| Date: | 2021, January 19 |
| Time: | 09:00 a. m. |
| Place: | Online |
| Author: | Glauer, Martin |
| Title: | Symbolic Compositional Representations in Grounded Question Answering |
Recent approaches in artificial intelligence led to major advances in object detection and natural language processing. Convolutional networks can be used to detect objects in images, while transformers can estimate the intended meaning behind written sentences and texts. Yet, a combination of both fields – the answering of complex queries on grounded images – is still a great challenge, as it often involves multi-step reasoning.
In this talk, we present and discuss a novel CYK-based approach by Bogin, Berant et al. which tackles this problem of grounded question answering by inducing an inductive bias towards tree structures.