This industrial doctorate program aims to apply Artificial Intelligence in solutions specific to the real estate market.
Artificial Intelligence lies at the foundation of the company, and it is the main component of our products. The project aims to advance the relevant research and ultimately apply the developed methods to tackle one or more of the business needs.
The project will make use of vast amounts of data (visual, textual, and numeric) and apply one or more of the following concepts: multi-task and multi-modal deep learning models, or generative AI. The goal is to improve the company’s solutions’ performance and to open new research directions leading to the development of competitive products.
The multi-task part aims at investigating the potential of deep learning models designed specifically to perform multiple tasks on different data. Matching or even surpassing the performance of multiple specialized models might require applying concepts like knowledge distillation or exploring other methods that improve the baseline results. After the necessary investigation, the expectation is to prepare a procedure to build a described type of model for the production purposes of the company.
Multi-modal models might have many applications in the real estate market. They can include improving the performance of the visual-based models by adding new modalities, creating new neural network architectures and training pipelines, or creating new state-of-the-art Automated Valuation Models. The related research might cover defining specific data requirements to apply different methods, investigating the concept of continual learning, and solving design challenges by investigating the specific causal relationships between applied machine learning model architectures, optimization methods, and data preparation.
Generative AI is probably the most commonly included phrase in recent headlines across all media. Given the significant advancements in the field, applying this concept might bring substantial business value to the company. The necessary research will include assessing the usefulness of the company data (visual and textual) across different tasks and applying stable diffusion models (or other techniques) to generate visual and textual content given the specified constraints. Research on the potential methods and their optimizations is necessary, especially to ensure the highest level of coherence in the produced results.