Musings from my development journey.Topics
Talk on machine learning given at Rails Conf 2017, by Matthew Mongeau
Pattern recognition is one of the keys to humans being able to understand what elements are based on historical fuzzy learning. Aka, being able to recognize when something is food vs when something isn't food.
Clustering and recognizing patterns of data to see how they fit together.
Supervised learning is the machine learning task of inferring a function from labeled training data. The training data consist of a set of training examples.
Iteration based learning, where programs are informed at each iteration if their perception of the world is accurate or not.
Building and training neural networks. Formal definition: an open source software library for numerical computation using data flow graphs. Nodes in the graph represent mathematical operations, while the graph edges represent the multidimensional data arrays (tensors) communicated between them.
A dimensional array. Formal definition: Tensors, defined mathematically, are simply arrays of numbers, or functions, that transform according to certain rules under a change of coordinates. In physics, tensors characterize the properties of a physical system, examples here: examples
Built on the ImageNet data set. Enables programs to perform image recognition, documentation here.
Taking the knowledge base previously learned and applying it to new data. In data set, it's important to have the machine learning algorithm have knowledge about the expected data, but also the data that should be revealed as false. Example: if you are looking to recognize flower pictures, you should have images that aren't flowers in addition to having pictures of flowers.
There always needs to be room for error. No machine learning algorithm can be completely accurate or it means that it may be overfit (made too specifically).