Transfer learning, where a model is first pre-trained on a data-rich task before being fine-tuned on a downstream task, has emerged as a powerful technique in natural language processing (NLP). The effectiveness of transfer learning has given rise to a diversity of approaches, methodology, and practice. In this paper, we explore the landscape of transfer learning techniques for NLP by introducing a unified framework that converts every language problem into a text-to-text format. Our systematic study compares pre-training objectives, architectures, unlabeled datasets, transfer approaches, and other factors on dozens of language understanding tasks. By combining the insights from our exploration with scale and our new "Colossal Clean Crawled Corpus", we achieve state-of-the-art results on many benchmarks covering summarization, question answering, text classification, and more. To facilitate future work on transfer learning for NLP, we release our dataset, pre-trained models, and code.

With the aim of building machine learning systems that incorporate standards of fairness and accountability, we explore explicit subgroup sample complexity bounds. The work is motivated by the observation that classifier predictions for real world datasets often demonstrate drastically different metrics, such as accuracy, when subdivided by specific sensitive variable subgroups. The reasons for these discrepancies are varied and not limited to the influence of mitigating variables, institutional bias, underlying population distributions as well as sampling bias. Among the numerous definitions of fairness that exist, we argue that at a minimum, principled ML practices should ensure that classification predictions are able to mirror the underlying sub-population distributions. However, as the number of sensitive variables increase, populations meeting at the intersectionality of these variables may simply not exist or may not be large enough to provide accurate samples for classification. In these increasingly likely scenarios, we make the case for human intervention and applying situational and individual definitions of fairness. In this paper we present lower bounds of subgroup sample complexity for metric-fair learning based on the theory of Probably Approximately Metric Fair Learning. We demonstrate that for a classifier to approach a definition of fairness in terms of specific sensitive variables, adequate subgroup population samples need to exist and the model dimensionality has to be aligned with subgroup population distributions. In cases where this is not feasible, we propose an approach using individual fairness definitions for achieving alignment. We look at two commonly explored UCI datasets under this lens and suggest human interventions for data collection for specific subgroups to achieve approximate individual fairness for linear hypotheses.

The Australian Security Intelligence Organisation (ASIO) has a problem, it collects too much data and might miss something. "That's the problem we are dealing with right now, given the threats are at the unprecedented level," recently installed Director-General of Security Mike Burgess said during his 38th day on the job. "There is the potential to miss something, the application of data analytics helps us to reduce the possibility of that being an event." ASIO is currently undertaking an enterprise-wide transformation that it believes will place it "at the forefront of agencies" using artificial intelligence and machine learning, according to its recent annual report. Providing an update on the project, Burgess said the organisation has so far put a new operating structure and model in place, as well as other foundational work subject to further government approvals.

The Defense Department's artificial intelligence strategy, released in February, calls for the use of standardized processes in areas such as data, testing and evaluation, and cybersecurity. Now, the DOD is starting to make that a reality. The Pentagon's Joint Artificial Intelligence Center plans to work with the National Security Agency, U.S. Cyber Command and numerous DOD cybersecurity vendors to standardize data collection across the department, JAIC chief Lt. Gen. Jack Shanahan said earlier this month, as Nextgov reports. Speaking earlier this month at the Billington CyberSecurity Summit in Washington, D.C., Shanahan discussed how the DOD wants to create a consistent way to curate, share and store cybersecurity data from across the Pentagon's entire IT environment. Doing so will make it to easier to deploy AI-powered cybersecurity programs, he said.

The most valuable weapon against the next deadly disease outbreak may be data. Scientists aiming to stop or prevent the spread of viral or bacterial pathogens need rapid, comprehensive access to datasets on their genomics, structure, function and more, combined with computational tools to quickly analyze data and make predictions using artificial intelligence techniques. That critical service will be provided by the new Bacterial and Viral Bioinformatics Resource Center (BV-BRC), based at the University of Chicago. The big data resource combines two independent efforts at UChicago and the J. Craig Venter Institute (JCVI) into a common infrastructure that will support richer scientific data and more powerful analytic tools. The data resource will be funded by the National Institutes of Health and the contract award totals $43,196,984 over five years, if all options are exercised. The primary awardee is the University of Chicago, with sub-awards given to JCVI, the University of Virginia and the Fellowship for Interpretation of Genomes.

