Unbiased is a deep tech startup determined to challenge the complex societal issues and to build next-generation business models with next-generation technology. Some of the problems we are addressing include Fighting Bias in AI, Fighting Fake News & Misinformation on the Internet. We love to encourage and work with passionate individuals who are crazy about innovation, technology and driven to do something good for the benefit of others.

Master Thesis - Predicting Bias of a given dataset or an algorithm[NLP Focused]

The thesis is focused on the NLP domain but the work should also consider other domains like Computer Vision.

Summary:

The goal of this master thesis is to build a framework or a practical methodology that can be applied on a given dataset or an algorithm to predict various biases like gender bias or racial bias or prejudice bias etc. There can be various types of biases in general but the idea with this thesis work is to build a working application that can take an input dataset and do a bias prediction, present the predictions or indicators as an output.

The end application would help enterprises and governments to ensure that the AI & ML applications are following ethical considerations.

This Master Thesis Project is performed in collaboration between Alten IT & Unbiased.

Background:

AI and ML models or algorithms depend on data. The algorithms are only as good as the data it's trained on.

Natural Language Processing(NLP) is a growing technology domain in the field of AI and with the recent innovations like BERT, we are now closer than ever to achieve results we haven't imagined before. NLP had a wonderful year during 2019 with the introduction of models like BERT and the adoption has increased more than ever. Many enterprises today use NLP algorithms and datasets to build critical applications.

As the adoption increases at the same time, this brings new concerns on how good these models are under real-world conditions. We can ask questions like How biased is my model? What kind of biases does my dataset have? Is BERT biased towards a certain culture or community?

The above questions raise serious concerns as they don't align with ethical guidelines or policies.

In the last year or two, there have been many advancements in this domain and as part of our mission at Unbiased, we would like to contribute in the fight by innovating and developing new solutions.

Key Requirements:

• Masters Student with Computer Science or Statistics background based in Göteborg.

• Driven to innovate.

• Good hands-on experience with programming & web technologies.

• Experience in Python

• Experience working with TensorFlow or similar frameworks

• Comfortable in finding the datasets, asking questions and taking initiatives

Expected Outcomes:

• 4-5 articles addressing the problems, challenges and existing solutions.

• Detailed theoretical analysis report presenting the research landscape in this domain.

• A framework built based on the identified research gap.

• Master Thesis report

• An end-to-end ready to use tool to get predictions from a given dataset.