SIGraDi 2021 – Workshops

AI-Form is a synthetic form-finding methodology for topology optimization generated using an artificial neural network system in the architectural design field. With the development of computers and technology in the 20th century, the topological optimization method has spread wordwide in various fields. This novel structural optimization approach has been applied in architectural form-finding. Especially Bi-directional Evolutionary Structural Optimization (BESO), which was proposed in the 1990s, is widely used by thousands of engineers and architects worldwide to design innovative, iconic buildings. The related software Ameba developed by our team has been tested in several well-known workshops and design studios. In this workshop, we introduce the deep learning-based method to integrate with topological optimization to leverage the power of artificial intelligence algorithms to improve the diversity and efficiency of the BESO topological optimization method. Deep learning-based generative models such as Generative Adversarial Networks (GAN) can accelerate the topology optimization process of architectural structures. The workshop will explore methods of predicting optimized structures under defined conditions through GAN. Participants will learn how to use GAN combined with BESO/Ameba to facilitate the conceptual and architectural/structural design simultaneously. Through the workshop, we will demonstrate the ability of the novel approach and its application in architectural form-finding.

Language of the workshop: English
Workshop Tutor(s): Ding Wen 'Nic' Bao; Xuyou Yang; Xin Yan
Workshop schedule [08/11/2021]: Morning and afternoon (9 am - 12pm and 2 pm - 5 pm
Workshop schedule [09/11/2021]: Morning and afternoon (9 am - 12pm and 2 pm - 5 pm

Aprende cómo adaptar tus diseños a diferentes tipos de usuarios para obtener la mayor confortabilidad posible. Cambia el aspecto automáticamente y explora miles de posibilidades a solo un click por medio de un algoritmo que te facilitará la práctica en el diseño de calzado.

Language of the workshop: Español
Workshop Tutor(s): Valeria Palacio Manco y José Ovidio Cardona Osorio
Workshop schedule [08/11/2021]: Morning and afternoon (9 am - 12pm and 2 pm - 5 pm
Workshop schedule [09/11/2021]: Morning and afternoon (9 am - 12pm and 2 pm - 5 pm

This workshop is going to be seven sessions of 3 hours. It consists of three major parts: first, Fundamentals of deep learning, which is about the history of artificial intelligence and the relationship between AI, machine learning, and deep learning. It is also about the mathematics of neural networks. After that, we are going to learn about the basics of neural networks and machine learning. Second, we are going more in-depth in computer vision; we train a neural network from scratch. Then we learn about “transfer learning,” which means how to use a pre-trained neural network. Finally, and most importantly, the last part is about “Generative adversarial networks” (GAN). We will have some experiments with “ neural style transfer” as an example. After that, we will learn about GANs, specifically CycleGAN, and how to implement them. After this workshop, you know the basics of artificial intelligence, the relationship between machine learning and deep learning. You understand how neural networks work; you can make and implement deep learning models to classify images. You can also make Generative Adversarial Networks to synthesize new images and broaden your horizons.

Language of the workshop: English
Workshop Tutor(s): Mohammed Behjoo
Workshop schedule [08/11/2021]: Morning and afternoon (9 am - 12pm and 2 pm - 5 pm
Workshop schedule [09/11/2021]: Morning and afternoon (9 am - 12pm and 2 pm - 5 pm

Abstract: Introducción del asistente al diseño de edificios para impresión 3D desde la teoría actualizada hasta el diseño interactivo de arquitecturas, donde operará a través de un flujo de trabajo desde el CAD-generativo hasta el archivo de impresión 3D, pasando por evaluaciones de desempeños a tiempo real, evaluaciones de cualidades arquitectónicas y prototipado por manufactura aditiva a escala. Objetivo General: Introducir al diseño de edificios para impresión 3D desde la proposición de proto-arquitecturas. Objetivos Específicos: • Introducción a las tendencias actuales para arquitecturas impresas en 3D. • Exploración de cualidades/posibilidades espaciales emergentes de acuerdo con la capacidad de complejización formal de la tecnología. • Interacción desde la problematización del proceso de construcción con impresión 3D, delineando estrategias y resolviendo problemas a tiempo real. • Fenomenología de la arquitectura paramétrica en escala 1:1. Alcances-Resultados: • Aprendizaje basado en plataformas pre-parametrizadas: El asistente propondrá edificaciones para impresión 3D. • Análisis de geometría, construcción e impresión: El asistente podrá evaluar el desempeño físico de sus propuestas. • Previsualización interactiva a través de visualización inmersiva, modelos 3D y visualizaciones arquitectónicas: El asistente podrá aproximarse a la relación humano-edificio. • Prototipado remoto: El asistente podrá aproximarse a los resultados finales como al proceso constructivo de la impresión 3D.

