Deep Reinforcement Learning Doesn’t Work Yet, by Alex Irpan

Source: Deep Reinforcement Learning Doesn’t Work Yet, by Alex Irpan

here are some of the failure cases of deep RL.

Deep Reinforcement Learning Can Be Horribly Sample Inefficient

If You Just Care About Final Performance, Many Problems are Better Solved by Other Methods

Reinforcement Learning Usually Requires a Reward Function

Reward Function Design is Difficult

Even Given a Good Reward, Local Optima Can Be Hard To Escape

Even When Deep RL Works, It May Just Be Overfitting to Weird Patterns In the Environment

Even Ignoring Generalization Issues, The Final Results Can be Unstable and Hard to Reproduce

The way I see it, either deep RL is still a research topic that isn’t robust enough for widespread use, or it’s usable and the people who’ve gotten it to work aren’t publicizing it. I think the former is more likely.

My feelings are best summarized by a mindset Andrew Ng mentioned in his Nuts and Bolts of Applying Deep Learning talk – a lot of short-term pessimism, balanced by even more long-term optimism.

Just Teach My Child The Maths, by James Tanton

Source: Just Teach My Child The Maths, by James Tanton

It is astounding to me that mathematics – of all school subjects – elicits such potent emotional reaction when “reform” is in the air.

So … Can we educators work to understand this sentiment, put the right words to it, and fully engage in conversation about it? Can we be fully transparent about our approaches and intents and listen to, honor, learn from, and respond to parental and societal reaction with clarity and grace?

Of course we can and of course we must.

We need to communicate the true goal of given exercises to parents.

If we are going to ask students to practice mathematics ideas, we need provide interesting or meaningful examples with which to practice them.

We need to be sure not to insist on one approach when analyzing a problem. We need to encourage students to generate efficient practices.

Work to have students show their work only when there is work worth showing. Let’s honor our students’ intellectual capabilities and time!

Make showing/explaining your work interesting.

If we truly acknowledge there is a change in mathematics education – as this internet piece purports to demonstrate – then we need to stand by what we value: understanding, flexibility of thought, innovation, problem-solving, reflection on solutions and approaches, and the search for efficiency and elegance.

We must acknowledge that testing agencies in most parts of the world have not yet caught up with what we educators value. We must find the means for students to experience tremendous success on all fronts – with speed testing and with deep understanding and mathematical innovation.

No one’s coming. It’s up to us. – Dan Hon – Medium

Source: No one’s coming. It’s up to us. – Dan Hon – Medium

If technology is the solution to human problems, we need to do the human work to figure out and agree what our problems are and the kind of society we want. Then we can figure out what technology we want and need to bring about the society we want.

Because if we’re a tool-making, tool-using species that uses technology to solve human problems, then the real question is this: what problems shall we choose to solve?

These are questions that cannot and should not be left to technologists alone. Advances in technology mean that encryption is a societal issue. Content moderation and censorship are a societal issue. Ultimately, it should be for governments (of the people, by the people) to set expectations and standards at the societal level, not organizations accountable only to a board of directors and shareholders.

As a society, we must do the work to have a point of view. What does responsible technology look like?