As far as seeing and ignoring fixed objects, you can also remove any returns that have a near-zero velocity in radar and focus only on those objects that are moving.
Of course, indoor settings have a lot of non-stationary objects as well that might not be targets of interest to you, like fans, curtains blowing in the breeze, etc. So you can also develop algorithms to remove those signatures too.
Seeing fixed objects can be beneficial as well, for example, if you have a sensor deployed in a room but you don't know a priori what the room looks like. Longitudinal results and long range statistics can take you pretty far in seeing the room extents and layout and furniture, etc. Though a lidar sweep is better if you can get it
Oh Jua, it's cool to see that name. They've been on my radar since I applied to (and was rejected by) them earlier this year. I'll be interested to see the research paper once it is published next week, especially since they claim their model surpasses Aurora and Graphcast
I was about to say "they still torture students this way" but stopped myself when I remembered I took Circuits 1 and 2 back in 2007. So maybe my knowledge is dated too...
It's a weird butterfly effect moment in my career though. I had an awesome professor for circuits 1, and ended up switching majors to EE after that. Then got two more degrees on top of the bachelor's
Very cool project. There are some presentations by the PI on youtube that I recommend searching for. One of the interesting takeaways I had was that they were able to do better with mesoscale phenomena and extreme weather prediction than the other players (like Graphcast and Pangu and FourCastNet), in part due to their technique for training a higher resolution data space (0.1 deg vs 0.25 or 0.5). I also found it interesting that they were able to show a scaling relationship where performance increased by 5% every time they doubled the model size - and their loss was still improving when they had to cut it off due to cost constraints.
I love the creativity that goes into naming these projects in the geosciences! I've been a part of several of these projects myself, and have used data and collaborated with teams from many more.
One point of clarification: GNSS is a term that has broader application than you describe, as it encompasses constellations from other countries and political associations as well. For example:
* Galileo - European Union's GNSS system, named after the astronomer
* BeiDou - China's GNSS system
* GLONASS - Russia's GNSS system
* JAXA - Japan's GNSS system
One backronym that I liked from my time doing my PhD was RELAMPAGO, which is a Spanish word for "lightning," but which some group of scientists gave this definition: "Remote sensing of Electrification, Lightning, And Mesoscale/microscale Processes with Adaptive Ground Observations". It was a very cool campaign that produced a ton of amazing data, and catalyzed many dissertations (including one of my close friend's).
I've seen other research and discussion on this topic. Some stats that may be validating for you (and others) to hear:
* There's a 0.08% job application -> offer rate when applying through LinkedIn (LI). An average of 1 in 1,250 applications lead to an offer
* The linked paper on this post finds that 21% of postings are ghost jobs, but I've seen credible estimates that the proportion is as high as 50%
* A Stanford survey found hundreds of fake LI profiles, AI-generated "recruiters" that are interacting with candidates and posting ghost jobs on behalf of big companies
* ~75% of resumes from qualified applicants are never seen by a human
* resumes get on average 6 to 8 seconds of consideration when they are reviewed by a human
* 300,000 jobs are outsourced annually (with respect to the US)
All this to say, you're right, something is fundamentally broken in the labor market, especially the tech labor market. And not that many people are talking about it, except for those of us who have been unfortunate enough to need to look for jobs in the past ~2 years.
In my own case, my previous employer (a startup) ran out of money and laid everyone off last Fall. I was fortunate enough to find a new position, but this job search was the hardest I've faced since 2008 - and it seems worse now than it was this time last year.
I once applied to a position like this. It was eerily similar to my background, and when I did a little digging I found that the group lead had even directly cited my research papers in his own research work.
I applied on the site, reached out on LinkedIn to the group lead and the recruiter, and even was able to find emails for those two, which I also messaged as well.
They didn't even bother to send me an automated rejection notice. There was nothing at all, no responses to any messages, no email, nothing. I have to assume that position was posted with someone already in mind that they wanted to hire.
I don't even mind paying up front anymore. I'm in a position where I can afford it now, though for most of my adult life I've relied on sites like this to make ends meet.
However the only thing I want from publishers is DRM-free e-books (same for music). If you offer a way for me to actually own the digital property I'm buying, I'm going to buy it. If you make it hard or impossible to transfer between my devices, or share with my wife and kids (i.e. how physical media works), you're not getting my money and I'll find another way to get the book.
Yeah for sure. So say your category is "fruits and vegetables". You might say Avocado, Banana, Celery, etc. The goal is to challenge yourself, and it's more of a stretch than you'd think (but a good exercise). Or at least, it was a stretch for me especially at the beginning
This happened to me following a concussion. I worked with a therapist who suggested a "game" that could help my word recall. Every night before sleeping, pick a category and try to name an example from the category for each letter of the alphabet. Choose a different category every night so you're always challenging yourself.
Of course there's no substitute for working with a PT who specializes in post TBI recovery.
I've gained a lot of my recall back (though not at 100%).
I wouldn't read too much into it. UK is one of my alma maters. Everyone in that area of the US means "University of Kentucky" when they say "UK". It isn't a dig at the United Kingdom nor is it (I assume) an attempt to gain undue credibility by associating with the country. For the people there, UK as the University is simply the first order association for that acronym, rather than what is to them a faraway country that has no bearing on their day-to-day lives.
Wi-Fi or another radar-type device (like the one we developed) is the way to go. That got around people not wanting, or simply forgetting, to wear a device, but it did have its own technical challenges.
A sensor fusion approach is a great option, and wearables + some radar-based system seems like a best of all worlds solution, if people will use the wearables. Another big bonus you get from wearables is vitals detection, although I implemented a couple over-the-air vitals detection algorithms with our radar device that were sometimes very reliable.
I've thought about the cultural change aspect too and I agree with you. Younger generations are more used to wearing smart watches and carrying phones, so the appetite for wearable fall and activity detection devices will be much higher in future.
So this is an area where I can speak from experience. I was previously employed as the ML research scientist for a startup that designed and implemented an mmwave radar-based solution for the use case of seniors in independent living situations. That company fell apart for reasons unrelated to the technical side, unfortunately.
What we found is that seniors had essentially no desire to wear any sensors, which rules out the wearable inertial sensors mentioned in the paper. Also as others have mentioned, a sensor that captures visual images is a nonstarter due to privacy protections on a regulatory level but also privacy concerns on a personal level.
I'd add one other set of challenges that is unfortunately never covered in the academic literature - non-ideal rooms for monitoring signals. The papers show empty conference rooms with line-of-sight between sensors and people, but real settings are much messier. Not only is there furniture to block or distort signals, but also many sources of noise like fans, metal objects, open windows (which cause breezes to move curtains and other objects), pets, visitors, etc. Not to mention the unique room configurations for every person. We overcame several of these challenges but didn't develop perfect answers for many of them.
It was a fun position that gave me a weirdly specific set of knowledge that isn't always transferable on a technical level but was still great, and I wish it could have lasted longer. I'd be happy to share more info though if you're curious!
Of course, indoor settings have a lot of non-stationary objects as well that might not be targets of interest to you, like fans, curtains blowing in the breeze, etc. So you can also develop algorithms to remove those signatures too.
Seeing fixed objects can be beneficial as well, for example, if you have a sensor deployed in a room but you don't know a priori what the room looks like. Longitudinal results and long range statistics can take you pretty far in seeing the room extents and layout and furniture, etc. Though a lidar sweep is better if you can get it