Lynx Roundup January 2nd, 2020

Lynx Roundup January 2nd, 2020

Version control for machine learning models! Getting started with deep reinforcement learning! No free lunch in Data Science!

Matthew Alhonte
Matthew Alhonte

How to version control your production machine learning models
With Category Theory, Mathematics Escapes From Equality | Quanta Magazine
Two monumental works have led many mathematicians to avoid the equal sign. The process has not always gone smoothly.
Getting Started with Deep Reinforcement Learning Can Be a Beast, Here’s a Way to Frame It
A lot of primers on DRL can be confusing and needlessly high level. There’s a simpler way to begin your journey. There’s a wonderful metaphor to be found in fantasy fiction that applies to learning…
There is No Free Lunch in Data Science - KDnuggets
There is no such thing as a free lunch in life or data science. Here, we’ll explore some science philosophy and discuss the No Free Lunch theorems to find out what they mean for the field of data science.

The Salk dilemma occurs in faulty markets that are optimized to allocate rival goods in win-win scenarios, resulting in win-lose or lose-win scenarios for non-rival goods.

Please note that value relativity does not imply subjectivity.

In free markets, transactions produce a win-win scenario, each of the parts gives something and receives something valued more by themselves. This works virtuously for rival goods, which cannot be shared, and are exchanged. Physical goods, as well as services or labor over them, are rival.

“Try not to become a man of success, but a man of value. Look around at how people want to get more out of life than they put in. A man of value will give more than he receives. Be creative, but make sure that what you create is not a curse for mankind.”

― Albert Einstein

Knowledge and the knowledge economy are increasingly relevant for the economy, and non-rival. They are treated as a degenerate case and fixed with laws as rival goods: patents, copyrights,… This results in a perverse economy where capturing value is rewarded and creating it is not, therefore producing the Salk dilemma.

The name of the Salk dilemma is inspired in the story of Jonas Salk, which is similar but more dramatic than Tim Berners-Lee and many others. Please share your favorite story in the comments, and remember: the plural of anecdote is not data.

“If I have seen further it is by standing on the shoulders of giants.”

― Isaac Newton

Jonas Salk discovered a vaccine for the polio, a critical disease at that time. He had to choose between:

  1. patenting the polio vaccine for hefty profits, acquiring a temporary monopoly, or
  2. giving it for free, possibly eradicating the disease.

He chose the latter. Motivations may include moral reasons and many others. Motivations make stories more interesting, but the story is not the point.

The point is that Salk should not have faced that dilemma. The relevance of this story is not in the quantitative or statistical aspects, e.g. expected profit, but in the qualitative of the possibility, i.e. it should not be possible in a well designed economy and markets. For a well-working society the incentives to contribute to the common good should be much greater than for selfish behavior. Instead, patents, lack of suitable business models for the knowledge economy, and a bias towards rivalrousness create the dilemma between win-lose and lose-win, In particular, among those two options, incentives are for bandit behavior, at a net opportunity loss that increases existential risks.

As knowledge becomes a larger aspect of our economies, markets (including international ones) should address the economy of non-rival goods and collaboration in better ways, even at the expense of less optimization for rival goods and competition. Current incentives shape the future. Yet once again, decision makers may have more incentives as bandits. In such a case, what are our options? The future is coming, and is up to us in the present to direct it towards utopia or towards dystopia.

2020 Conference
MLSys Website
AWS Machine Learning – Specialty Certification Preparation (Preview) - Cloud Academy
Accelerate progress up the cloud curve with Cloud Academy’s digital training solutions. Build a culture of cloud with technology and guided learning experiences.

Matthew Alhonte

Supervillain in somebody's action hero movie. Experienced a radioactive freak accident at a young age which rendered him part-snake and strangely adept at Python.