Pythology One-Day Conference: Machine Learning, AI, Genetic Programming

The next Pythology event is coming on Friday, September 22nd! Come learn about Machine Learning, AI, and Genetic Programming.

Space is limited, so don't miss out on this awesome event!

The event will include talks from several pythonistas in the Indianapolis area:

Check Out The Hay Before Diving After The Needle: Intro to Pattern Discovery
By: Trey Brooks
Data Engineer, Healthcare Bluebook
MCS-DS Graduate Student at University of Illinois Urbana-Champaign

This lecture provides a brief tour of how rules of association are formed, how we can mine patterns efficiently, and why these methods are so important for data science today. Pattern Discovery is a fundamental principle of data mining. It allows a data scientist to derive association and correlation. Patterns can illuminate latent clusters in data that would not be immediately obvious. Searching for these patterns can be applied to multimedia data, spatiotemporal data, time-series data, or stream data. Furthermore, the application of sequential pattern discovery can help us understand the likelihood of natural disasters, what medical treatments have been effective, and how the DNA has changed over time.

How to Build Skynet: 7 steps (with pictures), an Intro to Neural Networks
By: Brandon Boynton, CEO and CoFounder, Vemity AI

Learn about the history of neural networks, what they are and how they work. Then build your very own AI using a neural network to predict whether or not a mushroom is poisonous or edible. If that wasn't cool enough, you'll build a second AI to classify pictures of cats and dogs.

What you will learn:

1. What a neural network actually is

2. Why it is the premier data science tool.

3. How neural networks can be used in endless applications

4. What is truly important when creating an AI with neural network technology.

Most importantly, you'll leave with two basic AIs of your very own and the knowledge you need to build more.

Garbage In -> Garbage Out: Proper Data Handling for Machine Learning
By: Alyssa Batula, Recent PhD Graduate, Drexel University

Machine learning algorithms can do some incredible things, but they still rely on humans to make sure they're getting the right data. Even the most powerful algorithm will benefit from a well-prepared dataset, and a good dataset can be the difference between a working algorithm and a fancy pseudo-random number generator. This talk will cover some of the important steps required to make your data machine-learning-ready, including:

* How to find or create a dataset
* Initial dataset cleaning
* How to make use of training, testing, and validation data sets, and why they're so important
* How to select the best information to give to the classifier
* How to select a classifier for your dataset

Deep Learning for Text Analysis
By: Ananth Iyer (Pictured top) and Felix Wyss, Applied Research Group, Genesys

Text analysis has broad applications in information search/retrieval, business intelligence, data mining, sentiment analysis, online advertising, social media monitoring, etc. In this talk, we will cover several machine learning topics, including word2vec, deep neural nets, and natural language processing. We will showcase a system built with the popular and powerful ML Python libraries keras and nltk that classifies text queries into predefined categories based on just a small set of hand-labeled data. It is intended as entry point for chat-bots and natural language based call steering in interactive voice response systems.

Solving Jigsaw Puzzles with Genetic Algorithms
By: Tyler Foxworthy, Chief Scientist, DemandJump, Inc

In contrast to typical machine learning techniques, genetic algorithms (GA) are particularly adept at solving complex optimization problems in cases where there are multiple objectives and little information about the gradient of its error function. As a motivating example, we will design a GA in Python to solve jigsaw puzzles. Despite their leisurely appeal, jigsaw puzzles have a high degree of combinatorial complexity, multiple objectives to match both edges and images, and one global solution. The approach outlined in this talk is extensible to many real world problems and is intended to demonstrate the development and implementation of a GA solution to a class of problems which are not readily addressable by other gradient based machine learning techniques.