BiasRegularizer

Interesting. I am not seeing any availability here in Sweden. Perhaps all the international shipping options are gone. Since I am not in the US, I'd have to order it to ship directly to China.

Apparently (or so I was told) that the masks with ventilation holes are not nearly as effective. Hence my original link is one without a breathable hole. If you check Amazon again, all the hole-less N95 masks are all sold out.

Yes, scalpers swoop them up in all online stores. They are even sold out on Amazon US, Canada, UK and DE. I was told only the ones with a N95 rating can be effective against this virus.

https://www.amazon.com/3M-Particulate-Respirator-8210-Pack/d...

I was born and raised in Wuhan. My father and a few friends are still in Wuhan, one of the friend is a nurse. Here's what they told me:

* The hospitals in Wuhan are at absolute maximum capacity. they are lumping any fever cases (e.g. flu) patients with potential coronavirus patients in the same room.

* People with mild symptoms are sent home due to capacity, they are instructed to be quarantined at home unless symptoms get worse

* Despite reports, surgical masks are completely sold out in China, my dad is asking me to send him 3M (the brand) masks from abroad.

* Even hospitals are low on surgical masks for staff. Each nurse is given one or two masks per day, but per regulation, they needed to replace the mask every 3 hours.

* A wider scale locked down went in to effect in Wuhan this morning, forbidding anybody from entering or leaving on any roads, rail, water or flight.

* Domestic news are filled with non-sense, top 3 news today are praising Xi's CNY speech, which has exactly zero word mentioning the pandemic or Wuhan.

[Edit] One more thing,

* This coronavirus is potentially deadlier than SARS. Most of the 25 deaths are from the original 47 cases that was reported a few weeks ago.

Also in the VAE diagram, the latent space appears to have the same dimensions as input and output.

I'd like to chime in that Variational Autoencoder's theoretical framework is also quite useful in unsupervised learning tasks. It bridges variational inference methods (a powerful classical ML technique) with neural networks, thus allow very efficient representation, generalization and disentanglement.

One can report facts about unethical businesses without asserting personal beliefs, otherwise it's just bad journalism. If the background and evidence are sufficient, the readers can draw their own conclusions. There is no need to tell the readers what is evil and what is not.

In this case, the author emphasized too much on his/her own belief, which erodes into the integrity of the valuable facts he provided. His/her personal opinions turned an exposé into an editorial.

This exposé would have been more convincing if the author/whistleblower did not inject so much of his own political views into it. State the facts and let readers decide.

I am a big proponent for climate actions, but excessive vilification and oversimplification are counter productive to the cause.

True, but having some noise in the label is actually good for generalization. If it's only learned on perfectly correct sentences then its tolerance for mistakes will be very low.

I think maybe CLEVR[0] dataset is what you are talking about?

Keep in mind that a most of the current ML systems have diverged from biology. A majority of the recent breakthroughs come from mathematics, the rational is that just because human brain does it in a certain way does not necessarily mean it is the only way to do it.

[0] https://cs.stanford.edu/people/jcjohns/clevr/

The keyword is "high impact". The link you provided is only ranked by publication count, it doesn't consider the impact factor of journal/conferences. If you look at stats from ICML this year[0], the only Chinese institution TsingHua is ranked #13 in the top #20 list.

Take a look at recent major advances in deep learning: information bottle neck, mask-rcnn, transformer, normalizing flow, all originated from US/Canadian/EU institutions. Chinese academia has very serious systematic issues that discourages innovation and encourages quantity over quality.

https://medium.com/@dcharrezt/icml-2019-stats-4ba18fbc6543

This is not completely true. China lead in terms of AI startup funding, but due to the huge shortage of talents, a good number of startups don't hold much water.

The top Chinese talents (e.g. Kaming He, Fei Fei Li etc) all chose to move to the US. AFAIK, there are no Chinese institutions that employ elite foreign AI researchers. The US universities and companies still lead with a large margin in terms of high impact publications in NIPS, ICML, CVPR etc.

I am not necessarily advocating on the tree solution, as someone else has pointed out, it's terribly inefficient as a carbon capture method. I am more arguing that it's more effective to say "how can we find a better alternative" than "How can we change the way everyone does thing". General population does not behave in a rational manner and its often short sighted, we can't expect societal self discipline as the main solution against climate change.

In the US and EU we are generally conscious about environment on a personal level (lifestyle etc), but the majority of the growing pollution will come from the upcoming superpowers, and they have very little intention of hindering their economic development with very radical environmental policies.

Take air travel for example, there is absolutely no way China and India will slow down their airline industry's growth. We need to start treating this fact as roadblocks, and must find solutions around it.

Wouldn't carbon capture be a much better solution than slowing down production in the long run? We can potentially restore a large amount of ecosystem within forests as humanity move towards further urbanization.

