We wish to express our appreciation to the reviewers for their insightful comments on our paper. All responses are reflected in our camera-ready version. Thank you for the proposal. We are sorry for that our writing makes itself hard to follow. Thank you for the important comment.

We wish to express our appreciation to the reviewers for their insightful comments on our paper. All responses are reflected in our camera-ready version. Thank you for the proposal. We are sorry for that our writing makes itself hard to follow. Thank you for the important comment.

Data for pretraining machine learning models often consists of collections of heterogeneous datasets. Although training on their union is reasonable in agnostic settings, it might be suboptimal when the target domain -- where the model will ultimately be used -- is known in advance. In that case, one would ideally pretrain only on the dataset(s) most similar to the target one. Instead of limiting this choice to those datasets already present in the pretraining collection, here we explore extending this search to all datasets that can be synthesized as `combinations' of them. We define such combinations as multi-dataset interpolations, formalized through the notion of generalized geodesics from optimal transport (OT) theory. We compute these geodesics using a recent notion of distance between labeled datasets, and derive alternative interpolation schemes based on it: using either barycentric projections or optimal transport maps, the latter computed using recent neural OT methods. These methods are scalable, efficient, and -- notably -- can be used to interpolate even between datasets with distinct and unrelated label sets. Through various experiments in transfer learning in computer vision, we demonstrate this is a promising new approach for targeted on-demand dataset synthesis.

Face biometrics are now firmly established as a way for people to unlock their mobile phones, or sign up to a new online account. As a core means of identifying a person, however, former UNDP Policy Advisor and Program Manager for Legal Identity Niall McCann thinks facial recognition may be on its way out. Biometrics are often part of the registration process, linking a person to their ID number, and ID documents may encode the individual's biometrics, number, or both. McCann tells Biometric Update's Frank Hersey in episode two that because facial recognition can be carried out without the consent or knowledge of the subject, unlike fingerprint biometrics, it is likely to be restricted by the UN for ID projects in the coming years. "You don't know when a CCTV camera system based on street corners is identifying you via facial recognition means," McCann explains.

Current chatbots tend to be awkward and even agonizing to use, limited to answering a set of simple queries -- often incorrectly. But what if a chatbot could be designed to support more complex and multistep tasks, such as organizing a daily schedule or pinpointing a fault lurking inside a complex mechanical device? Conversational AI technology promises all of this and more. Conversational AI is built on natural language processing (NLP) and other machine learning (ML) technologies, with the goal of enabling human-like interactions between machines and people. So far, conversational AI has mostly been used to create sophisticated chatbots -- as opposed to scripted, rules-based chatbots.

Artificial intelligence is not the scary, half-human half-robot movie character some might think it is, says Precision AI founder Daniel McCann. "AI is just a data processing system that points out patterns in huge volumes of data and that's it," he said during a presentation at the virtual Canada's Farm Show. For agriculture, it represents the future. McCann said he believes within the next 15 years common pieces of farm equipment, such as the broadcast sprayer, will become like a BlackBerry -- still around, not too common and not too efficient. But he said AI in agriculture is at an interim step along the way to that.

Boost My Business shines a spotlight on a tech start-up whose app maps accessibility via its users and AI. Fiona Alston chatted to Matthew McCann CEO of Access Earth about how his app helps users find businesses which have the accessibility requirements for their particular needs. McCann has cerebral palsy and uses a rollator to get around, and it's his personal experience of trying to access businesses like restaurants and hotels which proved the need for the software. "Growing up it was difficult for me to find accessibility information whether it was going out to eat somewhere or going to the shops and for me that's a really important thing to know ahead of time," says McCann. "I realised then going into college I wanted to be able to do something about that - figuring out the world isn't accessible at the moment and I wanted to make that change," he says.

NAIROBI, Jan 31 (Thomson Reuters Foundation) - Ugandan doctors are giving new mothers artificial intelligence-enabled devices to remotely monitor their health in a first-of-its-kind study aiming to curb thousands of preventable maternal deaths across Africa, medics and developers said. Doctors at Mbarara Hospital in western Uganda will give devices to more than 1,000 women who have undergone caesarean section births to wear on their upper arms at all times. Algorithms detect at-risk cases and alert doctors. Joseph Ngonzi from Mbarara University of Science and Technology, which is conducting the study, said it would help "improve monitoring in a resource-constrained environment". The World Health Organization says almost 300,000 women worldwide die annually from preventable causes related to pregnancy and childbirth - that's more than 800 women every day.

The tools, announced at Amazon's annual cloud event in Las Vegas, aim to help the company's cloud customers incorporate functionality such as natural language processing, but without long waits often associated with AI-related projects. Amazon says it is making this possible by integrating machine learning into the two new stand-alone services. "There's no machine-learning expertise required for either of these services. They're just plug and play. You don't have to get into all the weeds and get the training data and label the data and all those sorts of things," said Matt Wood, vice president for artificial intelligence services at Amazon Web Services.

Figuring out which products are in stock and which stock is likely to run low is a never-ending battle, as shoppers spend an estimated 40 billion hours picking things off shelves. It's also error-prone -- employees regularly misplace an estimated one in 10 items, contributing to global retail revenue losses exceeding $1 trillion. But drones hold the answer to the inventory tracking problem, if you ask serial entrepreneur Richard Schwartz. So strong is he in this conviction that he cofounded Pensa Systems, which develops inventory systems equipped with computer vision algorithms that "understand" what's on store shelves. The Austin startup today announced the close of a $10 million follow-on seed funding round that brings its total raised to $17.2 million, and according to investor and Pensa advisory board member James McCann, the future is looking bright.