Page 47 of 50

Re: FIVE years from now: can you predict the Software Landscape in anno 2023?

Posted: March 6th, 2018, 2:46 pm
by Traden4Alpha
It is too late IMO to parallelise things _after_ having written your (sequential) code.
Parallelisation is an optimisation step;.
How does one reconcile these two statements? The first is certainly true. But second one seems wrong unless the plan is to create a sequential code prototype but then throw it out to create the "optimized" parallel version.

I'd think that one of the very earliest design decisions would be "sequential vs. parallel" based on the customer's statements about the sizes of input data, required throughput or response time, time complexity of likely algorithms, and parametric drivers for computation cardinality (e.g., requiring N-digit accuracy implying O(f(N)) samples or time steps).

Re: FIVE years from now: can you predict the Software Landscape in anno 2023?

Posted: March 6th, 2018, 7:43 pm
by Cuchulainn
Nothing to reconcile, These two statements refer to different parts of the Sapir-Whorf continuum. Thee are not referring to the same things. 
Anyway, if you write a sequential program and it is too slow, then you have a problem. It means the response requirement got lost somewhere along the continuum.. 

to create a sequential code prototype but then throw it out to create the "optimized" parallel version.
Aka as a feasilbility project and a budget is needed. What often happens is that this beast is not thrown away and it evolves into a production system.

Re: FIVE years from now: can you predict the Software Landscape in anno 2023?

Posted: March 6th, 2018, 11:42 pm
by Traden4Alpha
Nothing to reconcile, These two statements refer to different parts of the Sapir-Whorf continuum. Thee are not referring to the same things. 
They sure sounded like they referred to the same thing.
Anyway, if you write a sequential program and it is too slow, then you have a problem. It means the response requirement got lost somewhere along the continuum.. 
Quite so.

to create a sequential code prototype but then throw it out to create the "optimized" parallel version.
Aka as a feasilbility project and a budget is needed. What often happens is that this beast is not thrown away and it evolves into a production system.[/quote]Maybe the solution is to write parallel code from the start. Perhaps one should avoid "loop-obessed" procedural programming languages that tend to foster/enforce sequential algorithm thinking.

And now we are back to your original prediction about functional programming!

Re: FIVE years from now: can you predict the Software Landscape in anno 2023?

Posted: March 7th, 2018, 8:58 am
by Cuchulainn
They sure sounded like they referred to the same thing.

English is weak when it comes to time axis; Italian (and indeed Latin!) is more precise in the sense when you say something at which point or time span the statement is valid etc.

In a court of juistice it would be a case of 'what I said, when I said it and who was listening."

Aka as a feasilbility project and a budget is needed. What often happens is that this beast is not thrown away and it evolves into a production system.[/quote]Maybe the solution is to write parallel code from the start. Perhaps one should avoid "loop-obessed" procedural programming languages that tend to foster/enforce sequential algorithm thinking. 

Indeed, probably 'parallel thinking' should be default? Sapir-Whorf in action.

Re: FIVE years from now: can you predict the Software Landscape in anno 2023?

Posted: March 7th, 2018, 2:55 pm
by Traden4Alpha
They sure sounded like they referred to the same thing.

English is weak when it comes to time axis; Italian (and indeed Latin!) is more precise in the sense when you say something at which point or time span the statement is valid etc.

In a court of juistice it would be a case of 'what I said, when I said it and who was listening."
Aka as a feasilbility project and a budget is needed. What often happens is that this beast is not thrown away and it evolves into a production system.
Maybe the solution is to write parallel code from the start. Perhaps one should avoid "loop-obessed" procedural programming languages that tend to foster/enforce sequential algorithm thinking. 
Indeed, probably 'parallel thinking' should be default? Sapir-Whorf in action.
And programming languages are even weaker when it comes to the time axis. What are the past, present, future, etc. tenses in C++ or any programming language?

