Classical Chinese poetry is a jewel in the treasure house of Chinese culture. Previous poem generation models only allow users to employ keywords to interfere the meaning of generated poems, leaving the dominion of generation to the model. In this paper, we propose a novel task of generating classical Chinese poems from vernacular, which allows users to have more control over the semantic of generated poems. We adapt the approach of unsupervised machine translation (UMT) to our task. We use segmentation-based padding and reinforcement learning to address under-translation and over-translation respectively. According to experiments, our approach significantly improve the perplexity and BLEU compared with typical UMT models. Furthermore, we explored guidelines on how to write the input vernacular to generate better poems. Human evaluation showed our approach can generate high-quality poems which are comparable to amateur poems.

Standard accuracy metrics indicate that modern reading comprehension systems have achieved strong performance in many question answering datasets. However, the extent these systems truly understand language remains unknown, and existing systems are not good at distinguishing distractor sentences, which look related but do not actually answer the question. To address this problem, we propose QAInfomax as a regularizer in reading comprehension systems by maximizing mutual information among passages, a question, and its answer. QAInfomax helps regularize the model to not simply learn the superficial correlation for answering questions. The experiments show that our proposed QAInfomax achieves the state-of-the-art performance on the benchmark Adversarial-SQuAD dataset.

Recently, biomedical version of embeddings obtained from language models such as BioELMo have shown state-of-the-art results for the textual inference task in the medical domain. In this paper, we explore how to incorporate structured domain knowledge, available in the form of a knowledge graph (UMLS), for the Medical NLI task. Specifically, we experiment with fusing embeddings obtained from knowledge graph with the state-of-the-art approaches for NLI task (ESIM model). We also experiment with fusing the domain-specific sentiment information for the task. Experiments conducted on MedNLI dataset clearly show that this strategy improves the baseline BioELMo architecture for the Medical NLI task.

The ongoing neural revolution in machine translation has made it easier to model larger contexts beyond the sentence-level, which can potentially help resolve some discourse-level ambiguities such as pronominal anaphora, thus enabling better translations. Unfortunately, even when the resulting improvements are seen as substantial by humans, they remain virtually unnoticed by traditional automatic evaluation measures like BLEU, as only a few words end up being affected. Thus, specialized evaluation measures are needed. With this aim in mind, we contribute an extensive, targeted dataset that can be used as a test suite for pronoun translation, covering multiple source languages and different pronoun errors drawn from real system translations, for English. We further propose an evaluation measure to differentiate good and bad pronoun translations. We also conduct a user study to report correlations with human judgments.

The rapid growth of artificial intelligence and automation presents threats -- and opportunities -- for workers and businesses in the Miami Valley. More than 31,600 people in the Dayton metro area work in the five largest occupations at high risk of automation, according to data the Brookings Institution prepared exclusively for the Dayton Daily News. Those jobs include food preparation, waiters, stock clerks, tractor-trailer truck drivers and accounting clerks. But about 34,600 people in the region that includes Montgomery, Greene and Miami counties work in the largest low-risk occupations. Those include registered nurses, freight and stock movers, janitors, customer service representatives and general managers, according to the Brookings data.

In recent weeks I managed to decrypt a difficult cipher that, despite expert codebreakers' best efforts, had remained unsolved for 70 years. The code was created by the late Cambridge professor and scientist Robert Henry Thouless, who passed away in 1984. He created it as a "test of survival" to see if he could communicate with the living after his death. Thouless thought if he successfully transmitted cipher keywords to the living through spiritual mediums and the message was received, this would prove he had survived his death. In 2019, I was more interested in seeing whether computer speed, storage and networking capabilities had advanced enough to break a code that had outlived its maker.

The ethos of the band has been to remain experimental and stir a conversation around creativity in the age of machines. With its latest album, Chain Tripping, recorded between the band's home in Los Angeles and Marfa, TX, they've pushed this conversation even further: into AI. Yacht is comprised of Claire L. Evans, Jona Bechtolt, and Rob Kieswetter, and the trio transformed and crafted its working method for Chain Tripping. The result is a 10-song pop album which falls in the intersection of DIY and high-tech. "At a certain point, you need something to push you out of your habits and make you excited, challenged and a little scared again," Evans says.

Designing reinforcement learning methods which find a good policy with as few samples as possible is a key goal of both empirical and theoretical research. On the theoretical side there are two main ways, regret- or PAC (probably approximately correct) bounds, to measure and guarantee sample-efficiency of a method. Ideally, we would like to have algorithms that have good performance according to both criteria, as they measure different aspects of sample efficiency and we have shown previously [1] that one cannot simply go from one to the other. In a specific setting called tabular episodic MDPs, a recent algorithm achieved close to optimal regret bounds [2] but there was no methods known to be close to optimal according to the PAC criterion despite a long line of research. In our work presented at ICML 2019, we close this gap with a new method that achieves minimax-optimal PAC (and regret) bounds which match the statistical worst-case lower bounds in the dominating terms.

Major powers are rushing to strengthen their militaries through artificial intelligence, but the US is hamstrung by certain challenges that rivals like China may not face, giving them an advantage in this strategic competition. Artificial intelligence and machine learning are enabling cutting-edge technological capabilities that have any number of possibilities, both in the civilian and military space. AI can mean complex data analysis and accelerated decision-making -- a big advantage that could potentially be the decisive difference in a high-end fight. For China, one of its most significant advantages -- outside of its disregard for privacy concerns and civil liberties that allow it to gather data and develop capabilities faster -- is the fusion of military aims with civilian commercial industry. In contrast, leading US tech companies like Google are not working with the US military on AI. "If we do not find a way to strengthen the bonds between the United States government and industry and academia, then I would say we do have the real risk of not moving as fast as China when it comes to" artificial intelligence, Lt. Gen. Jack Shanahan said, responding to Insider's queries at a Pentagon press briefing Friday.

Neuroscientists and clinicians around the world are using machine learning to develop treatment plans for patients and to identify some of the key markers for mental health disorders before they may set in. One of the benefits is that machine learning helps clinicians predict who may be at risk of a particular disorder. "Machine learning really meets a specific need that we have in psychiatry -- and that's the need for personalization," he says. "For decades, we've been working on group averages and statistics that apply to populations who may have the same diagnosis but don't translate as well to an individual patient. Machine learning allows us to get at individual predictions in a way we haven't been able to before.", says David Benrimoh, MD, CM, a psychiatry resident at McGill University.