Collaborative decision making in multi-agent systems typically requires a predefined communication protocol among agents. Usually, agent-level observations are locally processed and information is exchanged using the predefined protocol, enabling the team to perform more efficiently than each agent operating in isolation. In this work, we consider the situation where agents, with complementary sensing modalities must co-operate to achieve a common goal/task by learning an efficient communication protocol. We frame the problem within an actor-critic scheme, where the agents learn optimal policies in a centralized fashion, while taking action in a distributed manner. We provide an interpretation of the emergent communication between the agents. We observe that the information exchanged is not just an encoding of the raw sensor data but is, rather, a specific set of directive actions that depend on the overall task. Simulation results demonstrate the interpretability of the learnt communication in a variety of tasks.

Rational humans can generate sentences that cover a certain set of concepts while describing natural and common scenes. For example, given {apple(noun), tree(noun), pick(verb)}, humans can easily come up with scenes like "a boy is picking an apple from a tree" via their generative commonsense reasoning ability. However, we find this capacity has not been well learned by machines. Most prior works in machine commonsense focus on discriminative reasoning tasks with a multi-choice question answering setting. Herein, we present CommonGen: a challenging dataset for testing generative commonsense reasoning with a constrained text generation task. We collect 37k concept-sets as inputs and 90k human-written sentences as associated outputs. Additionally, we also provide high-quality rationales behind the reasoning process for the development and test sets from the human annotators. We demonstrate the difficulty of the task by examining a wide range of sequence generation methods with both automatic metrics and human evaluation. The state-of-the-art pre-trained generation model, UniLM, is still far from human performance in this task. Our data and code is publicly available at http://inklab.usc.edu/CommonGen/ .

An important class of decidable first-order logic fragments are those satisfying a guardedness condition, such as the guarded fragment (GF). Usually, decidability for these logics is closely linked to the tree-like model property - the fact that satisfying models can be taken to have tree-like form. Decision procedures for the guarded fragment based on the tree-like model property are difficult to implement. An alternative approach, based on restricting first-order resolution, has been proposed, and this shows more promise from the point of view of implementation. In this work, we connect the tree-like model property of the guarded fragment with the resolution-based approach. We derive efficient resolution-based rewriting algorithms that solve the Quantifier-Free Query Answering Problem under Guarded Tuple Generating Dependencies (GTGDs) and Disjunctive Guarded Tuple Generating Dependencies (DisGTGDs). The Query Answering Problem for these classes subsumes many cases of GF satisfiability. Our algorithms, in addition to making the connection to the tree-like model property clear, give a natural account of the selection and ordering strategies used by resolution procedures for the guarded fragment. We also believe that our rewriting algorithm for the special case of GTGDs may prove itself valuable in practice as it does not require any Skolemisation step and its theoretical runtime outperforms those of known GF resolution procedures in case of fixed dependencies. Moreover, we show a novel normalisation procedure for the widely used chase procedure in case of (disjunctive) GTGDs, which could be useful for future studies.

Every natural text is written in some style. The style is formed by a complex combination of different stylistic factors, including formality markers, emotions, metaphors, etc. Some factors implicitly reflect the author's personality, while others are explicitly controlled by the author's choices in order to achieve some personal or social goal. One cannot form a complete understanding of a text and its author without considering these factors. The factors combine and co-vary in complex ways to form styles. Studying the nature of the covarying combinations sheds light on stylistic language in general, sometimes called cross-style language understanding. This paper provides a benchmark corpus (xSLUE) with an online platform (http://xslue.com) for cross-style language understanding and evaluation. The benchmark contains text in 15 different styles and 23 classification tasks. For each task, we provide the fine-tuned classifier for further analysis. Our analysis shows that some styles are highly dependent on each other (e.g., impoliteness and offense), and some domains (e.g., tweets, political debates) are stylistically more diverse than others (e.g., academic manuscripts). We discuss the technical challenges of cross-style understanding and potential directions for future research: cross-style modeling which shares the internal representation for low-resource or low-performance styles and other applications such as cross-style generation.

