The idea that simply having a lot of parameters leads to overfitting was shown to not be the case over 30 years ago by Vapnik et al. He proved that a large number of parameters is fine so long as you regularize enough. This is why Support Vector Machines work and I believe has a lot to do with why deep NNs work.
The issue with Vapnik's work is that it's pretty dense and actually figuring out the Vapnik-Chervonekis (VC) dimension etc is pretty complicated, and one can develop pretty good intuition once you understand the stuff without having to actually calculate, so most people don't take the time to do the calculation. And frankly, a lot of the time, you don't need to.
There may be something I'm missing completely, but to me the fact that models continue to generalize with a huge number of parameters is not all that surprising given how much we regularize when we fit NNs. A lot of the surprise comes from the fact that people in mathematical statistics and people who do neural networks (computer scientists) don't talk to each other as much as they should.
Strongly recommend the book Statistical Learning Theory by Vapnik for more on this.
Nope. Abuse should never be accepted. The right solution for actual abuse of a policy like this as opposed to an error is to immediately fire the person conducting the abuse. The inability or unwillingness to make decisions like this fast is the root cause of a lot of nonsense that goes on at businesses.
If it's an error, you correct and accept errors as the cost of doing business.
> A good strategy would be to have them be a lead on a minor project that
> requires technical chops and communication. Ensure there is a daily standup
> and grind them on details and timelines. Get them some juniors as direct
> reports to expose their lack of knowledge, then have meetings with these
> junior devs about project performance.
Do not do this. You do not need a “strategy”. Just do it! Rip off the band-aid. Either tell them the truth or give no reason even if asked. Reverse the mistake asap and get moving on your company’s mission. You don’t need one second more of this. Nor does anyone else at your organization. Do it today! Do it dispassionately. Do it as nicely as possible. You are going to feel bad because you are a good empathetic human.
Some people will have a reaction. Some people will “have a reaction” to try to get more out of you. You are not responsible for either. You are responsible only for your own actions. Figure out what you should give them to be fair and add a little extra so you are sure it isn’t unfair.
Plan what you will say and do in advance and have someone you trust with you in the meeting. Stick to your plan. Even the best hiring processes succeed only 50% of the time (though this goes beyond just a bad hire). Firing, and firing quickly is just as important as hiring well.
Also remember that if an applicant tailors to a hiring process to this extent, that’s not just a failure of the process. It’s dishonesty. Even if the applicant were competent, you absolutely don’t want to entertain someone dishonest in your company.
I've been using a Mikrotik for about 2 years now, switched from an Ubiquiti EdgeRouter X when I upgraded to 1gig at home. It works great and has been rock solid since setting it up. I even have 4 port bonding setup to my main switch because neither has SFP+.
However, it was kind of a bear to get all setup. In terms of setup difficulty it goes Mikrotik -> EdgeRouter - any consumer focused router. I've been putting off setting up VLANs for about a year and a half because I just know I'm going to break everything.
You do need to cap fountain pens because the inks are water based.
If you need to take quick notes and drying out is a problem because the pens are uncapped and capped a lot, get a Lamy 2000, vintage (not new) Parker 51 or a Pilot Vanishing Point. The first two have slip caps and hooded nibs so cap and uncap really quickly and don't dry out due to the hooded nib. The Vanishing Point also has a hooded nib but you click it like a ballpoint pen to expose and hide the nib. Also don't use quick drying inks.
If drying out is a problem because you don't write frequently enough, there are two solutions. Solution 1: Buy a Platinum (Preppy for low end, or 3776 higher end). It has a really good cap. Both will take a long time to dry out. Or even better, write more frequently.
It's extremely rare to get a nib that isn't tuned pretty much perfectly if you buy a brand new pen made by one of the Japanese big 3, Pilot, Platinum and Sailor from an actual fountain pen retailer (online or brick and mortar). I've never gotten a remotely bad nib from any of those nor have I heard of someone getting a remotely bad nib from one of them.
If you buy from eBay or used elsewhere or for that matter Amazon, you do sometimes get a bad nib because someone has tried to tune one of these nibs and botched it and then decided to try and sell the pen new.
A lot of times, with Pilots especially, if they're sold for the Japanese domestic market the tines are extremely tight and the pens write very dry, to the point of needing adjustment for writing in roman-derived script. I have a few nice Pilots and of the high end ones bought from Japan, a little bit of skipping or dryness was common out of the box.
Most nib complaints can probably be resolved with proper adjustment and maintenance. If people were doing that, I'd expect the complaints to be more focused on how the nibs defy their efforts, as opposed to how the nibs write poorly.
Totally agree. Properly restored vintage pens are absolutely awesome. You don't have to go as high end as a vintage Parker 51. Try an Esterbrook. They were the BIC of their day and are amazing.
Very important warning regarding Parker 51s or any modern Parker. Do not buy the modern Parker 51!! The brand Parker is now owned by a company that is choosing to take advantage of people who remember the Parker brand and foisting total garbage on them.
I love my Esterbrooks, I was gifted a green model J many years ago, and I use it frequently at home. I recently purchased an SJ from an antiques market, and revived it with a new ink sac ($20 all in), this pen now lives on my desk at work. I prefer their lighter weight to the Pilot Metropolitans.
The issue with Vapnik's work is that it's pretty dense and actually figuring out the Vapnik-Chervonekis (VC) dimension etc is pretty complicated, and one can develop pretty good intuition once you understand the stuff without having to actually calculate, so most people don't take the time to do the calculation. And frankly, a lot of the time, you don't need to.
There may be something I'm missing completely, but to me the fact that models continue to generalize with a huge number of parameters is not all that surprising given how much we regularize when we fit NNs. A lot of the surprise comes from the fact that people in mathematical statistics and people who do neural networks (computer scientists) don't talk to each other as much as they should.
Strongly recommend the book Statistical Learning Theory by Vapnik for more on this.