HNHacker News
TopNewBestAskShowJobs

kevinalexbrown

4,506 karma · joined August 28, 2011

Reach me here: kevinalexbrown@gmail.com

Yes, if

submissionscomments
kevinalexbrown··on Deep learning algorithm diagnoses skin cancer as well as seasoned dermatologists
It was an image segmentation task, and the features were similar across data sets. The other thing that made it work well was heavy use of data augmentation that captured ways in which different data points could reasonably differ.

There was a really cool medical imaging paper recently that literally just labeled several 2D slices in a 3D dataset consisting of 3 images and performed a reasonable segmentation:

https://arxiv.org/abs/1606.06650

kevinalexbrown··on Deep learning algorithm diagnoses skin cancer as well as seasoned dermatologists
I suspect one reason is the schleppiness of implementation. Healthcare is hard enough to navigate as a doctor or patient.

Another is the existence of edge cases. While not quite as bad as self-driving cars, they still require human review.

One radiologist I spoke with was under the impression that his field was going to disappear in the next n years. I don't think that will be the case entirely, but it will change.

One reason why image detection is ripe for this kind of "disruption" is that some of the highest paid medical fields (dermatology, radiology) basically employ expert image recognizers.

On the other hand, I'm not sure what percentage of health care costs goes toward employing doctors directly, but I don't think it's huge compared with the rest of the facility and equipment and administration cost.

The real benefit will come when people don't have to go in to the doctor at all. Which makes the smartphone "take a picture" aspect pretty sweet.

kevinalexbrown··on Deep learning algorithm diagnoses skin cancer as well as seasoned dermatologists
One major, major advantage that medical imaging has for deep learning is the similarity of each data point, especially the 'background data.' For instance, human brains typically look very similar across individuals (up to scanning parameter differences), except in the abnormalities - which are often precisely what you want to highlight.

As an example, I recently trained a neural neural network to perform a useful task for our lab using 3 (!) hand-labeled brains.

kevinalexbrown··on Deep learning algorithm diagnoses skin cancer as well as seasoned dermatologists
Before neural networks got deep, there was a lot of very impressive work applying neural networks to medicine. Example:

http://suzukilab.uchicago.edu/research.htm

IIRC they were outperforming the average radiologist on some tasks 10 years ago.

kevinalexbrown··on CMU's Libratus builds substantial lead in Brains vs. AI competition
> Making all this computerized seems to take the spirit of poker away. It is like playing online poker which is an entirely different ball game vs. real poker.

Maybe, but it seems like impressive performance nonetheless.

kevinalexbrown··on Brain Sodium Linked to Cognitive Deficits in MS
> right now all they're saying is that there's a correlation in a study of less than 60 patients

Sort-of. Many recent studies had already suggested a role of sodium accumulation in brain tissue loss in MS. Previous work had also shown significant sodium accumulation in regions where there was no apparent cell death, suggesting that cell-death does not drive measured sodium concentrations.

What's cool about this result is 1) that sodium measurements from the brain are better indicators of cognitive defects than actual death of brain cells. 2) Abnormal sodium accumulation occurred in neocortex, which is the first place you'd suspect cognitive impairment to originate. That's a correlation across patients, and brain areas, suggesting at the least that NaMRI is a good tool for early MS detection.

I doubt it's dietary sodium. Likely, sodium accumulation occurs when normal processes are disrupted. I wouldn't expect increased sodium in your diet to lead directly to increased sodium in selective areas of the brain (could be wrong here).

kevinalexbrown··on Information Geometry
Much of this is due to Amari, who got really into merging theoretical neuroscience and IG, e.g. http://www.mitpressjournals.org/doi/abs/10.1162/089976602602...
kevinalexbrown··on Advanced Data Analysis from an Elementary Point of View
This book is a great resource. I've gone back to it frequently. If you're interested in getting into statistics and machine learning, I cannot recommend this book highly enough. The motivations are good, and the explanations are good. Shalizi's class for this book at CMU is highly sought after, and some of that surely makes its way into this massive book.
kevinalexbrown··on Uber Self-Driving Truck Packed with Budweiser Makes First Delivery in Colorado
What if, say, the trucking industry loses 50% of its jobs/salary in 5 years? That's a lot of displacement even in an industry with high turnover (I don't know if those numbers are realistic, but I do think there will be faster and more massive displacement than self-serve checkout and automatic toll roads).

What I'm curious about is whether whatever industries truckers are heading to now can accommodate that many more workers without problems. Just because there's a high turnover doesn't necessarily mean that wherever truckers are heading now could handle 10x the influx of employees without funny consequences.

kevinalexbrown··on Uber Self-Driving Truck Packed with Budweiser Makes First Delivery in Colorado
If you read people's comments carefully, I think you'll find that almost no one is proposing to stall self-driving trucks for the sake of jobs.

