Reproducibility in reinforcement learning is challenging: uncontrolled stochasticity from many sources, such as the learning algorithm, the learned policy, and the environment itself have led researchers to report the performance of learned agents using aggregate metrics of performance over multiple random seeds for a single environment. Unfortunately, there are still pernicious sources of variability in reinforcement learning agents that make reporting common summary statistics an unsound metric for performance. Our experiments demonstrate the variability of common agents used in the popular OpenAI Baselines repository. We make the case for reporting post-training agent performance as a distribution, rather than a point estimate.

The infeasible parts of the objective space in difficult many-objective optimization problems cause trouble for evolutionary algorithms. This paper proposes a reference vector based algorithm which uses two interacting engines to adapt the reference vectors and to evolve the population towards the true Pareto Front (PF) s.t. the reference vectors are always evenly distributed within the current PF to provide appropriate guidance for selection. The current PF is tracked by maintaining an archive of undominated individuals, and adaptation of reference vectors is conducted with the help of another archive that contains layers of reference vectors corresponding to different density. Experimental results show the expected characteristics and competitive performance of the proposed algorithm TEEA.

Automatically reasoning about future human behaviors is a difficult problem with significant practical applications to assistive systems. Part of this difficulty stems from learning systems' inability to represent all kinds of behaviors. Some behaviors, such as motion, are best described with continuous representations, whereas others, such as picking up a cup, are best described with discrete representations. Furthermore, human behavior is generally not fixed: people can change their habits and routines. This suggests these systems must be able to learn and adapt continuously. In this work, we develop an efficient deep generative model to jointly forecast a person's future discrete actions and continuous motions. On a large-scale egocentric dataset, EPIC-KITCHENS, we observe our method generates high-quality and diverse samples while exhibiting better generalization than related generative models. Finally, we propose a variant to continually learn our model from streaming data, observe its practical effectiveness, and theoretically justify its learning efficiency.

Social media offer an abundant source of valuable raw data, however informal writing can quickly become a bottleneck for many natural language processing (NLP) tasks. Off-the-shelf tools are usually trained on formal text and cannot explicitly handle noise found in short online posts. Moreover, the variety of frequently occurring linguistic variations presents several challenges, even for humans who might not be able to comprehend the meaning of such posts, especially when they contain slang and abbreviations. Text Normalization aims to transform online user-generated text to a canonical form. Current text normalization systems rely on string or phonetic similarity and classification models that work on a local fashion. We argue that processing contextual information is crucial for this task and introduce a social media text normalization hybrid word-character attention-based encoder-decoder model that can serve as a pre-processing step for NLP applications to adapt to noisy text in social media. Our character-based component is trained on synthetic adversarial examples that are designed to capture errors commonly found in online user-generated text. Experiments show that our model surpasses neural architectures designed for text normalization and achieves comparable performance with state-of-the-art related work.

In order to drive safely and efficiently under merging scenarios, autonomous vehicles should be aware of their surroundings and make decisions by interacting with other road participants. Moreover, different strategies should be made when the autonomous vehicle is interacting with drivers having different level of cooperativeness. Whether the vehicle is on the merge-lane or main-lane will also influence the driving maneuvers since drivers will behave differently when they have the right-of-way than otherwise. Many traditional methods have been proposed to solve decision making problems under merging scenarios. However, these works either are incapable of modeling complicated interactions or require implementing hand-designed rules which cannot properly handle the uncertainties in real-world scenarios. In this paper, we proposed an interaction-aware decision making with adaptive strategies (IDAS) approach that can let the autonomous vehicle negotiate the road with other drivers by leveraging their cooperativeness under merging scenarios. A single policy is learned under the multi-agent reinforcement learning (MARL) setting via the curriculum learning strategy, which enables the agent to automatically infer other drivers' various behaviors and make decisions strategically. A masking mechanism is also proposed to prevent the agent from exploring states that violate common sense of human judgment and increase the learning efficiency. An exemplar merging scenario was used to implement and examine the proposed method.

Walt Disney World recently showed the Associated Press what it takes to put their shows together. It's a shift for a resort that hasn't allowed many peeks behind the curtains of the fantasy it creates. Maybe romantic Disney fairy tales come true after all. Data from the popular global online dating site Plenty of Fish reveals that singles who have expressed an interest in Disney are 3.6 times more likely to leave the app in a relationship compared to singles who more generally list interests in music and movies. That was certainly true for Disney fan Abby Schiller.

Amazon's Echo speakers have a broadcast feature that will help you send a message to family members that might be scattered around the house. If you have an Amazon Echo product, you aren't the only person privy to your private conversations. Thousands of people across the globe are employed by Amazon.com to listen to Echo recordings, transcribe and annotate them and feed them back to the software so that Alexa can better grasp human speech, according to a report from Bloomberg. The employees – ranging from Boston to India – signed nondisclosure agreements barring them to speak publicly about the program. According to Bloomberg, they work nine hours per day, with each reviewer going through as many as 1,000 audio clips per shift.

'I donate to the Guardian, so I'm paying you." So begins Chet Faliszek as we sit down to lunch in one of the San Francisco hotels that satellite around the Game Developers Conference. One of the industry's most respected comedy writers and lead developers, the 53-year-old is here to recruit developers to his new studio Stray Bombay, named after his pet cat Boris. With Riot Games veteran and AI expert Dr Kimberly Voll, he is leading a studio that will focus on smart cooperative video games, made for (they say) smart cooperative players. It quickly becomes clear just how much cooperation has been a vital part of Faliszek's life, from pivotal relationships growing up in Parma, Cleveland, to a comedy writing double-act at infamous early-internet website Old Man Murray, to his run of successful collaborations at a behemoth developer, Valve. With every key moment in his life, he cites the generosity of another person, a pattern which appears to have informed his entire approach to games development, and the sorts of games he wants to make. At 17, in the early 80s, Faliszek had dropped out of a computer-science college course. "I was taking a course in [programming language] Fortran," he explains, "and one time I tripped and dropped my punchcards.

When Amazon customers speak to Alexa, the company's AI-powered voice assistant, they may be heard by more people than they expect, according to a report. Amazon employees around the world regularly listen to recordings from the company's smart speakers as part of the development process for new services, Bloomberg News reports. Some transcribe artist names, linking them to specific musicians in the company's database; others listen to the entire recorded command, comparing it with what the automated systems heard and the response they offered, in order to check the quality of the company's software. Technically, users have given permission for the human verification: the company makes clear that it uses data "to train our speech recognition and natural language understanding systems", and gives users the chance to opt out. But the company doesn't explicitly say that the training will involve workers in America, India, Costa Rica, and more nations around the world listening to those recordings.

Golf fans who are planning to watch the Masters this weekend will have yet more ways to check out the action. For the first time at a golf tournament, practically every one of the more than 20,000 shots from the first major of the year will be available to view on the Masters website and app within five minutes of a player striking the ball. While these videos won't be live, you'll essentially be able to watch full rounds from the likes of Tiger Woods, Rory McIlroy and Jordan Speith without such trivial matters as watching them walk between shots. There is a caveat in that cameras might not capture shots in some instances, such as those from unusual lies, or if a group's tee shots end up in wildly different spots. The Masters attracts sports aficionados who might not typically watch golf as well as devotees, so it's a high-profile way to debut this technology after a few years of development. It should be especially useful over the first two days when the field is at its most expansive, and a player might be unexpectedly putting together a killer round and rampaging up the leaderboard when they aren't a focus of the TV broadcast.