Language of the workshop: Español
Workshop Tutor(s): Pablo Banda-Pérez, Rodrigo García-Alvarado y Ginnia Moroni-Orellana: Universidad del Bío-Bío, Chile Eduardo Valenzuela-Astudillo: Universidad Técnica Federico Santa María, Chile
Workshop schedule [08/11/2021]: Morning (9 am - 12 pm)
Workshop schedule [09/11/2021]: Morning (9 am - 12 pm)

In recent years, improvements in access, cost and availability of hardware and software have accelerated the adoption of technologies such as LIDAR, drones and photogrammetry for as-is surveys of historic buildings. Despite this promising renewal of capture methods, little attention has been given to the corresponding update on workflows, and how to actually incorporate and process this massive new data. As a result, the potential of these technologies is wasted, generating inconsistencies in the products obtained (planimetries, three-dimensional models, etc.), and errors in the subsequent studies. The workshop is based on an extensive search for best practices, recommendations from the heritage sector, and international standards to adequately process the point clouds obtained, adhering to the protocols promoted (CIPA, Cultural Heritage Imaging, Historic England, etc.), and maintaining criteria on replicability, metadata management, information preservation, among others. Finally, and as a way to avoid the “black box syndrome”, common with proprietary software, extensive use will be made of open source software, stimulating a reflection on the management and selection of processing algorithms to regain control over the data and what these processing stages do to them.

Language of the workshop: English
Workshop Tutor(s): Antonio Suazo
Workshop schedule [08/11/2021]: No activities
Workshop schedule [09/11/2021]: Afternoon (2 pm - 5 pm)

Design as a procedure involved in visual representations requires the expertise of designers who are able to process shapes in both 2-dimentional and 3-dimentioanl space. This procedure recently is simulated in various programming languages such as Grasshopper, Dynamo, Marionette, Python, JAVA and so forth, in order to obtain the merits of computation. However, this might cause much complexity and distortion while mapping the procedure from visual domain to numerical and text-based domain. Shape Machine is a novel visual computation system developed in the Shape Computation Lab at the Georgia Institute of Technology that can provide designers an end-to-end computation solution without mapping the visual procedure to text-based procedure. More specifically, designers can compute shapes directly with Shape Machine, so that the visual computation procedure can carry more possibilities and creativities without being limited by symbolic representations. This workshop is aiming for exploring the possibilities of this novel technology in design with the audience, who will be working with instructors with the Shape Machine in the workshop to test the ideas and extend the horizon of design.

Language of the workshop: English
Workshop Tutor(s): Tzu-Chieh Kurt Hong, James Park, Georgia Institute of Technology, United States
Workshop schedule [08/11/2021]: No activities
Workshop schedule [09/11/2021]: Afternoon (2 pm - 5 pm)

Beauty of cityscapes has long been studied anecdotally through accounts of novels and historical notes; and manually through analysis or scale, order, rigidity, etc. The use of machine vision allows designers to interrogate how urban fabric explicitly affects the perception of beauty, and consequently design interventions that may increase perceived beauty. This 2-day workshop is designed to be an introduction to the use of machine vision and machine learning in addressing the age-old question of “How to quantify subjective spatial experience?” Participants will be introduced to machine vision and supervised learning. As part of the exercise, machine vision algorithms such as semantic segmentation and object detection will be used to extract and quantify urban features (e.g., road, people, trees, vehicles, sky, building, signage, etc.) of photographs of urban scenes collected by the participants. Subsequently, participants will be asked to rate the photographs. The corresponding quantification of the urban features and beauty scores of the photographs will be analysed to extract spatial beauty principles. Finally, using the same dataset, a neural network will be trained to predict the perceived beauty of new urban scenes. Google Colab shall be used for semantic segmentation and training neural network.

Language of the workshop: English
Workshop Tutor(s): Joy Mondal
Workshop schedule [08/11/2021]:Morning (9 am - 12 pm)
Workshop schedule [09/11/2021]: Afternoon (2 pm - 5 pm)