Real Time Classification of Amazon Catalog

In Amazon EU Supply Chain, we are improving our customer experience and costs by answering complex questions using enhanced analytical methods ranging from Econometrics to more predictive methods (Machine Learning) ,optimization models (Operations Research) and Computer Vision.

In order to automate and improve the performance of classifying our worldwide product catalog at scale, we are working towards developing new text and image classification models. In particular, these models/pipelines need to be robust against imbalanced datasets and have low latency requirement.
Furthermore, as Deep learning models are notoriously data hungry, we want to develop generative models to augment our manually labeled datasets.
Finally, in order to better communicate model results and build trust in the models we want to apply explainable AI specific/agnostic techniques.

The ML solution developed by the students should not only satisfy the aforementioned requirements but also leverage state of the art methodologies around images and text data to improve our success metrics (precision, recall, F1 scores,  ecc..) of the currently used algorithms.

We see this as a great opportunity to familiarize the students with AWS technologies (EC2, EMR), which will provide on hand experience and be part of an exciting and innovative research and development project to develop new solution in state of the art of deep learning. This will provide fruitful on-hand experience to the students on highly demanded and value added skills (PyTorch, Tensorflow, Python; familiarity with AWS technology) and it will also be an opportunity for them to review and dive into machine learning models for classification.

Accepted students to this project should attend, unless they have proven knowledge, online workshops at the LRZ in April 2023, before the semester starts. More information will be provided to students accepted to this project.