But it does raise questions about worker displacement that are very interesting, and not just from a social-justice-warrior brigading perspective. I imagine this will happen much faster than the textile industry changed. Am I wrong there? That alone makes it fascinating.

kevinalexbrown··on Uber Self-Driving Truck Packed with Budweiser Makes First Delivery in Colorado
The trucking industry is huge (say 1.5 million truckers alone, plus rest stops etc), and it will largely have disappeared as a result of this. Uber may want drivers in the cab now, but when it's reliable enough, I imagine it makes sense to just have drop-off points for the last few miles.

In some sense it will be among the biggest economic transformations in recent history. There have been far bigger ones, but I can't think of one that has happened in the short timeframe expected of self-driving trucks. I don't think this point is controversial.

At the same time, I've been very curious to see the overall economic impact, especially with displacing such a huge percentage of the US workforce. Interestingly, there's already a shortage in truckers per wikipedia. I found this fascinating: https://en.wikipedia.org/wiki/Trucking_industry_in_the_Unite...

I imagine that people going into trucking see the writing on the wall. It will put a lot of truckers out of business, but the turnover is already crazy high, with many truckers already leaving the business entirely.

This could be a very interesting case study on how to manage worker displacement effectively. I don't think anyone here is seriously advocating paying more money and lives for goods just to keep trucking jobs up (although more and more on HN one has to write 'defensively' to curb misinterpretation, i.e. I feel compelled to shout that I think this is a good idea). But it would be interesting to see where truckers are getting new jobs already, and whether those industries could use more. Are truckers leaving the industry going back into some other industrial job, back to school, unemployment?

kevinalexbrown··on Bryan Johnson invests $100M in Kernel to unlock the power of the human brain
Throwing money at neuroscience hasn't been so bad, has it?
kevinalexbrown··on Challenges for Artificial Intelligence in Medicine
> Ct scans aren't really used to look for brain tumors.

I can see how my phrasing was confusing, but I didn't mean to suggest that Ct scans are used to look for brain tumors. My work segmenting brain scans was not tumor-related, just gray/white matter segmentation.

kevinalexbrown··on Challenges for Artificial Intelligence in Medicine
I've used deep learning for segmenting brain anatomical scans, and I worked in a lab that used neural networks to detect cancerous tumors.

I suspect the first major hospital-facing implementations of machine learning will be in radiology, e.g.: http://suzukilab.uchicago.edu/, which has been diagnosing cancerous tumors in CT scans with neural networks since before it was cool (one reason you won't see the terms 'deep learning' in the literature since they were originally just 3-layer networks, before the term was even coined). IIRC it outperformed the average radiologist.

I wonder if the label problem could be less difficult for some low-hanging fruit. The CT scan neural network required something like 40k labeled scans from a radiologist, but it could come for free: many yes/no disease detections will eventually be resolved by human labeling anyway by your doctor. If you had access, say, to every CT scan taken, and electronic health records for the patients, your labeling is noisy and biased but at least at massive scale. The problem is (legitimately) restricted access to health records in the US. Maybe some European countries have better data access?

And the implementation problem will eventually disappear. I remember talking with a radiologist years ago, who remarked "some people in my field have no idea it's about to disappear". I'm not so sure there will be no more radiologists, but their role will definitely change. Hospitals would be okay with this, actually, since radiologists are expensive. Eventually radiology scans will probably be like ordering blood tests, where fewer and fewer MD's are required.

kevinalexbrown··on The basic neurobiology behind the 12-dot illusion
I don't think that explains it, either: Even in low contrast, the illusion fails w/o the lines, as far as I can tell.
kevinalexbrown··on The basic neurobiology behind the 12-dot illusion
I'm not sure the central argument fully explains the illusion.

If you can simultaneously see several stars in the sky using your peripheral vision using averted gaze, why not several dots?

I suspect uncertainty plays a part, but image completion from higher-order feedback that complete the lines might drive the illusion more. Put another way, I believe if you remove the gray lines, the illusion ceases to work.

kevinalexbrown··on Scientists Watch Thoughts Form in the Brain
You don't need to genetically modify an organism at birth to use optogenetics.
kevinalexbrown··on Scientists Watch Thoughts Form in the Brain
strictly speaking, this is calcium imaging
kevinalexbrown··on Breakthrough solar cell captures CO2 and sunlight, produces burnable fuel
Is it excessive? It certainly doesn't seem to be BS. It's published in a reputable journal (which doesn't make it true). It discusses several of the questions raised by Animats directly (though I wouldn't say it answers them conclusively).

I do agree that the university press release isn't going to critically evaluate the work of its own researchers. That's too bad, perhaps, but it hardly renders the original result uninteresting.

Artificial photosynthesis may be old, but the advances in this work appear aimed at overcoming precisely the problems Animats raised, like catalyst preservation and efficiency.

