Mobile ML Weekly Newsletter vol. 4

“Behind the Magic: How we built the ARKit Sudoku Solver.” An excellent look into how Brad Dwyer, founder of Hatchlings, made a Core ML model to solve Sudoku’s from from camera images in real time. Link

Apple releases the code for coremltools on GitHub. Link

Zalando Research releases Fashion-MNIST dataset. The dataset contains 70,000 labeled images of various articles fo clothing. The perfect dataset for training a model to be deployed in a fashion app. Link

Development on popular Deep Learning framework, Theano, will cease following it’s 1.0 release. Theano is the first of the major DL frameworks to fall in the wake of consolidation around competitors like TensorFlow. Link

“Smart Gesture Recognition in iOS 11 with Core ML and TensorFlow” Link

Computer vision startup Vize releases a Core ML model builder tool. Link

Google’s New Camera “Clips” Uses AI To Automatically Get Great Shots. “Google says it wanted to automate the process of both capturing and selecting great images….So it evaluates those photos on the device as they happen to determine what to save to memory.” Link

“Introducing NNVM Compiler: A New Open End-to-End Compiler for AI Frameworks” Amazon teams up with UW research team to create a new compiler for AI frameworks. Results show >10x performance boosts evaluating models on low power devices like the Raspberry Pi. Link

Read More

Mobile ML Weekly Newsletter vol. 3

Prisma Labs, maker of the Prisma app letting users add AI driven artistic filters to photos, has announced it will license its technology to businesses through an SDK. Link

A guide to hotswapping Core ML models on the iPhone. Link

TensorFlow 1.3.0 released. Adds high level functions for DNN estimators. Mobilenet support was also added to TensorFlow for Poets training script. Link

Box partners with Google to add automatic image recognition to customers. Currently everything is still run through Google Cloud, though the companies dedicated camera app is a prime candidate for edge computing. Link

Google acquired AIMatter, a startup making neural-network based models for photo and video editing on mobile devices. Link

Former Google and Baidu executive Andrew Ng is raising a $150M AI fund to complement is Deep Learning courses. Link

Gartner report on megatrends in tech driving digital business. AI featured prominently, of course. Link

“Stop just assuming you have a full lifetime to do whatever it is you dream of doing.” Unrelated to Mobile ML, but Scott Riddle provides motivation and perspective after being diagnosed with cancer at 35. We wish him and his family the best. Link

Read More

Mobile ML Weekly Newsletter vol. 2

Google’s Jeff Dean talks to YC companies about AI. Lots of interesting stuff here including an overview of different Google Brain projects, open source projects, and state of the art “learning to learn” techniques. Link

“The End of Typing: The Next Billion Mobile Users Will Rely on Video and Voice.” Mobile machine learning will help power user interfaces for the next billion people coming online. Link

Audio super resolution with neural networks. I’ve seen a lot of cool super resolution projects using images, but this is one of the first I’ve seen with audio. Link

Run or Walk (Part 1): Detecting Motion Data Activity with Machine Learning and Core ML. Link

iOS ARKit Tutorial: Drawing in the Air with Bare Fingers Link

(Paywall) Back to the Edge: AI Will Force Distributed Intelligence Everywhere Link

Building mobile apps with TensorFlow. A free O’Reilly e-book. Link

Read More

Mobile ML Weekly Newsletter vol. 1

Edge computing could push the cloud to the fringe. Link

How to train your own model for CoreML. A full tutorial on how to build an app to detect food in live video. Tutorial covers training your mode and building the app itself using Caffe, AWS, Core ML, and XCode. Link

TensorFire. An implementation of TensorFlow running natively in the browser with WebGL. Link

Qualcomm opens up its AI optimization software, says dedicated mobile chips are coming. Link

Researchers develop a way to model neural network’s power consumption to help developers make them more suitable for mobile devices. Link

Graphic footage shows Apple entering the AR space. Link

Read More