viewtransform

I recommend Prandoni & Vetterli's coursera class on Signal processing[1]. They have really made an effort to teach this in an intuitive manner instead of just throwing equations at you. This was the course where I finally really understood Fourier transforms in a visual way.

The other course I recommend is the Audio Signal processing class by Xavier Serra[2] This is a practical class with lots of hands on programming examples and introduction to useful open-source software tools.

[1] https://www.coursera.org/learn/dsp [2] https://www.coursera.org/learn/audio-signal-processing

A HIP version is available. https://github.com/ROCmSoftwarePlatform/hiptensorflow

The 'curse of dimensionality' should be required reading https://en.wikipedia.org/wiki/Curse_of_dimensionality for anyone getting into genetic algorithms or deep learning.

Also previously on HN http://www.visiondummy.com/2014/04/curse-dimensionality-affe...

Agreed. NFL is general and we need to discuss a specific landscape.

Deep learning is continuous <nonlinear> multidimensional optimization - yes.

What defines the subset of problems ? a general nonlinear mapping from R^n to R^m n>m ? or are you limiting to image classification ? speech recognition ? which would be a subset.

We have empirical evidence that deep-learning works but I'm not confident that we have the mathematical tools to understand why.

True. High-dimensional non-linear systems (like deep NNs) are a frontier we are just beginning to explore with empirical success.

However, if I was a VC - my knowledge of the NFL would make me look skeptically at any grandiose claims of miracle learning algorithms. In a sense, it provides me with an upper bound to BS.

This mathematical hand waving plagued the field of genetic algorithms and its variants in the 90's until the 'No free lunch theorem' was published from the Santa Fe institute. It essentially said (my take on it) that if you had no information about the landscape you were searching - then couldn't say much about the algorithm you were selling.

https://en.wikipedia.org/wiki/No_free_lunch_in_search_and_op...

Humans progress involves consigning some models to the trashcan of history.

At my Google screening interview, I was asked "How would you teach recursion to a 7 year old child ?"

Apparently, the correct answer was Romanesco broccoli.

I failed the interview and briefly sank into a depression questioning everything I knew about raising young children, self-similarity and recursion.

Thanks for the book reference - I'll have to read it. A film editor told me that a great actor knows when to blink and that's the point where you make the cut. Ever since I can't stop noticing eye blinks in movies and animated films ...

Could you reveal whether the Bay Area shops were borderline profitable or operating with decent margins ? I was surprised that they managed to stay open for more than a decade and assumed that they had figured out a decent sustainable business model.

At the Redwood city location there were many artists and small businesses that depended on the facilities. It will be a big blow to them.

I think they got reckless with their expansion. The San Jose location moved to a new location after their lease expired. They chose to relocate to a prime location in downtown San Jose and ended up with a million dollar shortfall to renovate and open the new location. They somehow managed to scrounge the money through an appeal to the community and opened but it must have ruined their financials.

At the same time they were opening in other cities. If their financials in the Bay area were shaky (as they admit in their closing statement on their website) then it is not clear why they were taking on more risk by expanding.

Does anyone know what breakthroughs in control systems have gone into making this possible ? Are there any papers describing how Boston Dynamics is programming this ? are they hardcoding the backflip from human data and then overlaying some stability code ? or is the backflip entirely generated by the underlying controls ?

Trader Joe's Multi-Purpose Cleaner. The green stuff. Spray and wipe the board with a microfiber cloth or any cloth rag. Cleans perfectly and smells great too.

HIP allows developers to convert CUDA code to portable C++. The same source code can be compiled to run on NVIDIA or AMD GPUs https://github.com/ROCm-Developer-Tools/HIP

AMD has HIP. HIP allows developers to convert CUDA code to portable C++. The same source code can be compiled to run on NVIDIA or AMD GPUs https://github.com/ROCm-Developer-Tools/HIP

AMD has a Radeon Pro SSG card with 16GB + 2TB on board memory.

You address the 2TB NVMe SSG memory as a filesystem directly from the card itself - no PCIe system transfer involved. 24.6TFLOPS half-precision.

You can program using OpenCL or HIP (opensource cuda replacement )

https://pro.radeon.com/en/product/pro-series/radeon-pro-ssg/.

https://github.com/ROCm-Developer-Tools/HIP

When a recruiter contacted me for a position at Google in 2012, I happened to be also taking Daphne Koller's Coursera class on Probabilistic Graphical Models (PGM).

I was puzzled why the recruiter wanted to know my SAT score. He also wanted to know the college and high school I went to and my GRE score and my undergrad GPA ... Keep in mind that I have a BS, MS, Phd in three different engineering fields and over a decade of work experience at well know companies in the Bay area. For a 40 something engineer it seemed odd he would ask me information from two decades ago but I went along assuming he was filling in some boxes for HR.

