HamSession

You would get tit for tat and US would tax any offshore development products.

A lot is caused by the forwarding services that mask the numbers. Really each should be held liable for any damages.

I've talked to some neuroscience friends and was interested to learn that lucid dreaming hasn't been studied much if at all in lab conditions. I welcome such research and hope more people get inspired to study deeper.

Used Zotero with Firefox plugin and word plugin. Used to use Mendeley, but like the open source nature of Zotero.

I wouldn't put much faith into the non ldl hypothesis we have extensive studies that show that lower cholesterol (total and ldl) cause less myocardial infarctions.

[1] https://www.nejm.org/doi/pdf/10.1056/NEJMoa1405386 People with a gene that naturally produces lower ldl levels lead to less Hard CD events.

[2] https://www.thelancet.com/journals/lancet/article/PIIS0140-6... Statins cause less heart disease and strokes even more than originally thought.

[3] https://www.thelancet.com/journals/lancet/article/PIIS0140-6... Even if people with low risk factors cutting LDL with stains causes lower number of events

There are a lot of factors that can go into heart disease. Some of these factors are:

1. Under/over working out, especially for white men over working out can lead to increased shear stress on the arteries

2. Genetic familia hypercholesterolemia aka high cholesterol. This is caught by a blood test and fixed with statins.

3. High Blood Pressure again exams and medication

4. Size of ldl particles, this one is new

5. Genetic markers especially 9p21.3, sadly about a third of people have this marker. How it works we don't entirely know, but the best guess I've come up with is it has to do with interlukin

6 signaling which signals inflammation. Solution to this might be taking curcumin, lower weight, Mediterranean diet.

7. Low HDL again mostly genetic, niacin may help some as may exercise

From my research into the matter it seems that the initial damage is done via HBP, shear stress, or cell repair failure. Then inflammatory markers attach to site causing white blood cells and ldl to infiltrate between the inner and outer walls. Eventually the plaque grows and the inner wall continues to thin until one of two scenarios occur 1) hard cap forms (calcification) and the plaque grows until it restricts enough blood that it causes heart attack or 2) The much more dangerous scenario of the inner wall thinning and breaking causing a clot and instant heart attack. The bad part about 2) is that its undetectable except for angiograph or nuclear stress (treadmill can if the heart rate goes high enough and the doctor skilled at interpretation). Some hope for 2) is that statins seem to encourage calcification of those thin walls thus preventing the heart attacks.

This is just what causes the heart attack and the actual fix (stenting and balloon angioplasty) has its own issues. Where the only surefire fix after a heart attack is a bypass, but if more than 2 plaques are found you need to harvest veins which don't last as long. What the entrepreneur world needs to focus on in my opinion is easier at home detection of coronary plaques (sound, radar, electrical, magnetic, etc.). If the government got involved nanorobotics and bio compatible hearts could be explored along with angiogensis via external stimulation, but each of those would be a moonshot program.

Would be cool if you open sourced it, I would happily add my NLP expertise to it as I have the same issues.

It is good to note that Lauder's principal is a statistical argument for degrees of freedom in a system. The Landauer principal on a statistical averaging of this effect, meaning you will sometimes be able to store a little more information.

I think you have your calculations off for the amount of energy required to store a human brain. Assume 8.6 x 10^10 neurons (more accurate number as its simple density measurement) each holding a 64-bit weight which gives us 5.5 x 10^12 bits. This multiplied by the lower bound e.g. k(1)ln2 or 9.56 x 10^-24 J, gives us 5.3 x 10^-12 J which is lower than the energy required to type a letter on this keyboard. Now even if we increase this by body temperature say 37 C we get 310.15 K or about 1.7 x 10^-8 J. This is of course a lower bound on the problem but even if you pump up the number of neurons or weight value you still get less than a joule of energy required.

This is a rolling lawsuit waiting to happen. Those that invested must really be hungry for something unique.

This is very cool stuff. I was just wondering what do you see is the biggest technical problem with this technology? I understand that the hardest part currently is transferring the modified genome into all cells, is this still correct?

If I'm interested in this technology how do you recommend learning the required techniques. As a machine learning engineer I know maths more then biology, but want to contribute to the open source movement. Where do you view the biggest impact of software/machine learning engineers can make to the open sourced biology movement?

I agree with your suggestions for human assessment because of problem posed by food calories varying significantly due to different preparations.

Take for example soda, if I gave you a picture of soda in a glass could you tell it was diet or regular? You might scoff at such an edge case but it quickly becomes more common when looking into food perpetration techniques. This is why caloric estimation is a really difficult and causes restaurants to not list their calories as the calories of a meal do not equal the sum of it's parts.

All these solutions are common as people want a signal to tell them to stop eating, but these are insufficient as people will simply ignore it due to hunger cravings (as happens on diets). Any nutritionist service in addition to detailing calories would need to incentive the patient to recognize the need to lose weight or setup a helpline.

https://www.reddit.com/r/MachineLearning/comments/3dbrum/out... reddit discussion on the same thing.

Note my thoughts are included and thought they might be of interest to anyone looking into this problem.

Agreed fully, this and machine curation of news stories.

Nope, in fact you are going to see salaries start to retreat as more people get into the industry.

The field that will collapse the hardest is going to be data science. I speak to this as a data scientist, where most I meet don't have the knowledge to perform their duties. Eventually our salaries will decrease to those of traditional office workers/middle managers.

What is a engineer to do in this situation? The answer is to specialize, or gain exclusive access. Specialization is obvious, and exclusive access are things like clearances, certifications, and networks. Of course, this omits paths such as entrepreneurship.

