valley intellectual property attorney for more than 20 years, the potential of disruptive technology has long been of special interest to me. Artificial intelligence (AI) promises to be one of the most important innovations that powers many disruptive ventures and brings exciting changes to our legal system. AI is already influencing the way we work, travel, shop, and play. From autonomous vehicles to improved medical diagnostics to voice assistants, AI is increasingly at the forefront of innovation. As a continuation of the United States Patent and Trademark Office's (USPTO) policy leadership in the field of AI, the USPTO convened a conference on Artificial Intelligence: Intellectual Property Policy Considerations on January 31 this year.

The News: Huawei Technologies has officially unleashed its artificial intelligence (AI) chip Ascend 910, which it says has a maximum power consumption of just 310W–lower than its originally planned specs of 350W. The chip is touted to have "more computing power than any other AI processor", delivering 256 teraflops at half-precision floating point (FP16) and 512 teraflops for integer precision calculations. The Chinese tech giant also announced the commercial availability of its MindSpore AI computing framework, which it said was designed to ease the development of AI applications and improve the efficiencies of such tools. Analyst Take: The news of the Huawei Ascend 910 chip, a powerful AI processor, is a clear sign that Huawei is moving forward with its plans to cut any dependency on the U.S. We saw a similar aggressive course of action from the company just a few weeks ago when it announced Harmony OS, and the intent to make it avaialble to replace Google's Android in the company's smart phones. This would reduce its dependence on U.S. based companies and at the very least could give Huawei smart phones a boost in China.