Brilliant!
Excellent ideas for surviving zoom meetings there. Must start recording some footage, and learn to play guitar
4257 publicly visible posts • joined 24 Apr 2007
I frequently rail against claims made in AI papers that if the ground truth contains a percentage of errors, any AI system trained on them is likely to end up with a similar actual error rate. I have seen people claim an increase in performance from 97.6% to 98.1% (error bars not included) on data sets where there are two ground truths, drawn up by to medics, which are at odds with each other. In our own earlier work, we managed to get a sort of pareto optimum of 92.5 ± 0.6% on both ground truths, but were in places penalised for finding blood vessels the doctors had missed. It turns out, somehow ground truth 1 has been elevated to The Ground Truth, and the other demoted to "a human observer". And now AIs are better than the poor "human observer" simply because they have been taught to copy all the mistakes the other human has made.
If ImageNet contains up to 6% error, I will continue to take all claims of 99% or better performance with a considerable pinch of salt. Furthermore, if error bars ar not included, how can they claim to be better than an earlier method if the differences are sub 1%.
I am not saying deep learning and CNNs are useless, it is just that sloppy science does them a disservice.
I did once blow up a 80287 co-processor inserted in the image processing "workstation", inserted there by our supplier. When I opened the case to inspect the damage, I noted the lettering on the chip: 80287-10, indicating it was a 10 MHz part inserted in a 12 MHz machine. Not a massive degree of unintentional over-clocking, but enough to fry the part after some time. I complained about the issue, and the company in question first stated I was wrong, after which I gave them my copy of the Intel data sheet on said component. I got an 80287-12 as replacement after that.
I actually inserted a home-made, hand soldered ISA-bus board to control both the exposure time on our cameras and the shutters on the fluorescence and phase contrast light sources on our microscopes. Switching the computer on after installing the experimental board was quite scary, but nothing shorted. Much to my surprise, it worked straight away (once my heart rate was back to something approaching normal)
I skipped the 95 and 98 editions, and went from 3.1 directly to NT workstation edition on our home machine (in part so the missus wouldn't inadvertently "clean" some mess from the root directory (like config.sys or the like)). Rather liked its stability, even though it was quite resource hungry. Mostly used SUSE Linux on the machine, however.
I remember writing code for several PCs with different Matrox frame-grabbers and image processing boards (PIP-1024A or B, and the more powerful MVP-AT/NP boards). In software, the diversity was solved by cordoning off the diversity of hardware platforms in separate libraries that were linked as needed. This worked fine in the medical microbiology lab, where I developed the code. However, once the department of dermatology had got themselves a shiny new MVP-AT/NP in their lab, and wanted to run my code, things constantly crashed, or froze. It turned out, this happened only when using a huge beast of a power supply for their mercury vapour light source for their Leitz fluorescence microscope. Bit of a bummer, as the code was intended to capture and analyse fluorescence microscopy images. The power supply produced so much RFI that the MVP-AT/NP electronics borked if you used any of the hardware image processing accelerator electronics. In microbiology we used Olympus microscopes, with a much smaller, more modern power supply, feeding the same type of mercury lamp with no issues at all. I had to write a separate MVP-AT library that didn't use any of the hardware acceleration to get the code to work in the dermatology lab.
Adobe should certainly shoulder much of the blame. After all, if this bug effects everyone on said OS, how on earth did this bug escape the testing phase? However, if your photos are so valuable, why on earth do you only store them on just one device? What if your device was lost, stolen, or broken beyond repair?
I suddenly have this image of a swarm of drones, overwhelmed by a much larger swarm of seagulls, as in the film "Birds". Spraying the drones with something like fish oil might do the trick
I'd better be going. Doffs hat (grey Tilley once more) to the late, great Alfred Hitchcock.
I see, there was "a coordinated social engineering attack", or in layman's terms: "some of our staff fell for phishing"
or should that be "some of our soon to be ex-staff fell for phishing"
It does make you wonder how sophisticated it was, how they are going to prevent this in the future, and of course how many heads will roll.
