DeepMDF

The DeepMDF project expands upon the previous ModificAI installation, first introduced at the 2023 Bratislava Design Week. This year's theme, relationships, is explored through visual connections across different design domains, from furniture to automotive. The project uses machine learning to uncover latent relationships between objects, which may not be obvious to human observers. It demonstrates these relationships on multiple levels, including between human and AI-generated art, form and function, and entirely different product categories. DeepMDF opens new avenues in both practical industrial design and speculative design theory.

The exploration of the latent space allows the discovery of hidden connections, which inspire innovative design variations. This multi-dimensional space fosters the creation of experimental solutions, expanding the boundaries of conventional design thinking. The project offers both practical implications, such as the generation of alternate variations, amalgamations and a theoretical dimension through speculative design, which encourages deeper contemplation of the role AI/ML could play in future of creative design processes.

By visualizing these complex relationships, DeepMDF demonstrates how AI can influence not only the aesthetic dimensions of design but also its functional aspects. The relationship between human creativity and AI-generated results is explored as a dynamic interplay, with potential applications across industries. Through this project, the team continues to investigate the boundaries between machine learning and industrial design, pushing forward the discourse on how AI might shape future design paradigms.

Input data preparation

The input dataset for the DeepMDF project was created using 3D modeling software Rhinoceros. The visualizations featured a monochromatic color scheme on a white background with a preset camera position. Each product section was represented by various objects, which were then utilized in experimental research to explore novel forms within the dataset. The project's methodology enabled the investigation of object relationships, facilitating the generation of new design solutions through the combination and variation of existing forms, pushing the boundaries of design possibilities.

This approach taps into the flexibility of Rhinoceros to generate clean, highly controlled visual outputs, optimizing the way forms are analyzed across different sections. By keeping the aesthetic simple, the experiment focused on pure form, reducing distractions from color or texture. Each object from a specific product section, such as furniture, automotive, or consumer goods, was intentionally chosen to represent a unique shape language. These objects served as "building blocks" in further experiments with form generation, using techniques like blending, variation, and interpolation or combination to produce modified new shapes.

Through the manipulation of these visual elements, the project explored hidden relationships between objects that aren't immediately apparent to the human eye. The dataset allowed for identifying unique intersections of shapes and functions that traditional design methods might overlook. The idea of latent space is crucial here—it refers to the uncharted territory where new forms arise from known objects by tweaking parameters or mixing features, creating innovative designs while retaining a sense of connection to the original dataset.

This methodology emphasizes algorithmic exploration, where technology aids designers in discovering and generating forms that combine the familiar with the unexpected. Such experiments not only reveal practical variations for design but also push speculative boundaries, questioning how far we can stretch existing design languages using AI and machine learning techniques.