Machine Learning Deep Learning World (June 2020)
In this free issue: current machine learning deep learning trends, news, resources, sneak preview of paid subscriber content. Deep Learning at Tesla. Summer hackathon event and resources.
Launching private, searchable, paid subscriber blog
Having a searchable blog that requires authentication allows us to show every one what kind of resources are available. Free signups get previews and paid subscribers can quickly access and search for relevant resources. We also link to our Medium blog networks this way we have all the information in one place, organized by topics and keywords.
Current easter eggs
We routinely send easter eggs to paid subscribers. Current easter eggs including free deep learning ebooks worth $29.99, top datasets to download, important deep learning news, trend, code, papers.
Note: only active paid subscribers, after 1st payment, get easter eggs and access to paid only blog https://data-bootcamp.blogspot.com/ During paid subscriber trial phase, you can only see previews and subscriber only articles on uniqtech.substack.com.
Current events and trend
Pytorch 2020 Summer Hackathon (Due Aug 25 2020) - Pytorch summer virtual hackathon hosted by Facebook HQ. https://pytorch2020.devpost.com/
Current Hackathon and Hackathon Resources for paid subscribers
View resources here: Data Bootcamp hackathon resources and hot project ideas 2020
We know a thing about hackathons because we were top three at pytorch and angular hackathons where core developers and creators are the judges!
One pro tip for machine learnists is to always ask for GPU credits, hosting credits. As training cost can get expensive.
The learning opportunity at hackathons is invaluable:
Attending hackathons is the no.2 way for developers to level up! No.1 is having a developer job.
Previously, a writer went to a city hackathon by the San Francisco city. There was a small company presenting and advocating for Esri, a geo location service. She sat in, took notes, and used it in a scavenger app. Guess what Esri powers almost all of the COVID-19 dashboards, including the famous one by Johns Hopkins.
A typical hackathon schedule may include door opening, swags, sponsor talk, tech talk, workshops, API introductions, one-on-one office hours, overnight coding, coffee and simple pizza dinner, pitches from product managers product designers and technical founders, pitches presentations, judging, happy hour.
We find the API tech talk to be most useful.
Previously important news:
Pandas turns 1.0
Pandas 1.0 explained for paid subscribers. Official document What’s new in Pandas 1.0 . It also deprecated a lot of functions. Tip you can improve pandas performance and by doing so, significantly improve data preprocessing speed. Check out our Medium article about Pandas, Pandas Cheatsheet. Paid subscribers read for free.
Python turns 3.8
The concept of f-string formatting was introduced in Python 3.8.1
Our medium article on Python 3.x. Available for Medium members. Paid subscribers can request link here hi@uniqtech.co
It’s time to stop using Python 2.x. Moving from Python 2 to 3.
Pytorch turns 1.5 Pytorch 1.1 added support for tensorboard, announced previously in F8
How Tesla uses deep learning?
First presented by Tesla AI manager at Pytorch conference 2019. Guest post published, shared with permission. This is a special link available for all our email subscribers (paid and free).
Full transcript (writer, software generated) available for paid subscribers hi@uniqtech.co We also have transcripts in the pipeline for how Quora uses NLP and how Lyft uses deep learning and computer vision. Let us know if you want to see some content sooner.
Current news and hot topics
Generative neural networks, deep fake, how they differ from super resolution. Ethics and fairness in machine learning.
Duke’s Pulse generates high definition faces from pixelated images. But it explains this is very different from super resolution or SR, an image processing task.
"Super-resolution imaging (SR) is a class of techniques that enhance (increase) the resolution of an imaging system. In optical SR the diffraction limit of systems is transcended, while in geometrical SR the resolution of digital imaging sensors is enhanced." - Wikipedia
The researchers explain "We have noticed a lot of concern that PULSE will be used to identify individuals whose faces have been blurred out. We want to emphasize that this is impossible - PULSE makes imaginary faces of people who do not exist, which should not be confused for real people. It will not help identify or reconstruct the original image.
We also want to address concerns of bias in PULSE. We have now included a new section in the paper and an accompanying model card directly addressing this bias."
Ethics and fairness
Black Lives Matter movement is changing machine learning. Top tech companies like Amazon now hesitant to provide facial recognition technologies to law enforcement.
Due to lack of training data, and lack of quality control, fairness, sometimes, facial recognition gets it wrong! Remember previously Google Photo’s algorithm misidentified African Americans? Facial recognition performs notoriously poorly on female identification, minority identification and have misidentified innocent men as criminals.
Facebook invests in privacy improvement while Mozilla invests in ethnics on the internet.
A frequently seen bias is in GANs only generate Caucasian faces. For example, previously there was a network that will generate our Renaissance portrait from a photo. No matter what ethnicity is the original photo, the output is always white because the training data is mostly European oil painting. Such bias exists and researchers are working on it.
In fair housing and fair lending practice, regulations are requiring financial institutes to document and explain the unexplainable : the black box of machine learning algorithms. Remember Goldman Sachs + Apple credit card gives a husband higher credit than his wife even if they file the same tax return, under the same household.