Many people will call me crazy because I'm making this available before getting engaged on the actual project. Well, I guess this is the good part about not being a full time PhD student yet. Because my professional future is a mess right now, I have no idea what trend I will follow. May be I will implement one of these ideas, may be I will implement none of them, may be I will implement all of them. In the mean time, I'm just having fun. As such, thinking about this would not be as fun if I could not share with however is interested on reading it.
1)Computer Feelings – Because I wrote a short paper about this before (just browse http://labtricks.blogspot.com and check it out) I will not go into many details. The basic idea here is mixing several AI techniques in order to enable feelings on a computer. The procedure to accomplish (or having a starting point on) this would be by creating a frozen neural network “hard-wired” into emotion sensors. The purpose of such network would be interfering on the normal function of the emotion sensors just like Human Feelings do with the human “sensors” (tired, hungry, anger, etc). Similar to someone who loose track of time because it is reading something that he/she likes. Adding this “like” type of feeling to a machine would enable it to find a purpose to itself. A “common-sense” knowledge base would be used by them in order to balance whatever the computer “likes” to do versus what is best to its society. The common-sense plus the “like” feeling would enable the computer to guide itself during its learning activities. Always trying to perfect his leaning on whatever it likes most.
2)Ryodoraku temporal analysis – Ryodoraku is part of the traditional Chinese Medicine. It is a diagnose/treatment tool that allows the practitioner to have an energetic picture of his/her patient. The method is based on the evaluation of 24 acupuncture points. The practitioner uses an equipment similar to a multimeter in order to collect measures from each point. Those measures are plotted into a Ryodoraku chart. Based on the evaluation of the chart, the practitioner is able to diagnose the patient and know exactly what acupuncture points should be used on his/her treatment. I have already developed an Expert System that assist practitioner on this technique some time ago (available on the Apple App Store, i-Ryodoraku). My idea for a PhD degree would be composed by two parts. The first part would be running a detailed analysis on the temporal behavior of the Ryodoraku points. I'd collect and run data mining techniques in order to understand the cross-relationship between all Ryodoraku points over time. This initial analysis would provide resources to identify the behavior of health individuals and also the progression of states that bring a health individual to a sick state. On the second phase, assuming that there is enough evidence and understanding of the progression of states, I'd build a neural network which could be used to interpolate the several different Ryodoraku states of a single individual in order to predict the upcoming health state. This research would target the prediction of disturbing health symptoms into a currently health individual. Deploying such system into a Cell phone, PDA, or smart closes would allow people to prevent health problems before they happen.
3)Neural Network and/or Membrane Computing runtime deployed on a Cell Phone network. Neural networks and Membrane computing share at least one behavior. They both can be deployed on a highly parallel architecture. Membrane computing maps computer instructions into genetic cellular functions. I really do not think it is a good idea to write about this here. I promise I will write a more introductory paper about Membrane Computing at some point in the future. Right now, I know that Nei Soma has been researching a bit into this area in his lab at ITA (Air force Institute of Technology, in Brazil) – at least he was the one who introduced me into this topic. From the neural network perspective, the neurones are the smallest computing nodes of the system. I'm not sure if most people will agree but by now, Neural networks are, for me, a really clever way to build a mathematical functions. Specifically, it is possible to build mathematical functions that map anything into anything. You could map a digital representation of your face to your Social Security number. Your could map the digital representation of a flower smell into a description of the flower. It kind enhances the traditional mathematical functions into functions that map whatever you want into any other thing you want. The way the Neural networks operate is based on a lot of training. There are special algorithms that receive several (some times thousands or millions) of input x output pairs and train the network to do the mapping. This training is a very exhaustive process. However, after the network is trained, the actual execution of the “function”(neural network) is very fast. The PhD work here would be building a runtime and training environment that could be deployed on Cellular phones. The reason for using mobile devices is basically because there are millions of them widely available in the world and because the operation of a single node in a neural network requires low enough computer power that a mobile phone would be more than enough to execute it. Several problems would have to be solved; here is an interesting one: The neural network training requires lots of communication among the nodes, one way to mitigate this issue would be by using wifi-enabled or blue-tooth enabled devices closely located in order to train the network (a good scenario for this would be using the traditional high-school building to train all the nearby devices – imagine that it would be a cell phone high school as well). After the training is completed, the nodes could be activated from anywhere via SMS messages or the internet. This would not be a good solution for a problem that requires low processing power. However, for complex problems, the time to get a solution would only depend on how long it would take to send a message to the cell phones and receiving the reply from all of them. Tens of millions of nodes could be activated simultaneously. This has the potential to bit any super-computer available today.
Well, three is my lucky number. These are the ideas I have for now. I have to find a final candidate until the end of this year. So, do not be surprised if new posts like this come up soon.
-Luciano
