HNHacker News
TopNewBestAskShowJobs

jhartmann

1,340 karma · joined August 10, 2012

submissionscomments
jhartmann··on Rapid recovery from major depression using magnesium treatment (2006)
This link discusses the paper in question: https://www.psychologytoday.com/blog/evolutionary-psychiatry...
jhartmann··on Microsoft releases CNTK, its open source deep learning toolkit, on GitHub
Thanks for this, I personally have a license for MKL due to a licensed Intel C++ compiler and it really does a good job. I feel that Intel should do more to get MKL out in the open and more in use by the Open Source community. For instance I would love to have access to their FFT code so I could optimize it for my specific use case, but right now things are a black box. Very excited about CNTK btw, releasing a fully cooked multi machine distributed framework is just awesome.
jhartmann··on If being too clean makes us sick, why isn’t getting dirty the solution?
I have an autoimmune disease (Ankylosing Spondylitis) and I've been pescatarian for a little more than a year. I'm not on any biologics, my sed rate is pretty normal and I feel SO much better. Part of my diet change is that I eat a large amount of fiber (40+ grams a day generally), which is very prebiotic. I would say it really has helped me, its probably worth trying a diet change if you have an autoimmune condition.
jhartmann··on French drug trial leaves six hospitalized, one brain-dead
Some information about the trial seems to be here: http://www.breizh-info.com/2016/01/15/rennes-biotrial-biales...
jhartmann··on Writing high-performance servers in modern C++
A lot of people state that Python/PHP/Ruby/Javascript on the dynamic language side and Java and other JVM languages on the statically typed side and much quicker to develop than C++, but being a very experienced Java programmer and a decent Javascript programmer I find that I can program just as fast with C++ as long as you use the right choices of libraries and use C++11 or greater. With libraries like Boost, fbfolly, wangle, etc. I find it can be just as quick to develop C++ code as with other languages. I think too many people think C++ is a huge investment above other languages due to past experience, but I feel that it is just not true anymore. Thanks for this post, I think it shows just how quick it is to write performant modern C++, when you stand on the shoulders of other libraries. Just like the other languages.
jhartmann··on AMD GPUOpen Initiative, New Compiler, Drivers and OS SDK's for Linux and HPC
OS here is open source, this is pretty significant since they plan to offer MIT licensed versions of lots of their internal tools and SDK's. The C++ compiler is an attempt at capturing the HPC market by providing tools to convert 90% of CUDA code to portable C++ that can run on AMD and NVidia cards. Lots of bare metal GPU access stuff too, so lots more control for Game developers. Lots of goodies.
jhartmann··on Experimental drug targeting Alzheimer's disease shows anti-aging effects
Also interesting links about this compounds safety and effect as a treatment to Alzheimer's disease that was previously published (http://www.alzres.com/content/5/3/25)

Summary from that paper: J147 is an exciting new compound that is extremely potent, safe in animal studies and orally active. J147 is a potential AD therapeutic due to its ability to provide immediate cognition benefits, and it also has the potential to halt and perhaps reverse disease progression in symptomatic animals as demonstrated in these studies.

--- The new paper is at: http://www.impactaging.com/papers/v7/n11/pdf/100838.pdf

Abstract: Because age is the greatest risk factor for sporadic Alzheimer’s disease (AD), phenotypic screens based upon old age‐associated brain toxicities were used to develop the potent neurotrophic drug J147. Since certain aspects of aging may be primary cause of AD, we hypothesized that J147 would be effective against AD‐associated pathology in rapidly aging SAMP8 mice and could be used to identify some of the molecular contributions of aging to AD. An inclusive and integrative multiomics approach was used to investigate protein and gene expression, metabolite levels, and cognition in old and young SAMP8 mice. J147 reduced cognitive deficits in old SAMP8 mice, while restoring multiple molecular markers associated with human AD, vascular pathology, impaired synaptic function, and inflammation to those approaching the young phenotype. The extensive assays used in this study identified a subset of molecular changes associated with aging that may be necessary for the development of AD.