I don't want to be too pessimistic, but even if the US and EU become carbon neutral tomorrow, China and India with their 3 billion people will cause an explosion of CO2 emissions as their population enter the middle class with disposable income. China is about to overtake US as the biggest air travel market in 2020, and India is just beginning its mass industrial revolution. No amount of external political pressure will make these countries to halt its GDP machine.

"Importance" is a fairly overused word in DL. e.g. importance sampling and importance weighted gradients.

"Attention" works by creating inductive bias for the upstream network, which is analogous to human attention, and the word itself is much more intuitive.

Keep in mind machine learning is largely a descriptive science(modeling the behavior), whereas neuroscience is more prescriptive. So from the behavioral perspective, attention is better suited than importance.

While Transformer's Self Attention(SA) is great, there are many applications where SA doesn't apply. For a more comprehensive overview of attention mechanisms, I often find myself coming back to Lilian Weng's post:

https://lilianweng.github.io/lil-log/2018/06/24/attention-at...

Fittingly reported by The Register

Are you referring to SAAB?

If I recall correctly the company was not going to make it during the Great Recession, and no buyers wanted them. So the Chinese bought them but still couldn’t save them from their ultimate fate.

The Chinese owner tried to revive the brand a couple of years ago with an electric vehicle, but I guess that didn’t work out after all.

Care to expand on that? Why can't you do both?

What constitutes as operating? Does the drone need to be completely autonomous or can I stand outside the park and remote control it?

Honest I think this is worse than ReCaptcha. At least with ReCaptcha I know I am helping someone somewhere labelling their data and possibly used for autonomous driving.

With this captcha I feel like I'm just wasting world's energy on useless computations.

Nice job OP. I noticed that you used word2vec for the embedding space, do you think more modern ones like seq2seq or BERT might give better performance?

Non-american here. Can someone explain what's so shady about NRA aside from their lobbying for gun rights? Also what is NRA's rational/rebuttal for their shady behaviors?

I have read a few cases of identical twins having different sexual orientation. Perhaps they are also good subjects of study, since their genetic make up are so similar, it would be easy to identify if/which genes correlates to sexual orientation.

Not necessarily true. Uber the company maybe operating at a loss but rides are not operating at a loss AFAIK. Kind of like Amazon in its earlier days, where they reinvested their revenue in growth.

I am under the impression that the courses are designed for EE/CS engineers to get familiar with the foundations of modern ML, but it's not sufficient education to work as a full time ML engineer.

I returned to grad school for ML two years ago, and even now I still struggle with some ML job interviews when it comes to statistics and theoretical questions that I've studied two years for. One particularly challenging part of ML interview is that it covers much more than a typical CS interviews that I'm used to. I had a ML engineer internship interview with a famous ML company recently, and I was asked about sorting algorithms, hashing algorithms, non-convex optimization techniques, gaussian processes and manually compute the jacobian of a NN for backprop on the spot.

Grew up in central China (Wuhan), a notoriously swampy region, we were always taught about two types of mosquitos: the regular ones and the "poisonous" ones (tiger mosquito). I remember avoiding these tiger mosquitos like plague. To this day I still hate mosquitos with a passion.

These little buggers are so adaptive, they are like the stealth jets of mosquito world: They are smaller, quieter, swifter. When they land on your skin you wouldn't feel a thing even in plain sight. They've even learned to avoid blue lights (traps) and became resistant to various bug repellant that we used. It was an endless arms race. Worst of all, its bite would leave you itchy for days. I still remember in some of my elementary school photo, my legs looked like they were the surface of moon.

If there is one thing I learned is that if you want to get rid them, you would have to do it 100%, because even if you leave 0.1% alive, they will adapt and come out stronger than ever.

I believe nuclear power is inevitable. Human power consumption has been growing exponentially, at some point renewable energy will not be able to catch up without causing extensive damage to our ecological system. There are finite amount of renewable energy per sqkm on earth, at some point the area required will begin to erode local ecology. Right now nuclear power appears to be the densest form of energy generation. I agree in the short term renewable energy is a great alternative to coal/natural gas.

In machine learning we often use regularizers to reduce bias (amongst many other neat things that it can do), and I've been thinking if similar techniques could be applied to help us humans to reduce bias and strengthen our belief system against malicious information.

Some information contain objective facts (e.g. climates change, vaccination), thus citing reputable research is often enough.

Though many of the political and economical belief have little objective truth to them (e.g. government regulation, taxation plan), and it is useless to discuss bias, we can only hope to prevent from malicious information. Perhaps allowing both sides to show arguments can help reduce bias, but this technique is far from perfect.

Are you positive about that? I just tried to VPN from a few different locations in the US (Texas, Georgia and Seattle) with a fresh private browsing session each, all of the headlines appear the same and doesn't change by region.