Currently I get paid to work with A.I. I have developed a lot of personal and university projects in terms of artificial intelligence and machine learning. Looking back it's quite interesting how I focused on controll problems.
Super cool project I have dedicated the longest time out of all, I work on a full scale application.
This is my master's thesis project, I spent nearly 4 months researching. Spiking neural netowrks have the potential to be implemented into nano robots and could be used for a lot of meaningfull applications such as treating cancer. This is why I chose this project. Spiking neural networks have a non differentiabilty challenge and they closely replicate how human brain works. I learned a lot about artificial intelligence during this project from my supervisor. Who taught me how to use Tensorflow, Gradien tapes and has shown a lot of graphing techniques. Through the project we found 3 different ways of training the SNNs, one of them was differentiable.
The goal of the project was to find novel training methods. Which I did by combining probabilities and proportions creating a relationship map for each costumer. Later me and the team have integrated the algorithm into the app using flutter. I consider this to be a huge achievement :)
I used this project to teach myself DEEP Q learning. Suprisingly RL can gamble reasonably well :D.
This was one of my masters projects. I had to research techniques and apply it.
This was one of my masters projects too. I made a modified decision tree.
I tried building generative adverserial networks for generating pixel art. I was struggling for a good week, until I realised using gausian noise was very imprtant :D
I have done this for about 5 times. In 3 differnet languages :D
I have participated in few Kaggle competitions, because I was interested in Pandas, Numpy and Sci-kit learn when I was a student. I always was slightly better than average.
This is probably my favourite control problem, because we had a real robot. I wrote the program from scratch as well. Which got me very close to max score.
I was interested in Genetic Algorithms from Code Bullet's and other 2018s videos. Training SNNs with GA was one of the approaches. But I believe there is always that one algorithm that outperforms GA in any problem. I ussually end up using them to see the solution's logic and work from there.
I watched Patrick Henry Winstons Lectures on AI Which were really informative on older methods, I really wanted his book, but I could not find it. Another good channel is Tubingen University ML it feels more modern.