jhartmann··on DMTK: Distributed Machine Learning Toolkit from Microsoft
This is very interesting. I was very surprised that Tensorflow did not use a central parameter server, and this looks like a good foundational parameter server to build something like DistBelief on top of from Microsoft. I wonder if they have a Deep Learning module that they will open source eventually. Microsoft has a few tricks they used in some of their recent papers and it would be interesting to see them in a production quality system.
jhartmann··on Drug engineered from bananas shows promise in fighting deadly viruses
I really think this is very interesting in that this would be a truly broad spectrum anti-viral agent. Maybe this is the start of a class of compounds like antibiotics, but for treating viral infections. Could be amazingly tranformative of our ability to treat disease.
jhartmann··on Obstacles on the Path to AI
The error in backprop is a vector quantity, not a single scalar for each time step. In RL the goal is to optimize the overall sum of the reward over all time steps. Backprop attempts to minimize the magnitude of the error for a given loss function, by moving in the negative direction of the gradient of the function. Backprop just moves a lot more variables to a desired outcome, that is what LeCun is saying. The representational power of a single scalar 'score' doesn't have much ability to optimize such a large n-dimensional function in any efficient way.
jhartmann··on Joint Allocations in C++
There are prebuilt slab allocators that do this well. I used the arena allocator in facebook's libfolly recently as the base for a huge page allocator, and the code turned out quite nice. I think the allocator is a much better abstraction here. You could use something like the arena allocator here to carve out the memory for all the vertex and mesh storage and amortize the cost of hundreds of objects as well, so I don't like the argument in the first few sentences of the article actually that allocators would take more space.
jhartmann··on Keeping older muscles strong
Interesting page from the company that was founded by one of the researchers as well: http://www.emmyon.com/research/ Looks like they have linked AF4 transcription with age related muscle atrophy and have discovered two small molecules from Green Tomatos and Apple peal that inhibit AF4 activity and reverse age related atrophy in mice. Very exciting.
jhartmann··on Voltage-Gated Sodium Channels: Therapeutic Targets for Pain (2009) [pdf]
I'm curious, what is the name of the potential compound? I have to take opioids for chronic pain due to Ankylosing Spondylitis and would love to stay informed on potential alternatives.
jhartmann··on YC Fellowship
I'm glad you will be slightly more open to solo founders. I also like that you are open to a more remote experience.

I think this project could be very helpful in my situation, I have a great idea that is progressing (Deep Learning/AI Image Recognition SaaS) but would have a hard time putting together a suitable team that could survive with our current obligations in San Francisco (families, house payments, etc.) Maybe getting myself working on it full time could get it to the point where it could sustain itself enough that YC proper could work out financially for a bigger team. Very excited, I'm definitely going to apply.

jhartmann··on Skunk Works Reveals Compact Fusion Reactor Details (2014)
nextbigfuture.com is pretty good in that regard, a little political sometimes and sometimes some things from the edges of science but the admin does a good job on tracking news for emerging tech.
jhartmann··on Diet that mimics fasting appears to slow aging
There was a paper published last year by this group where they did a study of Cancer patients (and a bigger study in mice) receiving a particular kind of Chemo, and their immune systems post therapy were basically 'rebooted'. The protocol in that work was a periodic 2 day water fast in 6 cycles timed with the Chemotherapy adminstration. They then did a very small pilot study and saw the reversal of immunosenescence in people who did the cyclic fasting without Chemotherapy. It was a reasonable effect, but the human studies were smallish and probably could use more reproduction.
jhartmann··on Diet that mimics fasting appears to slow aging
No, they did a super small random pilot study. Here is the section from the actual paper on the results:

In the FMD subjects, fasting blood glucose levels were reduced by 11.3% ± 2.3% (p < 0.001; FMD) and remained 5.9% ± 2.1% lower than baseline levels after resuming the normal diet following the third FMD cycle (p < 0.05; Figure 6B). Serum ketone bodies increased 3.7-fold at the end of the FMD regimen (p < 0.001) and returned to baseline levels following normal food intake (Figure 6C). Circulating IGF-1 was reduced by 24% by the end of the FMD period (p < 0.001) and remained 15% lower after resuming the normal diet (p < 0.01; Figure 6D). IGFBP-1 was increased 1.5-fold at the end of the FMD regimen (p < 0.01) and returned to baseline levels following normal food intake (Figure 6E). These results indicate that the FMD group was highly compliant and generally did not consume foods not included in the FMD box provided to them.

jhartmann··on Diet that mimics fasting appears to slow aging
This builds on some earlier work where the Authors investigated Intermittent water fasting on Chemotherapy recipients and found that their immune systems regenerated better. This paper appears to have found a way to mimic some of the effects on their previous protocol with something that is safer to do. The earlier paper is here: http://www.cell.com/cell-stem-cell/abstract/S1934-5909(14)00...
jhartmann··on Diet that mimics fasting appears to slow aging
The paper in Cell Metabolism is here: http://www.cell.com/cell-metabolism/abstract/S1550-4131(15)0...
jhartmann··on What's Wrong with Deep Learning?
Actually, state of the art speech recognition has switched over to having a Recursive Neural Network directly run over the audio input. Take a look at the paper at http://arxiv.org/abs/1412.5567 and http://usa.baidu.com/deep-speech-lessons-from-deep-learning/

However combining learning features with other systems is a very powerful approach and combining SVM's on top of the learned features of a Neural Network I would say is common. I personally am more interested in approaches like Deep Fried Convnets (http://arxiv.org/abs/1412.7149) that combine kernel methods as part of the Neural Networks themselves.

