The Five Best Things: Oct 24, 2020
GANs GANs GANs, GANs GANs GANs GANs
 |
(I should have made it more obvious I was going for the Barbra Ann/GAN joke)
Today I go deep on GANs, or Generative Adversarial Networks. These networks are becoming relevant across a variety of domains - fashion, art, beauty, self-driving cars, medicine, synthetic simulations - due to their ability to generate new data that mimics the characteristics of input data.
The Five Best Things
My layperson understanding of how GANs work - there are two networks trained on the same input data set, but with different objectives. The ‘Generator’ network must produce new examples that mimic the input training data. The ‘Discriminator’ network must then determine if the new examples generated are fake or real. They go back and forth until the Generator network starts producing examples that successfully fool the Discriminator network. I like to visualize it as a game between an art forger and a detective, with the forger getting better and better until the detective is no longer able to tell a fake painting from a real one.
I found this good depiction of GANs on Reddit -
Some currently available applications of GANs -
L’Oreal uses GANs in this tool where you can turn on your camera or upload a picture, and see how different makeup and hair styles look on you.
GANimal - upload a picture of your dog, see how it would look like if it was another breed.
GauGAN - generate photorealistic images from just a couple of brush strokes.
There are many applications in self-driving cars research.
They are used to model phenomenology in space and in the Large Hadron Collider.
GANs for Good by DeepLearning.ai - I attended this virtual panel with Andrew Ng, Sharon Zhou, Anima Anandakumar, Ian Goodfellow and Alexei Efros a few weeks ago, where they gave an overview of GANs and discussed other potential applications.
Of course there are many serious ethical concerns around GANs, including Deepfakes, copyright ownership, spoofing, cyber security intrusions etc.
MSN: After His Son Co-Founded Coursera, He Became Its Earliest Student and Completed 146 Courses
This article is a profile of Ronald Ng, the father of Andrew Ng and presents a life filled with relentless pursuit of knowledge and self-improvement. Some quotes -
“These life lessons taught me that we need very little in life to make us happy, provided we have that frame of mind to enjoy whatever we have.”
“The joy of learning helps keep the mind sharp and allows us to appreciate the beauty of the subject matter.”
Andrew Ng -whose material on GANs is referenced above - is one of the foremost authorities on Machine Learning today. He is a Professor at Stanford, founder of Google Brain, founder of Baidu AI research, and an AI investor. He also kicked off the MOOCs revolution as the co-founder of Coursera, and published the most popular Machine Learning class on the platform. A peek into his Dad’s journey helps explain how he ended up with such a wide-spanning career. His dad grew up in Hong Kong and became a hematologist, then moved to Singapore and served in the army, before moving and settling in the U.K.
Invest Like The Best: Jesse Livermore
This podcast with analyst ‘Jesse Livermore’ (I explain the quotes below) dives into why the stock markets are back at all time highs, when by all reasonable indicators we are in a horrible economic downturn. He explains, by turn, the impact of the Fed’s response, the fiscal response, capital vs. labor, inflation concerns and the market’s biggest concern being gridlock in Washington, D.C. The best phrase in the whole recording - “We have to heal the wound before worrying about the bruises.”
‘Jesse Livermore’ is a pseudonymous analyst, who publishes a few, thoroughly researched pieces a year. He named himself after a brilliant stock broker of the early 20th century. The internet and podcast medium allows him to maintain his anonymity as he publishes brilliant pieces. Howard Marks’ latest memo arrives at similar conclusions as Livermore about today’s markets.
This piece made me feel a bit less worried about the inflationary pressures due to the unprecedented money printing that is going on right now. We know inflation will happen, we don’t know where, but we have the tools to fight it. The biggest roadblock to recovery is the lack of a fiscal response due to gridlock in our 3 chambers of lawmaking. Something to think about with the upcoming election.
History Extra: A Sikh tragedy: the Indian kingdom that fell foul of the British empire
This post goes into the history of the rise and downfall of the Sikh empire in India in the early 1800s. It chronicles the rise of Ranjit Singh, who united the various Sikh misls (clans) into a formidable empire that the British East India Company was unable to topple. He utilized a combination of war, diplomacy, personal and trading alliances, in order to cement his power and established an enduring legacy in the Golden Temple. The article also briefly spotlights the female mentors who aided his rise. Ultimately though, insufficiently grooming his son as successor led to the downfall of the empire, as the British were able to spread rumors of the new king being an ‘imbecile’.
I love stories from Indian history, especially when there are lessons of statecraft. The Sikhs are a particularly inspiring religious minority in India, whose teachings are rooted in equality, service to the country, and to the community. Funny how so much of Ranjeet Singh’s story rhymes with that of Toussaint Louverture - who led the only successful slave rebellion in Haiti, and Napoleon.
Evolution is so cool (click the tweet below to watch the whole snippet)
Honorable Mentions
Intel announced Q3 2020 earnings, and there’s not a lot of good news. Datacenter and IoT sales nosedived, and client computing (laptops, notebooks) are only up 1% in the pandemic.
The CEO declaring that “wonderful progress” has been made in fixing 7nm production since the last earnings call 3 months ago - sounds kinda Trump-y. They also announced a deal to sell their NAND memory (used in thumb drives, hard drives, cameras etc.) business to South Korea’s SK Hynix for $9B, which makes me suspect there is a cash shortfall. Tax rate increase by 4.3% is strange too; they claim it is due to more sales outside of the U.S. with higher tax rates - implies they’re not selling well in the U.S. which is their primary market - yikes.
WSJ: Barbie Sales Take Off as Parents Try to Cut Down on Screen Time Mattel’s sales are up 29% this quarter.
CNN: A fourth-grader walked to school to use its WiFi because he didn't have internet at home
Marie Claire: The #endSARS Movement In Nigeria: What to Know There is a lot of turmoil in Nigeria, and this was the best explainer I read. I find that some of the best reporting nowadays comes from fashion magazines.
Nathan Tankus’ tweet sent me down a rabbit hole of looking into Donald Harris, Sen. Kamala Harris’ father. He’s still alive, was the first Black tenured economist at Stanford, has ties to the University of Wisconsin (!!) and was once considered a heterodox for suggesting that uneven/unequal economic development is a feature, not a bug.
Marie Claire: 25 Years of ‘Dilwale Dulhania Le Jayenge,’ Bollywood’s Best Love Story 😱 was my reaction when it hit me that this movie came out TWENTY FIVE years ago, because every scene is etched in my brain. A lovely retrospective of why this movie is an enduring blockbuster.
WSJ: Coronavirus Pandemic Speeds Shift to Cleaner Energy
Job Drop
Product Manager, Google Cloud - my org is looking for folks with experience in the HPC/Supercomputing space particularly, and is open to grooming engineers wanting to make the transition to product. Please reach out if you fit this criteria.
Data Science Fellow, Ken Kennedy Institute at Rice University - go work with my friend Angela Wilkins! She is simply the best.
Disclaimer: The views and opinions expressed in this post are my own and do not represent my employer.