Deloitte - DINDER.
Description of the company
Dinder is a new innovative app in the market of machine learning tools. Dinder is a revolutionary data classification framework that combines the ability to collect subject matter expert knowledge with the ease of a swipe. This approach playfully utilizes the vast knowledge base of companies to label arbitrary data for desired classification tasks, thus supporting the development of machine learning models over the whole company. Amazing, right?
Machine learning is everywhere. In some situations, it is used to support people, in others it is used to replace people. Without sufficient labeled data, the training phase of the model cannot be conducted properly. This problem especially manifests in the recognition of contextual information across varying data sources. Subject matter experts are well suited to deliver answers to arising questions on contextual information, but the lack of a platform renders the information gathering process slow and complex.
Imagine the same column label appearing across multiple data tables, but their content differs. Assume experts in the area could easily identify whether this column label and sample information on the column data belong to a certain context. The challenge is to gather the information from those experts as they might be spread all over the world or have time intensive schedules. Additionally, the classification process itself may seem tedious and therefore has to be wrapped into an overall entertaining product.
Aims of the project
Dinder helps its customers to collect labels for individual data sets to train machine learning models for subject-specific tasks by distributing the application to suitable subject matter experts within a company. The framework should be universally applicable, so it is not limited to one field of expertise, but can be used by any department who requires labeled data sets for their machine learning task.
- Transform a given task to be a machine learning task in such a way that yes/no labels suffice to train the model to solve the task
- Develop a smartphone application to enable users to classify data by answering (swiping) previously prepared questions or corroborate a statement
- Integrate additional features such as global high scores, user statistics, to make the use of the application attractive and to enable visualization of those statistics
- Develop a machine learning model for identification of contextual information within data tables using binary labeled data from open sources and the Dinder application
- Define performance measures for this identification task such that every prediction is accompanied by estimated, deterministic or fixed confidence levels to quantify uncertainty in the results
- Develop a system capable of assessing the data quality and quantity and be able to communicate measures to be taken by the company to improve gaps in their data’s underlying setup
Target group (students)
Students from the majors Computer Science, Mathematics, Computational Science and Engineering, Business, International Management and Engineering preferred.
Please reserve these dates: Fishing for Experience Termine
You can apply here for Fishing for Experience.