Ashot Vardanian has been programming commercial software since middle school. His academic background is in Astrophysics, but his primary research areas are AI and computational complexity issues. He spends his free time doing anything from basketball and freestyle skiing to chess and travelling. He currently works on scalability issues of various machine learning algorithms.
Computational Essay: Linguistic Universals
Modern cities often reach populations over 10 million and form very complex systems that define our habitat and impact every part of our lives. Complex systems like that need a lot of planning and analysis, but nobody has tried to quantify the correlation between the number of parks and human well being or the features of the transportation network and quality of life. This work will provide a foundation for future research in city design. To make it possible, I will be using techniques ranging from regression models and various image segmentation methods to recurrent and convolutional neural networks to analyze maps, images and text.
Summary of Results
The problem of quantifying the quality of life proved to be challenging on its own, let alone the problem of relating the features of urban design to it. The conventional linear statistical models trained on simple, manually extracted features didn’t provide good results. The only systems that seem to converge to a low level of error are large black-box neural networks, which need to be trained on huge sets of data.
The rule of thumb in data science is “garbage in, garbage out.” To get a better understanding of the subject, we must process a lot more data than just 100 cities or 50 features. We should construct larger pipelines that will be able to process not just a one-mile radius map but an entire planet.