Register now After registration you will be able to apply for this opportunity online.
This opportunity is not published. No applications will be accepted.
ARC - Reasoning with Language and Vision
The Abstraction and Reasoning Corpus (ARC) is like an IQ test for machine intelligence, allowing us to quantify progress towards human-level AI. It was introduced in 2019
by François Chollet as a Kaggle competition. Rather than focusing on a single specific challenge, the aim of ARC is to provide an ensemble of challenges to measure the ability
to handle and adapt to new unseen environments. One task of the ARC usually consists of three given samples with input and output pairs. For a fourth sample, only the input is given and the output must be produced. These challenges seem easy for humans to solve, however machines, at least for now, struggle with them a lot. The implicit assumption is that we humans can rely on existing ”priors” and knowledge that help us tackle these tasks.
We are interested in following up an approach which shows potential by combining both vision and language to build a learned end-to-end pipeline to solve ARC.