The Austin Python Meetup Monthly Meetup

The presentations will start after 7, yet feel free to join starting 6:30.

Talk 1: Layne Sadler - "AIQC; framework for rapid & reproducible deep learning for open science."

Talk 2: Massimiliano Genta and Avinash Gopal - Metabob an AI-assisted tool for debugging Python code

Details about the presentations below:

-------------------------------------

Talk 1: "AIQC; a framework for rapid & reproducible deep learning for open science."

Description: AIQC is an open source Python package that provides a persistent, relational API for: preprocessing, tuning of models in batches, and automatically generating performance metrics/ plots. It provides a high-level API that reduces the amount of code needed to perform best practice machine learning by 90% so that scientists can easily integrate deep learning into their research.

This talk covers: (1) the origins of AIQC in (a) barriers that prevent scientists from adopting deep learning and (b) sources of bias that are hardcoded into machine learning toolsets, before moving into (2) a live demo of the AIQC high level API.

More specifically, we will see how we can use the AIQC framework to address these chronic problems which are hardcoded into current machine learning toolsets: (1) Data Leakage; when aggregate information about test/ holdout data is used to process training samples. Most encoders are not handling each split/ fold individually, so information about the test data “leaks” into the transformation of the training data itself. (2) Evaluation Bias; when a user makes changes to their topology/ parameters based on how the model performs against the test/ holdout data. Most programs are not using a 3rd validation split, so they are effectively training on their entire dataset when they make adjustments. (3) Partial Reproducibility; it is common for experiment trackers to ignore the sample splits/ folds as generic inputs to or upstream artifacts of the training process (e.g. `X_train, y_train`). Despite the fact that preprocessing can be just as important as hyperparameters (e.g. PowerTransformer vs StandardScaler), most experiment trackers are blind to how the samples were processed, only focusing on the parameters to be tuned.

Bio: An autodidact at heart, Layne began on the business side of technology, but curiosity drew him toward building applications and algorithms alike. This led to cofounding an API-driven startup, athlete.studio, and spearheading product development at a biotech, Genuity Science. While working with pharma and research institutes on national genomic biobank projects, he observed barriers that prevented the adoption of deep learning in scientific research, and gaps in machine learning tools. So he built AIQC to address those problems.

Talk 2: Metabob an AI-assisted tool for debugging Python code

Description: Metabob (www.metabob.com) is an AI-assisted tool used to debug Python code. Using conventional static analysis and attention-based models, Metabob is able to detect where problems are, how they interact with other aspects of your codebase, and offer plain-text recommendations on how to fix them. In this event, we will demonstrate how easy it is to get started.

Speaker: Massimiliano Genta
Speaker Bio: Massi is a serial entrepreneur and founder with an extensive track record of success, ranging within deep tech startups to the IoT & AI industries. In his previous endeavors, Massi was the founder of smartflex (now senstek), Swayup, Clyste, as well EIR at NEC X.

Speaker: Avinash Gopal
Speaker Bio: Avi has an illustrious background in Aerospace Engineering, with expertise in every breadth of development, specifically versed within AI and machine learning. He’s worked on multiple Aerodynamics Research Projects, specializing in novel airfoil design and autonomously controlled crafts for personal transport and area mapping. He also served as the CTO of Clyste.