jhartmann··on What's Wrong with Deep Learning?
It is always impressive to me how both Professor LeCun, Bengio & Hinton stuck to their guns and worked on these problems while others were not so interested. I'm very glad Hinton's group at DNN Research was able to blow away the competition in the Imagenet challenge. Now many more people are working on these ideas and some really amazing things have been accomplished in a very short time. I can't wait to see where we are in five years, and I love that LeCun's points out the areas where we should focus and what we are not good at yet.
jhartmann··on Spectral Representations for Convolutional Neural Networks [pdf]
Reading through this paper, this is a logical extension of the recent work published by FAIR (Facebook AI Research) that proved that with careful implementation of FFT convolution that the speed up is very significant. The Facebook work though still had all the learning happening in the spatial not frequency domain. They are doing all the updating in the frequency domain, and have introduced a new type of pooling that uses stochastic resolution reduction in the frequency domain that seems very useful. Very interesting paper, I'm keen to try out the techniques myself.
jhartmann··on No One Knows What to Do with Fukushima’s Endless Tanks of Radioactive Water
Here is a good summary of the stats from various sources. Death rate per terawatt/hr is 0.04 for Nuclear and 0.44 for Rooftop Solar. Basically Rooftop solar is a full order of magnitude more dangerous. Coal is WAY more nasty at a world average of around 100 deaths per terawatt/hr if you consider heating/cooking/electricity or 60 for just electricity. Those numbers for China are 170 & 90 respectively. If you include the worst hydro accident, Hydro is 1.4 deaths per terawatt/hr, or 0.10 if that is excluded.

Nuclear is the best game in town, even though the accidents are well publicized and the marketing for alternatives have done a really good job of making us scared of the nasty radioactivity, nuclear has the best safety culture and safety record.

http://nextbigfuture.com/2011/03/deaths-per-twh-by-energy-so...

jhartmann··on The Musk Who Wants to Change The Way We Eat
I love the following comment 'Musk won’t rule out GMOs permanently, but he feels that currently they are used in a way that encourages the use of pesticides and other chemicals'. I wish more people had this common sense view of GMO's. Genetically modifying things is not itself bad, but modifying things so we can spray crap tons of chemicals on them is probably not a very swift idea.
jhartmann··on In-Memory Computing at Scale: When to Choose and How to Effectively Use JEMalloc
This paper was a good read. I recently have done some work using libhugetlbfs to use huge pages for an application and you might want to consider using an Arena backed by huge pages. For very large structures TLB misses can really add up, and huge pages can improve performance by a large margin for workloads where lots of large buffers are being accessed and scanned. If tests suggest that you should go this route on your workload, I would suggest managing the pages using libhugetlbfs directly instead of transparent huge pages.
jhartmann··on I signed up to write apps for my Amazon Echo, Amazon sent me an NDA
It is not just a sell. I build Neural Networks as part of some work I am doing, and the kinds of multi-layer deep networks that do the recognition just don't perform well on small devices. Also the best systems often have many different trained networks and specialized routing to find the best network for your accent, cadence, noise level, etc. Having all of these weights on your local device, running all the tests, running through any of the required preprocessing, and asking the network what the answer is, converting the results, etc. is going to be too much for your phone or low powered device. While there might be other reasons for them to send data to the cloud, to get the accuracy you really need beefy servers (potentially with GPU's or something more exotic like FPGA's), and potentially multiple servers working in concert.
jhartmann··on Hundreds of Thousands of Google Apps Domains’ Private WHOIS Leaked
I had some domains that fell under this bug. Google has paid my registration on the domains for an additional year, more than the price of the privacy service.
jhartmann··on While other VCs seek Unicorns, Indie.vc is all about cashflow
Bryce, This seems very interesting. I have been working on a Neural Network SaaS startup for awhile now, and I think my plans could align with this program. I just wanted to say thanks to you and OATV for trying this experiment. I think traditional VC's miss out on lots of opportunities where there is some consulting revenue early on in the company lifecycle to help build the business. I have high hopes that INDIC.vc and programs like it could help those sorts of businesses get going.
jhartmann··on The Face Detection Algorithm Set to Revolutionize Image Search
There is also a paper published with technical details at http://arxiv.org/abs/1502.02766. They are using Spatial Pyramid Pooling like the Microsoft Deep Learning group out of China combined with a sliding window evaluation over the input images. Looks like we finally have an approach that does well on occluded and rotated faces. That is something that Viola-Jones does quite poorly, so its quite an exciting advance.
jhartmann··on Move Evaluation in Go Using Deep Convolutional Neural Networks
This is very interesting, they have trained a 12 layer network to play Go, and that does quite well by itself. They also have built an asynchronous Monte-Carlo search + Neural Network implementation as a proof of concept. More awesome stuff out of the Google DeepMind and Google Brain groups.
← PreviousPage 2 of 5Next →