New Persian Year
0Happy new Persian year to whom it may concern! Persian year starts on first day of spring.
While most of the Iranians went to vacation during holidays I was still working on my projects. I have to finish crisis simulation within next week , with documentation perhaps !. Data mining project is running well, it will be (hopefully) a money making website for me and its users will be Persian language people. First version will be ready for announcement next week. Its user interface still sucks ! I have to hire a designer.
Thesis is still on the fly
I sent my 70+ pages seminar to my advisor and still waiting for his response.
While I was implementing data mining algorithms for that website I figured out that most of the algorithms are not suitable for realtime response ! and also many of them are just useful for limited amount of data. I converted some algorithms to SQL language which has to calculate Cartesian multiply between 3 (and more) tables which each has more than half a million records then compare and use other calculations and show the result to user 7 times per request in realtime ! So I changed most of the common algorithm in this field. My advisor told me to prepare an academic paper for the whole idea, but I prefer to keep them secret and make money
Learning Rate
0Choosing appropriate learning rate can increase accuracy of NN. Large amount of learning rate can cause zigzag effect and NN cannot converge to a good min point (global min , hopefully !). You can see zigzag effect in these pictures. This NN has one hidden layer with 4 neurons and learning algorithm is Gradient Descent BP , all implemented in VB. Horizontal axis is for epochs and vertical is for MSE. Red line is MSE for test set and blue one is for train set. Thickness of red line shows amount of change in epochs. Thick areas show zigzag effect.
Exams are gone !
2Artificial Neural Net exam was 2days ago and I have to start implementing its projects. I need to implement Levenberg-Marguardt as one of the methods asked in the project description. I implemented most of the algorithms during the course but making a good and complete document for them is time consuming.
I am thinking about reloading my Persian blog too ! As all exams are finished I think I have more time to spend on some business ideas and starting up some startups ! I should find some co-founders.
Lower Levels are Always Better!
0What am I reading now?!
0Game Theory
Shannon’s Information Theory
Boosting in machine learning algorithms
PSO , Swarm Intelligence , Ant Colony
A huge , good and useful probability book
Reinforcement Learning and Multiagent systems
I have to start coding my multiagent system soon , this semester is going to expire ! in less than a month. Its object oriented view and goals of agents are ready , but I need to think more for their actions and cooperation between agents . Also “proactivity” of agents is still missing , DSS is too
By the way , my MAS project is a crisis management simulation .
Back from north
0I packed all my papers , books and other stuff and went to north to think about TOC of my seminar . I returned today with 3 pages TOC ! I have to start typing its first edition this week.
Tehran is too much polluted these days. I am planning to getting out again soon !
I used MCI’s unlimited 1 month plan for internet on my HTC HD2 . Their service is useful for checking emails and a little news , nothing more. That is good for vacation.
Take Care
New app , New Start
0I didn’t find enough time for changing old app so just installed WordPress. I am going to make this site active, AgAiN !
I will try to post something about my current studying field , Artificial Intelligence .
by the way , WordPress is too easy to run ! I like it ! specially its theme changing abilities .
I need to go to English class
Have fun

