We rerank with scores from pretrained masked language models like BERT to improve ASR and NMT performance. These log-pseudolikelihood scores (LPLs) can outperform large, autoregressive language models (GPT -2) in out-of-the-box scoring. RoBERTa reduces WER by up to 30% relative on an end-to-end LibriSpeech system and adds up to 1.7 BLEU on state-of-the-art baselines for TED Talks low-resource pairs, with further gains from domain adaptation. In the multilingual setting, a single XLM can be used to rerank translation outputs in multiple languages. The numerical and qualitative properties of LPL scores suggest that LPLs capture sentence fluency better than autoregressive scores. Finally, we finetune BERT to estimate sentence LPLs without masking, enabling scoring in a single, non-recurrent inference pass.

Continual learning aims to improve the ability of modern learning systems to deal with non-stationary distributions, typically by attempting to learn a series of tasks sequentially. Prior art in the field has largely considered supervised or reinforcement learning tasks, and often assumes full knowledge of task labels and boundaries. In this work, we propose an approach (CURL) to tackle a more general problem that we will refer to as unsupervised continual learning. The focus is on learning representations without any knowledge about task identity, and we explore scenarios when there are abrupt changes between tasks, smooth transitions from one task to another, or even when the data is shuffled. The proposed approach performs task inference directly within the model, is able to dynamically expand to capture new concepts over its lifetime, and incorporates additional rehearsal-based techniques to deal with catastrophic forgetting. We demonstrate the efficacy of CURL in an unsupervised learning setting with MNIST and Omniglot, where the lack of labels ensures no information is leaked about the task. Further, we demonstrate strong performance compared to prior art in an i.i.d setting, or when adapting the technique to supervised tasks such as incremental class learning.

Personality and aptitude tests (psychometrics) are key to managing people and also for better understanding yourself. For students and learners it is important to take personality tests at regular intervals to gain insights into individual traits and aptitudes.. This...

The Payment Intelligence group is responsible for optimizing each of the billions of dollars of transactions processed by Stripe each year on behalf of our users, in order to maximize successful transactions while minimizing payment costs and fraud. We own products like Radar from end to end and work across the technical stack: from machine learning over our users' data, to integrating ML intelligence and serving real-time predictions as part of Stripe's payment infrastructure, to building user-facing product surfaces like dashboards and controls. Your team has significant responsibility over the funds flowing through Stripe, and the ability to make or break Stripe's future. Stripe is helping the internet fulfill its potential as a platform for economic progress by building software tools that accelerate global economic access and technological development. Stripe makes it easy to start, run and scale an internet business from anywhere in the world.

In partnership with PBS KIDS, this year's competition will look at advancements in early childhood education. The results will lead to better designed games and improved learning outcomes, empowering children, parents, caregivers, and educators across the globe with insights into how young children learn through media and which approaches work best to help them build on foundational learning skills. The 90-day Data Science Bowl competition will award winning participants with a share of $160,000 in cash prizes. Research shows much of the most critical brain development in children takes place before they even reach kindergarten. Child development experts indicate it is during these first 5 years that children develop linguistic, cognitive, social, emotional, and regulatory skills that predict their later functioning in many domains.

Abstract: Academics and practitioners alike believe that continual learning (CL) is a fundamental step towards artificial intelligence. Continual learning is the ability of a model to learn continually from a stream of data. In practice, this means supporting the ability of a model to autonomously learn and adapt in production as new data comes in. The idea of CL is to mimic humans ability to continually acquire, fine-tune, and transfer knowledge and skills throughout their lifespan. CL of models in production will improve accuracy, and bring artificial intelligence one step closer to real human intelligence.

We are already experiencing how AI is affecting some facets of our lives. The movement sensing lights that go on when you enter an unoccupied conference room or a parking area, intelligent sensors in cameras now that detect a smile and help take good pictures, cars with automatic parking features, Alexa switching on the living room lights, and Google adding items to your shopping list! We have personal assistants in our smartphones that we have rather got used to. The buzz around AI in eLearning has also acquired lot of weight in the last few years. Needless to say, AI is the primary topic of interest at most learning and technology conferences these days.

For many years, e-learning designers have relied on chatbots -- those handy little digital "helpers" that pop up to answer questions or provide additional information. With the introduction of faster internet connections (5G and beyond) and better development tools, chatbots are no longer just digitized voices behind animated avatars. Chatbots are rules-driven services, sometimes powered by artificial intelligence (AI), that help individuals communicate in an online environment. In an e-learning context, AI-powered chatbots make learning more intuitive by helping learners choose, consume and understand content. Rather than clicking buttons, selecting drop-down menu items or tapping on a screen, with chatbots, learners can navigate content through gestures and conversational interactions.

Big things are happening in tech with AI and deep learning. That's not exactly a news flash when you look at how often companies use algorithms to …

If you live and breathe all things robotics you might want to consider reserving your place in any one of these great events coming up in 2020. We have scoured the internet for some of the best robotics events you might want to consider checking out next year. VEX Worlds 2019 is a robotics competition that was run between the 24th and 27th of October 2019. It was held at the Kentucky Exposition Center, 937 Phillips Lane, Louisville in Kentucky. The VEX IQ Challenge is a fun competition presented by the Robotics Education and Competition Foundation.