Welcome to a new installment of our Machine Learning News of the Week blog series. This week we cover an attempt to bring deep learning to the Internet of Things, share a couple of cool neural network use cases, and present the MLPerf results of our very own Zebra inference accelerator. Let’s dive in!
MIT System Brings Deep Learning AI to “Internet of Things” Devices (via SciTechDaily)
MIT is working on a system that will bring deep learning to the Internet of Things (IoT). IoT devices like smart thermostats and appliances historically had limited processing power and memory, which limited their use of neural networks. To circumvent this issue, MIT’s MCUNet system designs compact neural networks using TinyNAS, “a neural architecture search method that creates custom-sized networks.”
New method brings physics to deep learning to better simulate turbulence (via Science Daily)
It seems there’s a new use case for deep learning every day. From flood predictions to Covid-19 exposure estimates, and now turbulence simulations. Researchers at the University of Illinois Grainger College of Engineering have developed a new method that incorporates physics into deep learning to make better predictions for air turbulence.
“We don’t know how to mathematically write down all of turbulence in a useful way,” said Willett Professor and Head of the Department of Aerospace Engineering Jonathan Freund. “There are unknowns that cannot be represented on the computer, so we used a machine learning model to figure out the unknowns. We trained it on both what it sees and the physical governing equations at the same time as a part of the learning process. That’s what makes it magic and it works.”
Here is another great machine learning application in the healthcare industry. Vanderbilt researchers are working on a technique that uses deep learning to remove distortion in medical imaging procedures, making it easier to analyze the brain. Vanderbilt’s Synb0-DisCo algorithm “synthesizes what the MRI image should look like from anatomically correct images and uses that data to correct the MRI scan that was acquired.”
FPGAs Emerge Everywhere (via Electronic Design)
Great insight from Electronic Design’s Bill Wong on the emergence of FPGAs as a replacement for ASICs and other chips. Not only are FPGAs found in the cloud alongside machine-learning/artificial-intelligence (ML/AI) accelerators but, as Bill writes, they’ve also found themselves in the peripheral market. Our friends at Xilinx have created SmartSSD, a solid-state drive (SSD) that can utilize an FPGA to implement a variety of features that would otherwise need to be handled by a host processor.
Mipsology takes on GPUs with AI on FPGAs (via EENews Europe)
We announced some news this week too! Mipsology is proud to share that our Zebra ML inference accelerator, running on a Xilinx Alveo U250 accelerator card, achieved a peak performance efficiency 2x higher than compared to all other commercial accelerators in the latest round of MLPerf testing.
From our CEO, Ludovic Larzul: “We are very proud that our architecture proved to be the most efficient for computing neural networks out of all the existing solutions tested, and in ML Perf’s ‘closed’ category which has the highest requirements. We beat behemoths like NVIDIA, Google, AWS, and Alibaba, and extremely well-funded startups like Groq, without having to design a specific chip and by tapping the power of FPGA reprogrammable logic. Perhaps the industry needs to stop over-relying on only increasing peak TOPS. What is the point of huge, expensive silicon with 400+ TOPS if nobody can use the majority of it?”