Olis Robotics, a leader in next-generation AI-driven software for remote robotics in dynamic environments in subsea, terrestrial, and space applications, today announced that it has been selected by Maxar Technologies to provide robotic operator planning software for Maxar's Sample Acquisition, Morphology Filtering, and Probing of Lunar Regolith (SAMPLR) robotic arm. The arm will be mounted to a yet-to-be-named lander as one of 12 payloads that NASA selected as part of its Artemis program to send the first woman and the next man to the Moon by 2024 in preparation for a human mission to Mars. Olis Robotics' operator planning software will solve for the extreme latency experienced while operating robotics on the lunar surface by enabling operators to simulate and plan movements from the ground. Olis' software will provide a 3D visualization of the lunar environment and intuitive controls for operators on Earth, providing enhanced control during exploration missions. "The moon provides an excellent proving ground for our robotic operator planning software, allowing operators on Earth to successfully complete more complex missions faster and safer than ever before," explained Olis Robotics CEO Don Pickering.

Raytheon has delivered an experimental new anti-drone weapon to the Air Force. The High Energy Laser Weapon Systems (HELWS) prototypes will be put through a year of testing and training by Air Force personnel overseas before finally being ready for live use on the battlefield. The HELWS can be powered by either a standard 220-volt outlet or a generator. Operators that aren't great shots can take comfort in the fact that the energy efficient device can fire'a nearly infinite number of shots.' The laser also comes with a sophisticated targeting system, with an infrared sensor to track and identify enemy drones, according to a report from Gizmodo.

Your resume may not be the only deciding factor in landing your next job – it could be an'employability score' created by artificial intelligence that has the final vote. More than 100 big name firms are using HireVue's AI-driven assessment, which is technology that ranks candidates based on their facial movements, choice of words and speaking voice. Although employers can pursue any candidate, some have told The Washington Post that they usually focus on those the computer system liked best -- leading some experts to question how bias the process may be. More than 100 employers are using HireVue's AI-driven assessment that ranks candidates based on their facial movements, choice of words and speaking voice HireVue's technology is employed by many large name companies such as Hilton Hotels, Unilever and Goldman Sachs, according to The Washington Post. And with hundreds of applications flooding in for just a single position, the AI has made it easy for human employers to find the perfect candidate what i- but some experts believe the technology can do more harm than good.

In a low-volume distributed denial-of-service (LVDDoS) attack, an adversary attempts to overwhelm the server by making requests specially crafted to use an inordinate amount of the server's resources. The imbalance between the resources used by the server and attacker during an LVDDoS attack allows otherwise resource-constrained adversaries to mount effective attacks on large systems. Standard defense tools focus on metrics such as the number of requests and don't focus on nuanced metrics such as user experience. We propose Canopy, a novel approach for detecting LVDDoS attacks by applying machine learning techniques to extract meaning from observed patterns of TCP state transitions. We differentiate between malicious and benign traffic by employing a supervised learning approach, using features extracted from the temporal patterns of TCP state transitions.

Today, Microsoft is releasing The Future Computed: AI and Manufacturing. This new book provides an in-depth look at how artificial intelligence (AI) is transforming the manufacturing sector by optimizing digital operations and driving efficiencies, enabling new products and services, and allowing for safer work environments. The book also offers a timely look at how society can respond to some of the challenges AI creates, and the need to develop new laws and regulations to address workforce disruption and develop AI in an ethical and responsible manner. Written by Greg Shaw – co-author of Hit Refresh with Microsoft CEO Satya Nadella and The Ability Hacks about technology for people with disabilities – this book is the second in a series that began with the release last year of The Future Computed: Artificial Intelligence and its role in society. In an era when digital technology is changing almost every aspect about how people live, work, play, and learn, we believe it is important to think carefully about the complex questions that AI raises.