Treffer: 从分析到设计生成 ——人工智能作为一种设计方法.

Artificial Intelligence (AI) is changing every aspect of human existence, including architecture. AI is embedded in software tools and is changing the nature of design. This study aimed to explore how the discipline of architecture can help extend the design of constrained spaces using AI learning systems, how post-human frames of mind can be combined with AI in the built environment, and how new design possibilities can be explored through the synthesis of large, shared datasets and buildings. Based on prior theoretical studies and case studies, this reviews and introduces the possibility of underlying logic, learning mechanisms, algorithm models, and extended applications of AI as design methods. Deep learning is the underlying logic of AI as a design method, and artificial neural networks (ANN) provide design capabilities and methods to extract information from any given dataset. Through interpolation and extrapolation, AI extracts features from whole datasets and further generates new true samples. Complex constructional details and structural properties can be useful in more complex combinations. These can increase the design efficacy of combinations by providing access to a variety of combinatorial scales, thus enabling the deep understanding and development of formgenerating strategies. Deep neural networks are automated machines that mimic the plasticity of the human brain. These can "receive", "output", and "generate" forms. With good nonlinear modeling capabilities, deep neural networks are also able to handle complex feature relationships. They can learn and improve autonomously, learning features directly from datasets, and are more suitable for large datasets and complex, metaphysical logical learning. Deep neural networks provide a powerful tool for form generation by: (1) training their neurons and plastic connectivity; (2) sampling their latent space and generating probabilistic sequences. Based on feature datasets, training neural networks combine form generation with design sensitivity of architects, thus gradually realizing automation of the design process. Natural language modeling systems are another important branch of deep learning, in which form generation and visual outputs are realized through linguistic text. The produced interconnected system allows encoding of design intentions at different system levels through multi-mode algorithms, as well as text-, image-, and gradient-based inputs. It can further realize three-dimensional models, and finally transfer language information into completely developed architectural entities. This article summarizes the characteristics of deep neural networks and their influences on the development of architectural design methods. It emphasizes the importance of improved datasets created by architectural design to model the optimization and generation of results. Combined with specific cases, the applications and improvement of different algorithm models for design processes have been introduced thoroughly. Operations based on deep neural networks is the core of deep learning, which can learn abstract laws, thus enabling the system to deal with more metaphysical problems. Moreover, deep neural networks show characteristics of complicated nonlinear dynamic systems since they are the collective behavior of abundant neurons. During the processing of practical problems, the training and learning of system datasets based on neural networks can fit nonlinear functions approaching complex realities at any precision, and solve problems, such as complicated environmental information, ambiguous background knowledge, and ambiguous rules of inference. Architectural design is a typical representation of these problems. With the help of relatively mature machine learning and computer vision (CV) technologies at present, AI can design at various scales and assist the entire architectural design process. AI (currently represented by deep learning) truly responds to the "complex logic" of design thinking, and develops advantages of logic, generativity, proliferation, and autonomy of design thinking. Influences of AI on architectural design methodology has just began and will continue to develop. [ABSTRACT FROM AUTHOR]

建筑学科可以使用人工智能学习系统来帮助扩展受限的空间设 计, 旨在探讨如何在建筑环境中实现与人工智能的结合。通过相关理 论介绍与案例研究, 对人工智能作为设计方法的底层逻辑、学习机制、 算法模型、拓展运用等方面进行尽可能系统的梳理和介绍。以深度学 习为代表的人工神经网络提供能够从给定数据集中提取信息的设计能 力和方法, 处理复杂的特征关系。归纳了深度神经网络的特征, 及其 对建筑设计方法自身发展的影响, 结合具体案例对不同算法模型在设 计过程中的运用进行了较为详细的介绍, 强调人工智能的价值在于真 正应和了设计思维的“复杂的逻辑性”特征。 [ABSTRACT FROM AUTHOR]

Copyright of South Architecture / Nanfang Jianzhu is the property of South Architecture Editorial Office and its content may not be copied or emailed to multiple sites without the copyright holder's express written permission. Additionally, content may not be used with any artificial intelligence tools or machine learning technologies. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)