Master thesis: Time series forecasting in the fashion domain
Forecast product sales in the fashion domain.
Background and purpose
Forecasting trends in fashion is important for selling both new and used clothes. Understanding trend is difficult since it is influenced by seasons, brands, prices, etc. Using open-source datasets, we aim to understand whether one can build reliable models for forecasting fashion trends.
About the project
We are building a dataset for used clothes with annotations for brands, size, damage, trend, among other things. While our dataset is useful for time independent characteristics such as size and damage, it is difficult to estimate the longevity of these used clothes. Cloth (re)use is determined in large part from the fashion trends at the time.
We will be using open-source fashion datasets to build trend forecasting models with and without the use of deep neural networks. Our goal is to understand what causes people to buy new and used clothes and how does this relate to factors such as the brand, seasons, size, visual design, etc.
Together with researchers at RISE Linköping, the student(s) will train models on fashion data and analyse performance of models based on different metrics. We offer GPU resources that the student can access for training and evaluating models.
The student is expected to have some background in deep learning, time series and some experience in training models. We use the PyTorch framework.
Credits: 30 ECTS (in agreement with the examiner)
Farrukh Nauman, PhD, Applied AI and IoT, RISE Linköping (email@example.com). Applications will be evaluated continuously, and the start date will be agreed with the successful applicant(s). Last day of application is 31 of March, 2023.
Linköping, Stockholm, Remote
Student - examensarbete/praktik
2023-03-31Skicka in din ansökan