Drones and machine learning algorithms are helping beachgoers get live updates to alert them whenever sharks are in the water.
Academics in Southern California teamed up with Salesforce to develop a system capable of spotting the marine predators by flying drones overhead. The camera footage is then run through computer vision software to detect them swimming in the ocean.
Text messages are then sent to people, warning them whenever a shark is nearby. The project, dubbed SharkEye, aims to help marine scientists study the animals’ behavior and to protect people.
The study described in the New York Times is still in its pilot stage. The researchers hope to develop their alert system further and extend their coverage to larger areas of the ocean.
Speeding up TensorFlow on Macs
Developers using TensorFlow, a Google-made AI framework, can now train their models faster thanks to Apple’s latest Arm-based M1 hardware.
The chip packs a new eight-core CPU and up to an eight-core GPU that is better at accelerating machine learning workloads compared to Apple’s Intel hardware in its older laptops. The latest version of TensorFlow 2.4 has also been optimized to run faster for the company’s Intel-based computer, but results lag behind the M1 chip.
“Until now, TensorFlow has only utilized the CPU for training on Mac,” Apple announced this week. “The new TensorFlow 2.4 leverages ML Compute to enable machine learning libraries to take full advantage of not only the CPU, but also the GPU in both M1 and Intel-powered Macs for dramatically faster training performance.”
You can see how much training times are slashed in some preliminary benchmarking results here. If you’re a Mac user and want to download the speedier TensorFlow 2.4, download the software from Apple’s GitHub page here.
World’s largest chip better than CPUs and GPUs, CEO claims
Cerebras, the Silicon Valley-based hardware startup known for crafting the world’s largest chip, claimed that its hardware was 200 times faster than a supercomputer at crunching through math equations modeling fluid dynamics.
Its CS-1 system, containing a single 18GB chip fitted inside a complex cooling device about the size of a mini-fridge, topped the world’s 82nd fastest supercomputer in a series of experiments. There is an important caveat, though – the CS-1 was only pitted against the largest cluster, some 16,384 CPU cores, in the Joule supercomputer.
The massive machine, operated by the US Department of Energy, has a total of 84,000 CPU cores. “Because of the radical memory and communication acceleration that wafer scale integration creates, we have been able to go far beyond what is possible in a discrete, single chip processor, be it a CPU or a GPU,” said Cerebras co-founder and CEO Andrew Feldman.
The results should be taken with a pinch of salt, however, as the company has yet to publicly disclose its chip performance in more typical benchmarking tests used for AI and machine learning.
Create your own Google GAN monster
Google released a new tool capable of transforming computer drawings to full-blown images of monsters using generative adversarial networks (GANs).
The app named Chimera Paint “creates a fully fleshed out rendering from a user-supplied creature outline,” it said in a blog post this week. The wacky project started as an idea to create fictitious creatures for a virtual card game. Engineers trained GANs on a dataset of 3D creature models drawn by artists that were brought to life using a game engine.
You can have a go at sketching monsters in the web app here, and watch it get transformed into a character of its own. ®