Does Latin have enough time precision to talk about a condition in which two events (A & B) might happen in any order or simultaneously (as distinct from talking about two events with a specific sequential order)?

Re: FIVE years from now: can you predict the Software Landscape in anno 2023?

Posted: March 7th, 2018, 3:52 pm
by Cuchulainn
However both Java and C# are unlikely to be mainstream for analytics for quite a while

DCFC, 2005

Re: FIVE years from now: can you predict the Software Landscape in anno 2023?

Posted: March 8th, 2018, 9:35 am
by Cuchulainn
They sure sounded like they referred to the same thing.

English is weak when it comes to time axis; Italian (and indeed Latin!) is more precise in the sense when you say something at which point or time span the statement is valid etc.

In a court of juistice it would be a case of 'what I said, when I said it and who was listening."
Aka as a feasilbility project and a budget is needed. What often happens is that this beast is not thrown away and it evolves into a production system.
Indeed, probably 'parallel thinking' should be default? Sapir-Whorf in action.
And programming languages are even weaker when it comes to the time axis.  What are the past, present, future, etc. tenses in C++ or any programming language?

Does Latin have enough time precision to talk about a condition in which two events (A & B) might happen in any order or simultaneously (as distinct from talking about two events with a specific sequential order)?
The fact that time is not absolute leads to Actor system(motivated by QM and relativity). Agents only have mailboxes.

According to Hewitt (2006), the Actor model is based on physics in contrast with other models of computation that were based on mathematical logic, set theory, algebra, etc. Physics influenced the Actor model in many ways, especially quantum physics and relativistic physics. One issue is what can be observed about Actor systems. The question does not have an obvious answer because it poses both theoretical and observational challenges similar to those that had arisen in constructing the foundations of quantum physics. In concrete terms for Actor systems, typically we cannot observe the details by which the arrival order of messages for an Actor is determined (see Indeterminacy in concurrent computation). Attempting to do so affects the results and can even push the indeterminacy elsewhere. e.g., see metastability in electronics. Instead of observing the insides of arbitration processes of Actor computations, we await the outcomes.

Don't know Latin anymore but Italian is good with simple, remote and  continuous time (events, duration) and the role of the associated observe. Not needed for 'tourist Italian'. 
You can take the viewpoint that time is a fabrication that was invented by Bishop Ussher.

Re: FIVE years from now: can you predict the Software Landscape in anno 2023?

Posted: March 28th, 2018, 8:55 am
by Cuchulainn
I think the Actor model has a good chance depending on how entrenched software developers remain.

viewtopic.php?f=10&t=101006

Re: FIVE years from now: can you predict the Software Landscape in anno 2023?

Posted: February 14th, 2019, 3:59 pm
by Cuchulainn
How white engineers built racist code – and why it's dangerous for black people

https://www.theguardian.com/technology/ ... ple-police

Even if facial recognition software is used correctly, however, the technology has significant underlying flaws. The firms creating the software are not held to specific requirements for racial bias, and in many cases, they don’t even test for them.

Re: FIVE years from now: can you predict the Software Landscape in anno 2023?

Posted: February 15th, 2019, 12:01 am
by Alan
How white engineers built racist code – and why it's dangerous for black people

https://www.theguardian.com/technology/ ... ple-police

Even if facial recognition software is used correctly, however, the technology has significant underlying flaws. The firms creating the software are not held to specific requirements for racial bias, and in many cases, they don’t even test for them.

Interesting, although 'racist' is not the word I would use. But buyers of the software, like police depts, should insist on testing the frequency of Type I and Type II errors by race.  And good defense lawyers should call them on it:

 "If the pic don't fit, you must acquit"  :D

Reminds of a case in yesterday's news. I wonder how many facial recognition programs would identify these photos as belonging to the same woman?

Image

Re: FIVE years from now: can you predict the Software Landscape in anno 2023?