Recent advances in deep reinforcement learning have demonstrated the capability of learning complex control policies from many types of environments. When learning policies for safety-critical applications, it is essential to be sensitive to risks and avoid catastrophic events. Towards this goal, we propose an actor-critic framework that models the uncertainty of the future and simultaneously learns a policy based on that uncertainty model. Specifically, given a distribution of the future return for any state and action, we optimize policies for varying levels of conditional Value-at-Risk. The learned policy can map the same state to different actions depending on the propensity for risk. We demonstrate the effectiveness of our approach in the domain of driving simulations, where we learn maneuvers in two scenarios. Our learned controller can dynamically select actions along a continuous axis, where safe and conservative behaviors are found at one end while riskier behaviors are found at the other. Finally, when testing with very different simulation parameters, our risk-averse policies generalize significantly better compared to other reinforcement learning approaches.

In this workshop, you will get a glimpse into how we can teach machines to analyze complex scenarios at a much larger scale than we're able to. After you've cleaned and organized your data, you will have an opportunity to train and test machine learning models, and even publish your predictor online for others to explore. New parts are made and shipped from factories, people continuously tweet, and companies grow and fluctuate causing major changes in the market. With the addition of more data comes the difficulty of being able to process that data. As humans, we can understand complex scenarios, but computers are much better at being able to analyze large datasets.

Microsoft officially entered the robotic process automation (RPA) marketplace this week with some major changes to its Power Platform. It's not the first incumbent enterprise software vendors to feel the need to have a product in this category (SAP purchased French RPA vendor Contextor late last year), and it won't be the last. We should expect to see a steady increase in RPA investment in from big enterprise vendors, with a mix of internally developed and acquired technologies. The fundamental challenge driving these investments is that RPA is the first step in a journey to reinvent how companies build the software they use to run their businesses (see my recent article on how UIPath is reinventing the RPA category). RPA lets companies record a series of computer-based processes done by a human so that the series can then be repeated automatically without human involvement.

Physicians may soon be able to get a second opinion on your diagnosis from a form of homegrown artificial intelligence (AI). Toronto's University Health Network (UHN) is teaming up with researchers at the University of Waterloo and the Vector Institute to develop software that can read and provide feedback on medical images like x-rays and ultrasounds. The groundbreaking technology is specifically aimed at assisting radiologists and pathologists. "We are trying to basically go after the biggest problems of medical imaging, which we call user variability," said Dr. Hamid Tizhoosh, director of the KIMIA lab at the University of Waterloo. "Different radiologists may have a different diagnosis looking at the same image – and it seems that AI can assist in removing that variability."

Asker, Norway-based Tomra Sorting Recycling has launched Gain, a deep learning-based sorting technology to further enhance the performance of its sensor-based sorting machines, according to a Tomra news release. The gain technology will be made available as an add-on option for the company's autosort machines. By classifying objects from sensor data, gain enables the sorting of objects, which could previously not be seperated with high levels of purity and without compromising the throughput speed of the autosort. "By bringing deep learning to our sorting technologies, Tomra is adding further sophistication and effectiveness to its market-leading autosort sorting machines," says Alessandro Granziera, sales manager for Tomra Sorting Recycling in Italy. "The gain technology will also help sorting machines adapt to new waste streams, which will be increasingly important as we move towards a circular economy."

Earlier this month, the Stamford, Conn.-based research organization ISG published its 2019 ISG Provider Lens Social Business Collaboration (subscription required) report. Like all such reports, it makes for interesting reading and focuses on the vendors that are developing quickly in this space. It showed that Workplace by Facebook, for example, has gained significant traction since its initial launch almost three years ago with the familiarity and popularity of the Facebook user interface making Workplace deployments faster, as employees require little or no training. Igloo Software was also named a leader for its "modern outlook toward the traditional intranet," the report noted. Other leaders in ISG's Enterprise Social Collaboration Solutions quadrant included Microsoft and Slack.