The first assignment in the Probabilistic Graphical Models class was to predict the likelihood that a candidate would be a good fit for a job given prior Bayesian probabilities for ivy league school attended, SAT score, GRE score, all forming nodes of a decision tree.

I looked at the problem assignment and the recruiters questions and realized what was going on. Some Stanford grad at Google had convinced HR that they needed an objective way to evaluate hundreds of candidates and what better way than to use the same methods that they had learnt from Daphne Koller at Stanford.

I quit the PGM class 'cause it seemed like a tool I would never use. I decline interviews at Google since I don't have the right Bayesian probabilities that would trigger a positive outcome for their PGM model.

"each cell from the same person will have different genomes"

Did you mean each cell will have different gene expression ?

I thought all the cells have the same nucleotide sequences barring maybe occasional mutations (like in cancer)

I second this. This was the explanation where I finally understood the big picture of ledgers, blockchain & proof of work

Asafoetida - "The devils dung" http://archive.aramcoworld.com/issue/200904/devil.s.dung-the... The most awful smelling spice in my cupboard. I have to put the spice bottle in a zip lockbag that is then put in another larger airtight glass jar otherwise the entire house smells like it.

Is used in Indian cooking - a pinch in hot oil transforms into a subtle flavor and aroma that people claim to be a replacement for onions and garlic. Used extensively in Jain cooking since the Jain religion prohibits use of onion and garlic in food. See http://www.manjulaskitchen.com/ for delicious vegetarian Jain recipes. Almost all her traditional recipes use asafoetida (aka "hing" in indian recipes). Look for her recipes around 2012/13 since those are the home cooked recipes before she went viral in the indian community.

This recipe with potatoes is simple http://www.manjulaskitchen.com/aloo-jeera-potatoes-with-cumi... . Use lemon juice as a replacement for mango powder.

These are amazing! I grew these in Santa Monica CA a couple of years ago and they came out the size of large mangoes with the best flavor I've tasted in a tomato. http://www.growitalian.com/tomato-cuor-di-bue-oxheart-106-24...

This bottom up approach was investigated in the late 90's after researchers became skeptical that the symbolic AI approaches in vogue then would achieve general AI.

You can look for the topic of Artifical Life, Emergent Computation, Artifical Evolution. See the Santa Fe Institute conferences on ALife. There were some promising initial results but the field never managed to take off.

We generally accept that we 'know' something when the model used to explain the system is simpler than the system itself.

The brain is a very high-dimensional non-linear dynamical system. The number of neurons in the brain is on the order of the number of trees in the amazon rain forest and the number of synapses the number of leaves on those trees. https://youtu.be/8FHBh_OmdsM?t=1165

We do not have the mathematical tools to understand such systems in general. What if reductionism doesn't work and the best model of natural intelligence is as complicated as the system itself ? Can we say we understand natural intelligence?

It could be the case in a distance future that we evolve an artifical intelligence purely as a computation that is capable of understanding us but not us them.

I would argue that intelligence is connected to the evolution of sensing at a distance. Vision in particular, allowed life to evaluate the state of the environment at a distance and allowed for the evolution of strategies to predict and respond to the environment in real time. The progression in intelligence from sponge to amphibian to mammal is related to the evolution of finer sensing of the environment at a distance: vision, smell, sound etc.

Not a wheel but a rotary motor: topologically the same. https://en.wikipedia.org/wiki/Flagellum

Put the cilantro/parsley stems immersed in a glass of water. Cover the leaves with a plastic bag and it will keep in the refrigerator for at least a week without wilting

Most immigrants strive to fit in with those aspects of the host culture that they appreciate. Every immigrant to the US changes their behavior: don't litter, obey traffic rules, stand in the queue, it's okay to smile at strangers etc because they see that everyone else is doing it and it makes life better.

However, this article is about tourists who have no stake in fitting in and will be on a plane back home in a few days.

I second that recommendation for Daniel Yergin's pulitzer prize winning book "The Prize". Oil as an energy source changed mankind. In addition to the history of oil in the US, it gave insight into the rising importance of oil during WWII (Hilter diverting his troops to capture the Baku oil field instead of focusing on Stalingrad), the post-war division of the world oil fields between the British and Americans (U.S. took Saudi Arabia (aramco), the Brits took Iran and Kuwait (anglo-persion and kuwait oil).

I remember when voice recognition was barely usable (90-00's) then all of a sudden within the last decade voice recognition started showing up and it was amazingly good. Siri can understand most of what I speak to it even while driving with music in the background.

What are the successful algorithms that made this leap forward possible ? HMM's ? Neural networks ? more data/compute ? What changed since Dragon NaturallySpeaking in 1997? Can anyone recommend overview papers/blogs on this topic I could use to get up to speed ?

As kids in Mumbai we had this Hindi rhyme

  Adha padha, kaun padha   # half a fart, who farted
  mamaji ka gohda padha    # uncle's horse farted      
  Doom, Doom, Puss.        # farting sounds