Posted: February 17th, 2019, 12:25 pm
by Cuchulainn
I think there are several issues

1. AI immature (the algorithms such ad SGD converge to a local minimum. if at all)
2. Algos are black boxes (they cannot explain how they arrived at an answer).
3. People have blind faith; if the media say it 3 times then it must be true. Yuge amount of hype. Everyone is doing AI, so you don't want to be left out.
4. When will "ethics" enter the software?
5. Developers are only interested in performance and dinky new technology, thus diverting their attention? In general, this is a valid assumption based on historical evidence.

Of course, I could be wrong. AFAIR there's a few lurkers  who can tell us.

//
I actually met the founder of AI (John McCarthy) in TCD in 1975 where he gave a talk in the maths department. He was a 1st cousin of my supervisor's wife.

Re: FIVE years from now: can you predict the Software Landscape in anno 2023?

Posted: February 17th, 2019, 12:58 pm
by Cuchulainn
Chinks in the armour?

https://www.bbc.com/news/science-environment-47267081


The “reproducibility crisis” in science refers to the alarming number of research results that are not repeated when another group of scientists tries the same experiment. It means that the initial results were wrong. One analysis suggested that up to 85% of all biomedical research carried out in the world is wasted effort.

Re: FIVE years from now: can you predict the Software Landscape in anno 2023?

Posted: February 17th, 2019, 4:01 pm
by Cuchulainn
How white engineers built racist code – and why it's dangerous for black people

https://www.theguardian.com/technology/ ... ple-police

Even if facial recognition software is used correctly, however, the technology has significant underlying flaws. The firms creating the software are not held to specific requirements for racial bias, and in many cases, they don’t even test for them.

Interesting, although 'racist' is not the word I would use. But buyers of the software, like police depts, should insist on testing the frequency of Type I and Type II errors by race.  And good defense lawyers should call them on it:

 "If the pic don't fit, you must acquit"  :D

Reminds of a case in yesterday's news. I wonder how many facial recognition programs would identify these photos as belonging to the same woman?

Image
Image

Re: FIVE years from now: can you predict the Software Landscape in anno 2023?

Posted: February 17th, 2019, 5:00 pm
by TinMan
1. That guardian article is drivel.  You should be embarrassed to post something as inane as that.  Classifying dark faces is mathematically harder than classifying light coloured faces.  It's not because the programmers are white.  That's like claiming pricing Asian options is easier than pricing Bermudan options because so many quants are Asian so it must be racism.

2. What the guardian consider racism to be and what most people consider it to be are very different things.

3. You didn't read the BBC article very well.  The replication crisis has been going on for years and it mostly refers to old fashioned significance testing.  ML may contribute to it but given that the culture of CS is to empirically test everything on new data to see if it generalizes, most issues in that regard come to light pretty quickly.  Unlike in other fields like medicine or psychology, where once a result is widely known it is very hard to shake.
See for instance the study Ben Goldacre just published.
https://www.sciencemag.org/news/2019/02 ... cal-trials
So ML isn't producing 'a crisis in science', science is in a a crisis all by itself.

4.  Nobody is trying to find the global minimum.  That would be overfitting by definition.  It's mind boggling how many people struggle with this

Re: FIVE years from now: can you predict the Software Landscape in anno 2023?

Posted: February 17th, 2019, 5:30 pm
by Cuchulainn
 Nobody is trying to find the global minimum.  That would be overfitting by definition.  It's mind boggling how many people struggle with this.

Please explain, as I am struggling with this. Up until now, it was always considered that a global minimum >> local minimum. Are you saying that it doesn't matter anymore?

What about adversarial examples that attempt to distinguish between a panda a gibbon. Is this caused by the GD algorithms? Just perturbing image a very small amount leading to something completely different image makes you wonder about the robustness of the algorithms used. In 19th/20th century mathematics this is called an ill-posed problem. I blame GD,.. feel free to correct me.

https://arxiv.org/abs/1712.07107

BTW you forget to mention the ethics, apart from rantish point 1.