Deep reinforcement learning has been successfully applied to several visual-input tasks using model-free methods. In this paper, we propose a model-based approach that combines learning a DNN-based transition model with Monte Carlo tree search to solve a block-placing task in Minecraft. Our learned transition model predicts the next frame and the rewards one step ahead given the last four frames of the agent's first-person-view image and the current action. Then a Monte Carlo tree search algorithm uses this model to plan the best sequence of actions for the agent to perform. On the proposed task in Minecraft, our model-based approach reaches the performance comparable to the Deep Q-Network's, but learns faster and, thus, is more training sample efficient. Keywords: Acknowledgements Reinforcement Learning, Model-Based Reinforcement Learning, Deep Learning, Model Learning, Monte Carlo Tree Search I would like to express my sincere gratitude to my supervisor Dr. Stefan Uhlich for his continuous support, patience, and immense knowledge that helped me a lot during this study. My thanks and appreciation also go to my colleague Anna Konobelkina for insightful comments on the paper as well as to Sony Europe Limited for providing the resources for this project.

Motivated by Supervised Opinion Analysis, we propose a novel framework devoted to Structured Output Learning with Abstention (SOLA). The structure prediction model is able to abstain from predicting some labels in the structured output at a cost chosen by the user in a flexible way. For that purpose, we decompose the problem into the learning of a pair of predictors, one devoted to structured abstention and the other, to structured output prediction. To compare fully labeled training data with predictions potentially containing abstentions, we define a wide class of asymmetric abstention-aware losses. Learning is achieved by surrogate regression in an appropriate feature space while prediction with abstention is performed by solving a new pre-image problem. Thus, SOLA extends recent ideas about Structured Output Prediction via surrogate problems and calibration theory and enjoys statistical guarantees on the resulting excess risk. Instantiated on a hierarchical abstention-aware loss, SOLA is shown to be relevant for fine-grained opinion mining and gives state-of-the-art results on this task. Moreover, the abstention-aware representations can be used to competitively predict user-review ratings based on a sentence-level opinion predictor.

Training neural networks involves finding minima of a high-dimensional non-convex loss function. Knowledge of the structure of this energy landscape is sparse. Relaxing from linear interpolations, we construct continuous paths between minima of recent neural network architectures on CIFAR10 and CIFAR100. Surprisingly, the paths are essentially flat in both the training and test landscapes. This implies that neural networks have enough capacity for structural changes, or that these changes are small between minima. Also, each minimum has at least one vanishing Hessian eigenvalue in addition to those resulting from trivial invariance.

The finding that very large networks can be trained efficiently and reliably has led to a paradigm shift in computer vision from engineered solutions to learning formulations. As a result, the research challenge shifts from devising algorithms to creating suitable and abundant training data for supervised learning. How to efficiently create such training data? The dominant data acquisition method in visual recognition is based on web data and manual annotation. Yet, for many computer vision problems, such as stereo or optical flow estimation, this approach is not feasible because humans cannot manually enter a pixel-accurate flow field. In this paper, we promote the use of synthetically generated data for the purpose of training deep networks on such tasks.We suggest multiple ways to generate such data and evaluate the influence of dataset properties on the performance and generalization properties of the resulting networks. We also demonstrate the benefit of learning schedules that use different types of data at selected stages of the training process.

Deep learning, a field in machine learning relying on learning data representations as opposed to task-specific algorithms, is experiencing a great and fairly sudden storm of enthusiasm and excitement after over twenty years of relative quietness. Delivering big technology breakthroughs recently, it has been regarded as a major driver towards artificial intelligence by many observers. The AI industry has spurred countless investments and acquisitions over the past years. In 2017, AI startups in the UK raised £488m, according to Pitchbook, more than twice as much as in the previous year. Furthermore, the country hosted four of the biggest acquisitions of AI startups over the past five years, including Google/DeepMind, Apple/VocalIQ, Microsoft/SwiftKey and Twitter/Magic Pony. Very generally speaking, deep learning models are based on information processing and communication patterns in biological nervous systems and represent coding attempts to define relationships between certain stimuli and associated responses in the brain in a mathematically convenient way.

Mark Zuckerberg on Wednesday issued a statement on the growing controversy around Cambridge Analytica's acquisition and use of tens of millions of people's personal Facebook data. In the 935-word statement, Zuckerberg reassures users that "the good news is that most important actions to prevent this from happening again" were already taken in 2014, when the company limited the amount of data that could be acquired by third-party apps on the social media platform. While the statement acknowledged the company "made mistakes," it avoided an explicit apology or the word "sorry." Zuckerberg's move comes four days after the New York Times reported that Cambridge Analytica, a company that provides political operators detailed information on millions of voters, obtained data on more than 50 million American Facebook users from a University of Cambridge researcher named Aleksandr Kogan. Cambridge Analytica's connections to Republican megadonor Robert Mercer and Steve Bannon, an ex Trump campaign chairman and a senior White House adviser, may have allowed the Trump campaign to access and use the data to target potential voters, according to the Times.

UiPath is creating a strong ecosystem of partners to help deliver increasingly efficient enterprise RPA deployment by ensuring that a rich marketplace of leading partner technologies can offer a powerful, pre-integrated, value-added function to the already highly successful UiPath solutions in place. "Conversational Natural Language engagement is a huge growth area for both personal and business use across the globe," said Boris Krumrey, Chief Robotics Officer of UiPath. "Speech and text driven access to initiate and operate robotic process automations is a new channel of user interaction to RPA using conversation. Humley has demonstrated how a virtual assistant can be deployed to support RPA management and build Voice or Chat User Interfaces which will help customers innovate in regards to the'how and where' they access robotic process automations. What we like about Humley is that it works with other leading technologies such as IBM Watson and Microsoft Cognitive Services."

People look back at an autonomous self-driving vehicle, as it is tested in a pedestrian zone in Milton Keynes, north of London, on October 11, 2016. Suburbs have largely been dismissed by environmentalists and urban planners as bad for the planet, a form that needed to be eliminated to make way for a bright urban future. Yet, after a few years of demographic stultification amid the Great Recession, Americans are again heading to the suburbs in large numbers, particularly millennials. So rather than fight the tide and treat suburbanization as an evil to be squeezed out, perhaps a better approach would be to modify the suburban form in ways that address its most glaring environmental weakness: dependence on gas-powered automobiles. The rise of ride-sharing, electric cars and ultimately the self-driving automobile seem likely to alter this paradigm.

Willis Towers Watson's The Global Future of Work Survey approached 909 companies worldwide, and found that they expect automation will account for 22% of the work being done in the next three years. That compares with the 12% of work companies say is being done using artificial intelligence (AI) and robotics today, and just 7% three years ago. The research suggested that HR is not currently fully prepared for potential organisational changes brought on by technology. Just 31% of organisations have taken steps to address talent deficits, only 32% have attempted to identify the emerging skills required for their business, while 29% say that they have tried to find appropriate talent for a digital workforce. While few employers have measures in place, respondents said they were planning to take action in the future through establishing which tasks can be automated (50%), and identifying ways to reskill talent whose work could be automated (48%).

Yesterday in Berlin, I attended a workshop on the use of artificial intelligence in governing communication online, hosted by the Humboldt Institute for Internet and Society. Section 230 of the CDA provides broad immunity to platforms, with the express goals of promoting economic development and free expression. Daphne Keller has a good summary of the legal landscape on intermediary liability. Platforms are now facing increasing pressure to detect and remove illegal (and, in some cases, legal-but-objectionable) content. In the United States, for example, bills in the House and Senate would remove safe harbor protection for platforms that do not remove illegal content related to sex trafficking.