TechChat Tuesdays #28: TensorFlow with Rich Freedman

Tags: , , , ,
Category: ,

Today Sujan takes the reins as host, and welcomes long-time Charioteer, Rich Freedman! Sujan and Rich cover dev news, then delve into machine learning, TensorFlow and TensorFlow Lite, and some Arduino work Rich has been doing recently.

Show Notes

  • Shameless ETE plug: Philly ETE Conference is back for 2021, and Alan Kay is our first keynote! Other notable speakers include Kent Beck, Jessica Kerr, David Nolen, and more. We’ll be virtual again this year – get your Early Bird tickets before February 15, 2020 for only $70.
  • Google and Android join The Rust foundation. “For example, on Android, we’ve found that more than half of the security vulnerabilities we addressed in 2019 resulted from memory safety bugs.”
  • Android barcode scanner infects millions of devices
  • “I don’t want to do front end anymore”. This developer is hanging the towel on front-end development, citing the complexity of the whole stack, quit his job and is looking a full time backend job, and the comments on HackerNews are great. “Oh boy. I sympathise entirely with this post, but the story’s exactly the same on the backend.”
  • Just when you thought Elon Musk couldn’t surprise you more… Tesla buys $1.5 billion of Bitcoin and says it will soon accept payments in Bitcoin.
  • If you’re looking for a lightweight, fast API server in Python, check out this Toptal blog post on building high performing apps with FastAPI.
  • Data structures and algorithms! Personally I’m not a fan of regurgitating them in coding interviews, but to be a great programmer, you need a solid understanding of them. If you’re looking for a great, free book on algorithms, check this one out by a University of Chicago professor.
  • MIT is 3D fabricating fully functional drones using a laser cutter. Check out the video.
  • Ubuntu switches to Google’s Flutter for it’s installer!
  • Tensorflow: a free and open-source library for Machine Learning developed at Google and released in 2015. Mostly focused on training and inference of deep neural networks.
  • TensorFlow Lite: released in 2017, and aimed specifically at mobile and embedded processors (i.e. relatively low-power). Available on Android, iOS, Linux (including Raspberry Pi), and on some embedded microcontrollers.
  • Models are developed and trained on full TensorFlow, and then optimized and converted, using the TensorFlow Lite converter, to the format required by the TensorFlow Lite interpreter. The act of running a model through the interpreter is known as “inferencing.”
  • TensorFlow Lite is now available for some microcontrollers via TensorFlow Lite for Microcontrollers. Also avaialble on Arm-Cortex-M Series processors, ESP32, as an Arduino library for some Arduino boards (notably the Arduino Nano 33), and on Mbed OS.
  • Person Detection is a somewhat advanced, but standard example of TFLite. We worked to make it run with the relatively new Arduino Portenta H7, along with the Portenta Vision Shield.
  • The Portenta H7 uses Mbed OS
  • Breathtaking 10 Terapixel image of the night sky contains 1B galaxies. Now, go hunt for the number 42!

See you next week!

This livestream took place on February 9, 2020 at noon. This is a recording. Subscribe to our YouTube channel to be notified of our next livestream – or bookmark this link and tune in every Tuesday at noon EST to hang out with Ken Rimple and Sujan Kapadia.