Dr. Stanley at UCF has taken this stuff to the next level especially with his HyperNEAT implementation http://eplex.cs.ucf.edu/hyperNEATpage/

I do wonder what type of effect this will cause with job requirements, will master's now be the new bachelors?

The more worrying issue is the government taking on more student loan debt. There is a student loan bubble but it has not hurt the economy partially due to federal requirements that the loan not be discharged. What happens if current graduation rates hold steady and the US is stuck with a large bill but nothing to show for the effort?

People are not worried about individuals of equal skill taking their jobs, but those of lesser skills.

Example. Individual A : Master's of CS from Stanford, Passed interviews, Java, SQL, Hadoop, JSF - Requested salary - 115k

Individual B : PhD Student at Stanford, Passed interviews, Java, SQL - Requested salary - 95k

Job requirements: Java, SQL, Hadoop, JSF

What would you decide on as the hiring manager, you have a limited amount of funds per year and need to meet a deadline. You know A is more qualified, but B is 20k cheaper. As a manger if time permits you will select B since he is cheaper and can acquire the skills needed in his spare time from his fellow engineers or from online materials. In making this decision B has the advantage due to price manipulation, and you allowing him to work even though he in under qualified.

Some might say this scenario is unrealistic but both candidates are using their salary to improve their lives, it just so happens B takes a lot less money (due to being born in a poorer country) to improve his life then A. Typically what you will see in the valley is B living in an apartment with 4 or 5 other people for 3 to 4 years periodically sending his money to his mother country. He can then start a company easily in that new country with his capital, meanwhile A plans to raise a family locally.

Much of this is happening right now http://www.nbcbayarea.com/investigations/Silicon-Valleys-Bod... Google, Facebook, Linkedin, Everyone does this in the valley.

Right now QRISK2 is considered the most accurate when it comes to predicting morbidity and mortality of cardiovascular events. This calculator is great, but missing one important factor if either of your first degree parents had a heart attack before 60 then you are at greater risk. This mostly ties into things like homocystine levels, hrcp, and LDL particle size, but its a good ballpark for the genetic variation.

My advice get genetically screened if you are worried and make changes. If you carry alterations on the 9p21 chromosome chances are you are at higher risk. If so you can simply eat more fruit and vegetables this has been shown in multiple studies to dramatically lower rates independent of other factors.

Its because textbooks are usually more in depth than lectures can go due to time considerations, this is why any graduate program is 90% papers and textbooks.

A great example of this is Andrew Ng's course, even though he is the co-inventor of LDA (complicated Bayes network) he does not explain Bayesian analysis in his course.

The biggest issue with the article is

"Last March, he enrolled in App Academy, a 12-week web development program with locations in San Francisco and New York, and found a job coding within three weeks of interviewing. In his new position as a web application engineer at Yola, a San Francisco-based website building company, Morrison says he earns considerably more than what he had made doing administrative work."

Same thing was happening before the crash of 2000, people would take a month long course and then get jobs based on that course. I ask everyone what is driving up the valuations, and to look critically into the future. As soon as the companies that have the large valuations (Twitter,Facebook,etc.) start to post negative growth you will see an exodus.

Will some of these be available as live streams?

This is a interesting take and why I love machine learning and its intersection with HPC.

First the seemingly blind decision to implement an SVM for improved performance. An SVM isn't magical in fact an SVM and neural network are equivalent with the SVM being the general case. SVMs suffer from the the same problems as neural networks in that your # Hidden Nodes/Activation function is the same as trying to choose your kernel function. When looking at your problem you have to ask yourself.

1) Is the model time varying? -> NN

2) Very large N dimensional search space -> SVM

Secondly even without changing algorithms you can get a significant accuracy improvement by examining your features. Features are the most important part of machine learning (garbage-in/garbage-out). Even simple classifiers such as Naive Bayes can do well if given the right feature set. There are multiple methods to examine your features such as ReliefF another is ANOVA. If you find your features are not good enough try unsupervised feature detection and learn more about the problem domain and coming up with your own features.

The final issue specifically with HPC and machine learning is even given 100 cores your algorithms may not speedup. Many machine learning algorithms are built to be iterative in nature and do not lend themselves to the becoming parallel. This means that MapReduce must be invoked at each iteration. As you scale up the number of available cores the overhead of startup and shutdown of your cluster at each iteration overrides your gain in performance, as many nodes will finish faster than others and just sit and wait.

The solution to all of this is simple

1) Get your features correct

2) Try new algorithms

   - Try online learning algorithm first like vowpal wabbit
     here https://github.com/JohnLangford/vowpal_wabbit/

   - Apache Spark http://spark.incubator.apache.org/ 
     or 
     GraphLab http://graphlab.org/
     or (if personal computer only)
     GraphChi http://graphlab.org/graphchi/

   - Both support HPC with graph centric framework

   - Orders of magnitude faster than Hadoop  

   - Both built on top of Hadoop HDFS so connect and go

Have fun and try to solve some cool problems.

Couldn't you just use something like Circular polarization http://en.wikipedia.org/wiki/Circular_polarization. Each point on the glasses is a TFT and polarizes the light coming in to the opposite direction thus creating a good black surface.

His second argument about the processing time is good if the processing is done on the phone but with cellular networks becoming better organized you could easily have a computing cluster do most of the work. Using that and some basic statistical inferences (to fudge some of the processing) you can get pretty impressive response times.

"I see the future of mass long-distance travel being underground vacuum trains. It's a huge engineering effort but would solve so many problems (eg air congestion, travel times)."

I don't think this will happen if the universe has taught me one thing its that things tend to take the least path of resistance. If anything I think the future will be a realized matrix, in that increasing the fidelity of the current internet.