Publications

2025

1. 

   Generative AI approaches for architectural design automation

   Adeer Khan, Seongju Chang, and Hojong Chang

   Automation in Construction, 2025

   Abstract HTML

   This review examines the potential and challenges of Generative Artificial Intelligence (AI) in automated building design within architectural practice. A comprehensive analysis of advanced generative models is conducted to evaluate their performance across eight architectural criteria. The qualitative assessment indicates that hybrid approaches combining diffusion models with autoregressive techniques provide the most promising outcomes for architectural applications. Despite advancements, significant challenges remain, including scalability limitations, fragmented workflow integration, and the lack of standardized evaluation frameworks. Potential solutions are identified through interdisciplinary collaboration and strategic research directions, such as developing unified evaluation metrics, enhancing model adaptability, integrating energy-optimized design generation for sustainability, and incorporating designer input in AI-driven workflows. This review provides a structured evaluation of current generative design approaches while proposing a roadmap for future research that bridges the gap between AI innovation and practical architectural implementation, ultimately advancing the field toward more efficient, creative, and sustainable building design automation.

2. 

   Neural Network Coding Layer (NNCL): Enhancing Deep Learning Robustness against Feature Erasure

   Chae-Seok Lee, Adeer Khan, Seong-ju Chang, and 1 more author

   IEEE Access, 2025

   Abstract

   This paper proposes a novel Neural Network Coding Layer (NNCL) that applies network coding theory to provide structured redundancy and enable reconstruction of lost features, thereby mitigating information loss problems in deep learning models. Unlike existing skip connections or attention mechanisms, NNCL embeds reconstructible redundancy into intermediate features through a learnable coding process, providing an explicit algebraic restoration mechanism for the original features. This design adapts the core principle of network coding—encoding information recoverably across a channel—for application within the computational graph of a neural network. Extensive experiments on the CIFAR-10 and CIFAR-100 datasets validate NNCL’s effectiveness. Even under a more conventional feature erasure rate of 20%, NNCL consistently improved classification accuracy by up to 8.3 percentage points. The model’s robustness becomes even more pronounced under extreme conditions. In a deliberate stress test involving 60% feature erasure—a scenario where baseline model performance collapses—NNCL dramatically boosted accuracy by up to 40.3 percentage points (e.g., from 19.8% to 60.1% on CIFAR-100). The proposed layer is designed to be modular and has been successfully integrated into various modern architectures, including ResNet and EfficientNet, and Vision Transformer, proving its broad applicability.

2022

1. 

   Machine learning-based monitoring and modeling for spatio-temporal urban growth of Islamabad

   Adeer Khan, and Mehran Sudheer

   The Egyptian Journal of Remote Sensing and Space Science, 2022

   Abstract HTML PDF

   LULC maps are important thematic maps that provide a baseline for monitoring, assessing, and planning activities. This study incorporates spatio-temporal land use/ land cover (LULC) monitoring (1991–2021) and urban growth modeling (2021–2041) of Islamabad, Pakistan to deduce the changes in various LULC classes in the past and the future by incorporating realistic influential thematic layers and Artificial Neural Network-Cellular Automata (ANN-CA) machine learning algorithms. Three decades of Landsat satellite imagery were used to classify LULC maps using a random forest algorithm with high Kappa indexes ranging from 0.93 to 0.97. Simulations for 2011 and 2021 were done for well-calibration of the model with Kappa (>0.85) and spatial similarity (>75%) using the MOLUSCE plugin in QGIS software. Future predictions were done for the years 2031 and 2041 to analyze and study the future urban growth patterns. The satellite-based LULC maps during 1991–2021 exhibited a 142.4 km2 increase in net urban growth. This had detrimental effects on other classes: net decrease of forests by 38.4 km2 and waterbodies by 2.9 km2. The projected increase of urban areas in 2021–2041 will be 58.2 km2. Visual urban sprawl assessment on LULC maps was done to highlight the type of sprawls. Overall, it was sensed that the city’s urbanization has been unplanned and erratic; leading to dire consequences on the environmental and urban systems. Therefore, the study necessitates better monitoring and better planning of urbanization by enforcing policies and necessary measures.

2. 

   A Simple and Sustainable Approach for Structural Health Monitoring of Structures

   Adeer Khan, Haider Ilyas, M Jalil Khan, and 1 more author

   Capital University of Science and Technology, Islamabad, 2022

   Abstract PDF

   Structural health monitoring (SHM) is an advanced tool that revolutionizes the capability of a structure to act as a responsive system by detecting changes and responding with performance analysis. But, for developing countries, its need is undermined due to its costly deployment. However, contrary to the costly belief, its use is direly needed in densely populated developing countries. Therefore, a simple and cheaper technique (despite lesser precision and accuracy) can help in the early detection of damages in structures. This can lead to avoiding financial and human loss. The primary objective of the project is to analyze the gaps in the application of SHM in developing countries and then recommend and achieve a simple approach to achieve its amenities through experimental and numerical validation. A critical review is made keeping in mind the previous research and the high-end deployed SHM on various structures across the developing countries. The advanced and simple approaches for SHM with their basic principles are thoughtfully analyzed. Then a prototype structure is prepared with induced cracking damage stages in columns and two cases based stiffness provided at joints. Snapback and harmonic tests are performed for both phenomena to assess the structural responses. A snapback test was performed to assess the natural frequencies and the damping ratios of the system. Whereas, the harmonic test was performed on the structure using a locally made shake table that was varied with increasing frequency and specific loading amplitude. The results were tabulated into acceleration-time and displacement-time histories which were initially used to assess the structural response. They were used to compute base shear and energy dissipation in the structure. Both methods produced reliable results. It was analyzed that the adopted strategy in the project is a viable and simpler approach to utilize on real-time structures. The instrumentation deployment is cheap and easy to handle. A combination of two approaches leads to better correlation results for the structure. Due to the increase in computational power, and the ability to handle large data through machine learning algorithms. An automated system can be devised that would detect the sudden changes in energy dissipation and time histories. It would then generate a warning through an automated smartphone system. This would allow better implementation of SHM in developing countries.