FACULTY OF

COMPUTER SCIENCE

Our most recent addition to the Faculty of Com­puter Science, which is respon­sible for the BiBa, is Dr. Alexander Binder, who joined us from Singapore in April, 2024.

Professor Binder studied Mathematics at Humboldt University in Berlin and earned his PhD in Computer Science at the Technical University of Berlin in 2013 with a thesis entitled ”Bag of machine learning concepts for visual concept recognition in images“. He spent most of the past decade teaching and researching in Singapore, first as an Assistant Professor at Singapore University of Technology and Design (SUTD) and subsequently as an Associate Professor at Singapore Institute of Technology. Between these two positions in Singapore, Alexander Binder was on the faculty of the University of Oslo for about a year.

In Magdeburg, Prof. Binder is starting a new research group in Computer Vision. This area has moved a lot from the initial classification of images and object detection. Image generation from text with diffusion models and deepfakes are known to a wider audience.

Nowadays, computer vision is heavily influenced by large language models, which gave rise to large multimodal models (LMMs).

Examples of his work are shown in Fig. 1 and Fig. 2. Such models for example can detect objects described by an arbitrary given textual input, answer natural language questions to images, or segment images based on textual queries. Other methods, like visual prompting, are learning to execute tasks on images based on example pairs of input and result images.

Fig. 1: Explanation for a dog class predicted by a vision transformer model. Note the focus on the head of the dog because the predictor is a discriminative model and the dog body is not discriminative between dog races present in the ImageNet dataset.

Fig. 2: Explanation for the concept of starfish (left). Unlike in the dog example, the whole object is important. Yet, not everything of the same orange color contributed equally to the prediction.

In this area, Dr. Binder develops methods and tools in explainable AI for vision which produce scores for every pixel of an image telling how much it contributes to a prediction.  He will be teaching courses in 3rd year of our bachelor’s degrees as well as in our master’s programs, and serve as advisor for PhD students. He still enjoys writing code by himself as time allows. Apart from that he likes landscapes, instant noodles and paneer butter masala, as well as watching sports from afar.

By the way: He shares his name with a famous movie director and a photographer (with the horned shadow photo) – but the three of them are different people!

Adapted from the BiBa Newsletter 2024/1, May 15, 2024