Computational Design Laboratory

Ardavan Bidgoli PhD Defense

PhD-CD Candidate Ardavan Bidgoli will defend his PhD dissertation A Situated Approach to Machine Learning-Based Toolmaking for Creative Computing on September 14 at MMCH 121.

A Situated Approach to Machine Learning-Based Toolmaking for Creative Computing

Ardavan Bidgoli, PhD Candidate in Computational Design

September 14, 2022, 2:00 pm – 5:00 pm

MMCH 121

The latest boom of Machine Learning (ML) in the early-2010s has raised a new wave of interest among creative practitioners to explore the intersection of Art and Artificial Intelligence (AI), specifically Generative Machine Learning. A growing number of artists, designers, and architects, appropriate these algorithms to make their creative computing tools.
This dissertation introduces and documents a situated and collaborative framework for machine learning-based creative computing toolmaking. The framework embraces the idiosyncratic nuances and physical context of creative practices. It takes a new point of view on data and data curation as the primary method of interacting with the ML algorithm. The framework achieves this goal by utilizing small user-generated datasets, which are biased toward the creative practitioners’ personal preferences, subjective measures, and physical context. Through collaboration with machine learning expert toolmakers, the framework makes ML algorithms more accessible to creative practitioners. It highlights the affordances of ML algorithms, specifically Variational AutoEncoders (VAE), that can be efficiently trained and overfit on small datasets and produce outcomes that are closely tied to the creative practitioners and their context.
In the two case studies, the framework serves as a high-level blueprint to develop bespoke tools that support various stages of machine learning-based creative computing toolmaking process. In SecondHand, I collaborated with a group of participants to develop handwriting typeface generation tools. A dashboard, based on Dash Plotly, featuring interactive data visualization and data curation tools, was developed for this study. In ThirdHand, I collaborated with a musician to create a robotic tool to play santur, a traditional Persian musical instrument, using an ABB IRB 120 robotic arm and a real santur.
The case studies demonstrated that the proposed collaborative framework meaningfully brings ML experts’ technical literacy to complement creative practitioners’ domain knowledge and skills, overcome the technical ML challenges, and help integrate various idiosyncratic elements, physical context, and nuances of creative practice in the toolmaking process.

Dissertation Committee
Dr. Daniel Cardoso Llach (Chair)
Associate Professor, School of Architecture, Carnegie Mellon University

Dr. Eunsu Kang
Visiting Professor, Art and Machine Learning, Carnegie Mellon University

Prof. Golan Levin
Professor of Art, School of Art, Carnegie Mellon University

Dr. Barnabás Póczos
Associate Professor, Machine Learning Department, Carnegie Mellon University

See Final Draft

Author: Daniel Cardoso Llach
Category: News