Someone recently asked me whether I was on Twitter. The answer was "no", and that doesn't look like it is going to change any time soon, not just because of privacy concerns, but I also have the El Reg Commentard section to vent my more unhinged opinions
Reminds me of the user claiming that since my last software update the mouse was consistently moving in reverse: pointer moving left as the mouse went to the right, or up as the mouse was moving down. I asked her to demo this, and noticed she was somehow holding the mouse with the wire (no wrireless in the early 90s) towards her. A quick 180 degree turn of the offending rodent solved it.
I honestly did toy with the idea of introducing a software option that would invert directionality of the mouse, preferably switching on or off at random intervals for seriously annoying users (complte with undocumented key combination that would kill that behaviour), but I thought the better of it.
This is why scientists always check each other's work, and don't take results posted by others at face value. More data are needed, as so often. Whatever the outcome, something interesting is going on. Either there is a black hole so close to earth (for a typically astronomical value of "close"), or a binary star system has been caught in the midst of a short (for a typically astronomical value of "short") evolutionary phase
of my first PC. Quite a beast it was in its day, sporting an 80386DX with Cyrix 387 floating point coprocessor, a whole 4 MB of RAM (later upgraded to 8), a graphics card with a whole further MB of RAM) and an Adaptek SCSI Controller with 88 MB disk! Cost quite a fortune at the time. Sped up my development work no end, and Windows 3.1 and MS-Office worked quite happily in 4 MB. I don't think that would do for the latest incarnations, would it now.
I know many a developer (including myself) that has yelled at his or her computer at some point in time. Indeed, yelling by itself doesn't solve the problem but:
a) it is good to blow off steam, and,
b) it did warn off users not to bother me with questions about a word processing package I never use, at that point in time, if they wanted to reach beer o'clock unscathed
Now where is that BOFH-grade cattleprod
Way back when in the days I was coding MS-DOS machines for image processing, I wrote two (harmless) programs that simply trapped interupt 9 (keyboard), discarded keyboard input, and did nothing except blink the screen and say the computer has crashed. Only a hard reset worked. I called these two variant HANG.EXE and CRASH.EXE. Everyone had to try them at least omce
I have a 64-core Opteron compute server at work (512 GB RAM is also nice), and tried make -j 64 on a big code base and was very impressed to see that fly. In practice make -j 32 generally maxes out the speed, simply due to IO limitations, but it is fun to watch things compile and install really fast. The machine is getting old, so I hope to weedle out funds for a replacement, and maybe while I am at it try to slip a 32 core desktop machine into the budget. Maybe I should connect to my inner BOFH to make this happen
I once observed an Italian (from Naples, too) who was confronted with the Chicago style deep pan pizza. An explosion resulted, not quite Trinity Test level, but getting there (hence icon).
Granted, several Neapolitans I have met tended to explode at any pizza NOT from what they consider the only authentic pizzeria in Naples (and they tend not to agree on what the ONLY authentic place is), but this person really got incandescent.
I tend not to be too bothered with what is authentic, food wise, just what tastes good. My preference is definitely with the Italian thinner base, but I refuse to get all religious about it.
Recently saw several of his swarms fly by. Quite apart from being annoying to astronomers, crowding LEO with loads of little satellites make the issues of space debris much worse, as the likelihood of collisions increases rapidly. Fewer, bigger satellites are much easier to manage (also not trivial) than swarms of little ones. Space may be big (you might think it's a long way to the chemist, but that's peanuts to space), but we only have a little speck of it, surrounding our pale blue dot, at our disposal.
I almost expect hardware manufacturers use this exploit as an excuse to for the use of glue in their laptops. Cheap cutting of corners during manufacture? Mais non!! We were thinking about security!!! Honest!!!!
Oh dear. I feel an extra exclamation mark coming up!!!!!
I'd better be going. The one with "Maskerade" in the pocket, please.
to build a little ultrasound broadcasting device, that emits all sorts of random signals just to screw this kind of eavesdropping up. A smart system could even listen for covert ultrasound broadcasts, and either try noise cancelling techniques on it, or (much simpler) do the ultrasound equivalent of Brian Blessed bawling "Blood! Death! War! Rumpy pumpy!!"
Electronic countermeasures, if you like. Icon, well, because of Brian Blessed, of course
Now where is that Arduino kit?