Parallelism & Intelligence


thesis - motivation - approach - overview - position of research - results


"Intelligence is necessary for succesfull parallelisation."

Current (non-intelligent) techniques will never be able to create an ideal machine. They can always improve the machine, but the ultimate goal will never be reached . Only ?intelligence? can solve the problems. This is proven by the human intelligence, as a human plus a computer form the ideal machine.
Moreover, we are getting stuck in more and more complex software, generating more and more problems to be solved.

Parallelisation is a complex matter:

Thus, a performant parallelisation algorithm should match the problem and the hardware. Therefore generality and nice, structured solutions should be sacrificed for optimisation. A parallelisation expert will have to develop specific solutions and this results in a combinatorial explosion of work. Moreover, parallel algorithms get complex. Then, if something goes wrong, it is hard to find the bugs. A good organisation is necessary.
For succes however, parallelisation should be ?sold in a box?. A user should not worry about any parallel aspect, parallelisation must work automatically for every hardware and every problem! In my research I want to investigate the problems that should be tackled to create an ultimate parallelisator. I want to proof that current techniques are extremely heavy and that we need something better, something that resembles human intelligence. It is not a matter of theory about parallelisation, but a matter of how to put this theory in a system.

The VUB PADX lab is specialised in parallelisation and discrete event simulation. I will use this experience to do my research on both topics.


My research is specific in 4 aspects: the problem analysis, the technique evaluation, the model of human intelligence and the synthesis.

1. Evaluation of techniques
A structured analysis starts with defining the ideal machine, here, the ideal parallelisator and the ideal simulator. Then, we should make an inventory of the things that should further be automated and analysed how to do this.
But the real problem is that most current techniques fail in doing this! Most research goes into developing and refining techniques, without making a thorough analysis. I belief that criticical analysis is an important lack in informatics research!

2. Problem analysis
My research is mainly a thorough analysis, not a construction of another technique. There is a lack of vision, a lack of answers to important questions. Questions like
what do techniques solve and what not? What can be automated? What not? And above all, why not? What are the fundamental problems to be solved.
These questions have to be answered.

3. Human intelligence
In doing this analysis, an important aspect in defining an ideal machine is applying the well-known Turing Test, because a human can write a perfect parallelised program. The combination of the computer and the human intelligence forms an ideal machine. So in evaluating current techniques, we have a test and in analysing the problems we have a clue.

4. Synthesis
The succes of this research will be measured by the succes of structuring and synthesizing the analysis. Moreover, I know that formalisation is crucial and is the ultimate challenge.


=> scheme

Position of Research
It is clear that it is a multidisciplinar topic (combination of parallel systems, programming, software engineering & AI). I hope that I can show the necessity of this.
Furthermore, I belief that the same fundamental problems can be detected in other areas as well and that the solution combines several research fields (this will be elaborated). Therefore a global approach is necessary and it can be considered as fundamental research.
Futhermore, I belief there?s good chance that this analysis will lead to a AI-hard problem, what justifies fundamental research.