Put another way, just because a university gushes about the work of its faculty doesn't mean the faculty's work is crap.

kevinalexbrown··on Breakthrough solar cell captures CO2 and sunlight, produces burnable fuel
The press release at least attempted to answer some of your 'usual questions'.
kevinalexbrown··on Breakthrough solar cell captures CO2 and sunlight, produces burnable fuel
It's actually a major approach under the name biomass. Despite releasing carbon when wood is burned, the tree has taken in significant CO2 over the course of its life. In addition, decomposing wood byproducts are going to release carbon anyway as they decompose, I believe. Places with excess wood byproduct can generate power onsite.

It's not going to replace other power sources, but it is commonly done.

kevinalexbrown··on Lumosity fined $2M for deceptive ads of “Brain Training” app
One intuitive way to categorize products is according to the observability of the value. In one class, it's readily apparent: the storage is cheaper, it doesn't fail very often, there's little downtime, my payments always go through, the user experience just feels better, all my friends use it. This doesn't have to be objective, just 'noticeable'.

On the other hand, it's hard to judge the value of an encryption app, or a brain training program, because the true value is often hidden. It's difficult to tell, at least for the average consumer whether an encryption app works. There might be signs for the more immediately knowledgeable ('why won't the passwords take special characters, or passwords longer than 12 characters?'), but it's usually a small group.

Unfortunately, this is probably going to make it harder for the latter kind of company to grow quickly, because trust takes time to develop. I don't just need to trust that my data will be safe - the entire value of the product depends on my trust that it does what it says.

I have no idea if lumosity does what it claimed, but I doubt I could easily tell in a day or two. Contrast that with instagram or a better computer - the value (however large or small) is obvious.

kevinalexbrown··on A Quasipolynomial Time Algorithm for Graph Isomorphism: The Details
I sat in on a couple of those classes. At the beginning of the class there were 50-100 people registered. At the end, there were 6 (including graduate students).
kevinalexbrown··on Different Brain Regions Are Infected with Fungi in Alzheimer’s Disease
totally - but if you can effectively treat a substantial percentage, that's huge. And some effective antifungals have what many consider tolerable in the face of AD.
kevinalexbrown··on Different Brain Regions Are Infected with Fungi in Alzheimer’s Disease
First, there are antifungal drugs that pass the BBB, with varying efficacy:

https://www.medicine.wisc.edu/sites/default/files/domfiles/i...

Second, antifungals have positively affected AD-like symptoms, though possibly through misdiagnosis.

kevinalexbrown··on Volkswagen's U.S. head: individuals engineered emissions cheating
"This was a couple of software engineers who put this in for whatever reason," Horn said

Except when obviously true, companies that make public statements like this will have difficulty recruiting the best engineers.

kevinalexbrown··on YC Research
To be honest, the freedom to fail alone would make a huge difference. On one hand there's a lot of freedom for intellectual growth in academia - in the last few months I've used deep learning, assembled hardware under a microscope, performed neurosurgery and genetic engineering (this is somewhat standard for my field). On the other hand, I make less than I did as a new cable guy, won't make more for 10 years, and every postdoc and phd student in my lab works 7 days / week standard. And if I fail, I will probably be done career-wise. LOTS of talented people leave because they see startups as the 'safer' bet - at least if your startup fails you're not doomed.
kevinalexbrown··on Trans-Pacific Partnership Trade Deal Is Reached
The fact that the negotiations were done in secret probably means that most of the TTP content is being given to journalists by those officially authorized to speak about it. This isn't unexpected, but it does affect how the TTP is framed (even if you don't buy the Greenwald puff-piece-for-access argument).
kevinalexbrown··on “This mistake will become the thing you are best known for”
Fwiw, the increased scrutiny that goes with death penalty cases raises uncomfortable questions about the justice system in general.

Would this case be acceptable if Glossip had only received a life sentence? I'm not sure either way. But it does seem that the justice system should have more accommodation for uncertainty than a complicated appeals process that seems to focus on procedural issues.

What form would an uncertainty-conscious justice system take? Economists might suggest that you can account for uncertainty merely by lowering punishment accordingly to account for "risk," (e.g. we will lower punishment by X percent because we're only Y percent sure you are guilty) but that doesn't seem to work here. On the other hand we clearly can't have a justice system that never acts for fear of wrongful incarceration.

kevinalexbrown··on If Materialism Is True, the United States Is Probably Conscious (2014)
Anesthesia practically eliminates consciousness yet the brain states are probably more similar than brains and the United States. I'm not sure I agree that apparent similarity is good enough to conclude that the US is probably conscious under a materialist framework.

Definitely possible, however. And fun to consider whole galaxies exhibiting consciousness.

← PreviousPage 2 of 10Next →