Day 1

Lectures / Group overview with all instructors

Day 2

Introduction to Natural Language Processing and Word Vectors

Nothing is more essentially human than linguistic communication. But when programmers, data scientists, and computational linguists work with language, the abstractions they work with sometimes don't line up with your intuitive understanding of spelling and grammar. In this workshop, we'll investigate the state of the art of natural language processing, including: a whirlwind tour of spaCy for parsing English into syntactic constituents; a discussion of techniques for classifying and summarizing documents; and an explanation and demonstration of "word vectors" (like Google's word2vec), an innovative language technology that allows computers to process written language less as discrete units and more like a continuous signal. Workshop participants will develop a number of small projects in text analysis and poetics using a public domain text of their choice. In becoming familiar with contemporary techniques for computational language analysis, critics and researchers will be able to reason better about language-based media on the Internet. Artists and writers, meanwhile, might just learn a few new techniques to add to their creative palette.

No previous programming experience is required. But students will need to have Anaconda installed on their laptops before the session begins (instructions will be sent prior).

Day 3

Emotional / Subjective data and dataset creation

This session will be on emotional/subjective data and dataset creation, with coding in Python. We’ll talk about what subjective data means, and how to capture abstractions like emotions and other subjective experiences in datasets. We’ll talk about identifying subjectivity in otherwise “objective” datasets, looking at how the datasets were created, who tagged them if they were tagged, who funded the dataset, what the dataset leaves out, etc. We’ll talk about “artisanal data” and creating explicitly subjective datasets for artistic purposes. We’ll talk about how a culture’s current set of values is a type of bias that gets captured in datasets, and how we can avoid bias retention over time. We’ll talk about labels and vocabulary for these issues - how we can create these, and use them with regularity. Finally, we’ll create and explore our own subjective dataset. We’ll talk about how to be cognizant during the collection, analysis, and usage stages. We’ll also examine currently available emotional datasets, and explore them critically.

Basic programming knowledge, particularly in Python, would be helpful, but this will be an introductory session aimed at beginners.

Day 4 & 5

Generative art with neural networks

This will be a crash-course on a number of methods for using deep neural networks for the creation of generative art. Students will be given a tutorial on how to use several general-purpose tools, including image transformation networks (pix2pix, style transfer), visualization of convolutional neural networks (deepdream), as well as generative models for text and audio generation. Emphasis will be on prototyping and practice; students will be given temporary virtual machines on a cloud-based service with all the tools already installed. Additionally, we will cover "learning how to learn," that is learning about the fundamental skills and landscape of resources for learning how to keep up to pace with this fast-moving field.

Ideally, students can have basic familiarity with the command line / terminal. how to change directories, execute scripts, and so on. If you are a beginner, we will go over this, too.

Day 6

Deep Learn Web

This one day workshop will be focused on doing deep learning in the browser with javascript. We will use p5ML, an experimental library that does not require any setup tools or complicated frameworks to install. This will be a beginners friendly workshop and no previous knowledge of machine learning is require. Although, some basic experience with javascript would be helpful.

Creative games with AI

In this session participants will get to put together their own datasets, train a model based on this data sets and then use that model in a game or story they will create. (eg give examples of positive or negative messages and have a game that reacts to your messages). I know that sounds very complicated but believe me it's not and you need no prior knowledge. The goal of the activity is to make the technical aspects as intuitive and seamless as possible and focus on the creative aspect while discovering the constraints and affordances of our DIY datasets. We will use Scratch for building the games and and stories and Watson api for compiling the data sets and training the models. See examples of activities here. Note that for this activity we are not limited to digital data alone and we will also be able to use hardware (arduino, lego wedo, etc)

Bring your own laptop computer to the session.

Day 7

Sunday Brunch Panel

On Sunday, February 18th, SFPC is hosting an afternoon of panels and discussions around machine learning, ethics, and systems. Moderated by Caroline Sinders and featuring multideciplinary panelists ; this panel and round table discussion how machine learning affects journalism, art, and the greater public at large. Thinking about how algorithms are utilized for advertising, and how wrong and invasive that can go, how do we critically create while maintaining self-preservation and skepticism inside of these large, moving systems?