workingon

Previous discussion on hn: https://news.ycombinator.com/item?id=27960865

Anyone play WOTV?

It’s a gacha that seems slightly less gacha then a lot of these.

HN is a bubble and it’s sad so many people live in it they have no idea what is happening to 50% of the population in this country and needs it drilled into their head repeatedly that they even exist.

Did you even read what I wrote? People are experiencing lack of housing even in not typically expensive areas. I’m from the middle of nowhere west virginia, and the rents have doubled and the only local jobs are fast food tier. These people have nowhere to go and no hope to find a place to live. Get your head out of your ass.

The cognitive disconnect is pretty insane here. Affordability issues are a widespread phenomen. You may be able to remote work in a low cost of living area, but the type of people who have housing instability can't. They have to be supported by the shit-tier wages in the low cost of living area, which is why it's a low cost of living area. Except now the "low cost" housing there takes up more than half someone's income at a median job in Bumfuck, State.

If you couldn’t afford a place to live you may feel differently.

The idea is they will buy it from Russia.

This isn’t a good faith comment. You assume that there is actually a problem with the authorities not having perfect access to criminal communications. We’ve gone the entirety of human history with ridiculously simple methods of analysis for crime etc., and the world is doing just fine. There will always be some crime, stomping out .1% more crime at great expense to the public isn’t even something that everyone would agree is necessary.

This all sounds like the solid advice I've heard before, but your very first line is where my disconnect lies. How is it not robbing my future self to divert that money into a savings account instead of having it sit in the 401k? All the money I would've had sitting in the savings account is likely experiencing growth (over time) in the 401k, making the chunk of cash that would've been sitting in the bank larger through growth. I have a decent percentage of the 401k in bonds, just in case the economy were to tank I'd still have some 401k to draw on (like a savings account).

I do see the issue with having to immediately start making payments back toward the 401k loan, but then I think you just take out a little bit more than what you would otherwise to cover those payments for the duration of the emergency.

Sorry if I sound combative, I am genuinely curious to the responses to these arguments. I'd be happy to find the flaw in my outline if it will help me make better decisions, I just don't see it yet.

I've heard this for awhile, and I never really understood it but perhaps it's just my particular situation. I can take an interest-free loan from my 401k-type plan at any time. I figure if I lose a job or have an emergency taking $5,000 out of that would be basically the same thing as taking $5,000 out of a bank, except it's been earning returns before that emergency. Do most 401k plans not allow you to take out money from your plan as long as you pay it back?

Thank you too for this information and for the resource, I'm looking at it now it seems very interesting.

The breakdown you give above seems to me to be more akin to something DL types tend to call 'feature engineering'. I also have a fun example, in this case it would be identifying land cover from satellite imagery. You can obviously just feed the raw reflectance values (RGB etc.) into a DL model to create a semantic segmentation of classes. However, it's been well established in literature at this point that that is not the most effective way to create a classification. This is similar to my previous comment, where there are lots of solutions that can be found through SGD based on these raw values.

There's a lot of traditional satellite imagery analysis algorithms that are based on very simple 'band-ratios', i.e. NDVI (normalized difference vegetation index) is calculated by Near Infared - Red / Near Infared + Red. This index will visually highlight areas of vegetation that was extremely useful in human sight-based analysis to identify vegetative areas. Now, you'd expect a deep learning model that takes in all the bands to have this information already, it has the NIR band and the Red band. However, explicitly doing the NDVI calculation and using it as an input feature leads to increased accuracies for classification. The exact reasons for this are unknown, but I think you touched on some of this above. With machine learning, and DL even moreso, sometimes it's necessary to hand-hold the optimization to optimize for exactly what you want. It helps 'explainability' and oftentimes helps accuracy, at the cost of some preprocessing.

I don’t at all. Sometimes living “paycheck to paycheck” has its perks. Make sure to have enough money with the last check of the month to pay the mortgage, while having the largest 401k contribution you can manage. Without active investing or any real savings there doesn’t seem to be much benefit to calculating cash in vs cash out any more than my bank app does. If there’s something out there useful for my case let me know! I’m just not sure how much optimization can be done with this lifestyle.

I’ve only ever played with constraints in the loss at this point. Would love more time to also deal with constraining the optimization. Do you have any examples that come to mind that worked well when doing this? Thank you for your suggestion!

I have never dabbled with constraining the parameter values themselves. Ive mostly put the constraints into the loss function. This works well when working with regression, a super simple constraint that adds penalty when the regression goes outside of the possible solution space has been extremely helpful in our work. Heuristically, I think it’s more useful during the first few iterations to find the correct local minima. If you were already finding the “correct” local minima this might be less important, but if you’ve ever dealt with convolutional artifacts (boxes, lines, edge effects) in your predictions, well informed constraints tend to help avoid these, as they are a symptom of being in an incorrect local minima.

It’s like any other system. A middle man has to take their cut and will increase costs. There is no reason (that I know of) to think the dealership model would ever decrease costs.

I would take the opposite approach here. Instead of overfitting, bake probability into your model. Whether using a Bayesian weights or using an ad-hoc version (I.e. dropout and batch normalization), you can do your predictions in ensemble and look at the deviation of predictions. the combination of this method and a small dataset usually leads to wider ranges of prediction, which can be interpreted as a model uncertainty. When using these techniques we have found smaller datasets lead to more uncertainty, but it often will bound the error. I think this is much more useful than assuming a small dataset means an overfit model will do good on future data.

I’ve found constraints in the loss functions are key to finding the correct solution space. With small amount of training data and SGD you can get a lot of mathematically “correct” answers, but a well informed constraint based on your problem space can eliminate 99.9% of the mathematically correct but practically incorrect answers.

It’s not a problem now, it will be a problem when you can easily get a Ford direct at MSRP but you have to pay dealer markup for a Chevy.

I technically live paycheck to paycheck after deductions, but I have a fairly high percentage (20%) going into 401k/Roth type stuff (before taxes income of ~80k). I bet these people are saving a lot for retirement as well.

Are you a late career researcher by chance? Your comments feel like how things used to be done. I think the most important thing to getting your paper read now (other than being in a nature level journal) is being indexed on google scholar.

I have multiple papers and none of them have university affiliations. The publishers are more than happy to take basically anyones money to just post your pdf online.

I was hoping for insight into a bunch of decomposition methods (SVD, POD, DMD, AA etc.). This was nice too. If anyone has any really good links discussing this type of decomposition please don’t hesitate to list some here for me :)

That’s how basic research works though. If you were older you’d be starting the see the results of the clickbait science from 40-50 years ago.

So you’ve come full circle here, though. What’s the difference between trying to have people not cheat the test at university and not cheat the test at the certificate authority? I was with you until you brought in this part, because it’s literally the same thing now but at a different building, basically.

Employer verification made sense, you mention they have to deal with it if their hire is dumb. This secondary certificate authority idea undermines your entire argument though. Maybe I’m missing something though and you have a good idea for how the CA will mitigate cheating that a university can’t do.

Because they are cheaper? I’ve hired unlicensed people to fell trees in my yard that weren’t really going to fall on the house etc.

Also, I’m fairly certain car mechanics are not licensed where I live. Sometimes licenses are just frivolous, especially in modern day society.