Roundup Let's get stuck into another roundup of AI news beyond what we've already reported.
Detect weeds with AI: Here’s a fun little machine-learning project: train a model to detect weeds among grass to they can be removed or sprayed with weed killer, using an AWS DeepLens camera and a microcontroller.
Amazon has described how to build and train such an image-classification system using its SageMaker service for data scientists and machine learning enthusiasts. You upload your training data, made up of images showing grass and weeds, to an AWS S3 bucket. These pictures are passed to a convolutional neural network that has already been trained on ImageNet to learn the features in your training dataset. The model can then be deployed on a DeepLens device to sniff out weeds from subsequent snaps of your lawn.
But if you want to go a step further, you can hook up the gizmo to a microcontroller to get it to spray weed killer when the camera detects unwanted foliage. And if you’re a really serious gardener, you can hook up a few sensors that detect soil moisture to a Raspberry Pi to show you where you need to water your plants. Be warned, the instructions are for a pretty crude system that is capable of handling a small patch – a 2 ft x 1ft plant bed.
The project is more of an educational exercise in how to deploy systems using AWS. “[Deep Lens] lets you learn and explore the latest artificial intelligence tools and technology for developing computer vision applications based on a deep learning model,” Amazon said.
You can follow the step-by-step instructions here, and customize it as needed.
BERT powered search engine for coronavirus research: Google has launched a search tool based on its language model BERT to help boffins explore the latest research papers on the COVID-19 coronavirus.
The US Office of Science and Technology Policy put together a dataset containing more than 29,000 scientific studies relating to the coronavirus pandemic with the help of Microsoft, the US National Institute of Health’s National Library of Medicine, and the Chan Zuckerberg Initiative. These were then integrated into an academic search engine known as Semantic Scholar built by the Allen Institute of AI.
Semantic Scholar uses natural language processing algorithms to help researchers search for relevant papers by analyzing the text in each study. Now, Google has emitted the COVID-19 Research Explorer, another search tool that works with Semantic Scholar to refine search even further.
Googlers trained its large language model BERT on a dataset containing sample queries alongside the corresponding answers, to teach it how to identify information. The goal was to create something more flexible than traditional algorithms that search text by simply matching phrases. The explorer can help scientists answer more open-ended technical questions, such as, “what regulates ACE2 expression?”
A text-matching tool will probably highlight papers that contain the word ACE2, which describes an enzyme, rather than finding studies that attempt to answer that question. Google’s COVID-19 Research Explorer, however, lists a number of papers that are relevant to the specific question by highlighting specific passages. Researchers can then narrow down their search further by reading those passages to decide whether the paper is useful to them or not.
“The COVID-19 Research Explorer is freely available to the research community as an open alpha,” the web giant said. Over the coming months we will be making a number of usability enhancements. Try out the COVID-19 Research Explorer, and please share any comments you have with us via the feedback channels on the site.”
You can play around with the tool here.
Waymo is back on roads in Arizona: Waymo suspended all testing its self-driving cars as America grappled with the COVID-19 pandemic. Testing required engineers and drivers working in close proximity for long periods of time. To prevent the spread of the coronavirus, Waymo suspended its operations in March.
However, as Arizona slowly reopens businesses, it’ll resume testing again starting May 11.
“Even under the extraordinary circumstances of COVID-19, we’ve continued to advance our technology and mission to make it safe and easy for people and things to get where they’re going,” the ustart announced. “Our continued focus on hardware and software development, driving in simulated environments, and ongoing investment in advanced algorithms, machine learning, and evaluation means we have not taken our foot off the pedal during these unprecedented times.“
Waymo staff will be encouraged to wear face masks and keep six feet apart. Drivers will also have to wear protective equipment, and vehicles will be disinfected and cleaned multiple times a day. Although testing will be resumed, it will not be picking up any riders around the Metro Phoenix area as part of its Waymo One service for now.
Facebook cancels its AI residency program: Facebook will not carry out its AI residency program this year due to the COVID-19 outbreak.
Jerome Pesenti, head of AI at the antisocial media giant, said in a statement: “We continue to put the health and safety of our community first as we respond to changes caused by the COVID-19 pandemic."
“As a result, we have decided not to move forward with the 2020-2021 AI Residency Program,” Pesenti said. “An optimal and consistent hands-on residency experience is critically important to the program’s success, and this isn’t possible in a remote format. We look forward to keeping in touch with this year’s candidates for future opportunities."
That’s bad news for applicants hoping for a place on the one-year paid position to work with a Facebook AI researcher and engineer on a machine-learning problem.
Facebook has announced strict steps to curb the coronavirus spread. It announced it would ban all events or conferences with more than 50 people until June 2021, and has said employees could continue working from home until the end of the year. ®