Study of Advanced Artificial Neural Network; Study of Machine Learning Algorithms in General; Understanding the Usage in Physical Data Analysis Based on Recent Developments in the Field

Kalra, Rakshit (2014) Study of Advanced Artificial Neural Network; Study of Machine Learning Algorithms in General; Understanding the Usage in Physical Data Analysis Based on Recent Developments in the Field. Masters thesis, Indian Institute of Science Education and Research Kolkata.

PDF (MS dissertation of Rakshit Kalra) Rakshit_Kalra.pdf - Submitted Version Restricted to Repository staff only Download (1MB)

Abstract

Machine learning is a branch of AI which is mainly concerned with the construction and study of systems that can learn from data. The core of machine learning deals with representation and generalization. Basics and main types of approaches are described and discussed. There are a wide variety of machine learning tasks and successful applications. Optical Character Recognition (OCR) in which characters are recognized is a classic example of machine learning. This system was designed after understanding the underlying approach here. A description of Genetic Algorithms is presented which is the key to understanding Genetic Programming. Once this framework is ready for ready for exploration, we shift our focus to describe its use in evolving machines in simulations as done in GOLEM Project. Taking this as the inspiration, a method was devised where robots could evolve in simulation based on their sensor data in reality. This normally causes a problem so a way to achieve coevolution i.e., Exploration Estimation Algorithm (EEA) is explored here. Using basis of coevolution and correspondingly EEA, a new method for reduction of evaluations when doing Symbolic Regression, is explored and discussed here. This is mainly achieved by subsampling dataset based on their ability to distinguish between individuals in population of solutions (in Genetic Programming). This model is further optimized to successfully reverse engineer non-linear dynamical systems. And lastly a variation in this setup can give us a way to find implicit equations in the system. This approach can thus help come up with Hamiltonians and Lagrangians easily (assuming that it‘s dependent on recorded observables).

Actions (login required)

View Item