Future of AI
Get inspired by the human brain and collaborate with Merck to shape the future of AI! Take a deep look at the problem of understanding the formation of invariant representations. Cross the boundaries of deep learning!
- Congratulations Team NeuroTHIx, well deserved victory!
- €5.000 + Paid Merck Research Fellowship + Joint Scientific Publication + University degree thesis
Get inspired by Mercks vision
The human brain has the fascinating ability to create invariant representations of real-world entities, i.e., to memorize and recognize `things' (which could be physical objects or abstract concepts like a song or a story) largely independent of how they present themselves to sensory perception. Until now the underlying mechanisms have been poorly understood, creating a major roadblock for the development of systems which approach human-level intelligence. It has been conjectured that in the brain essentially one `algorithm' - executed in parallel by many brain regions - is responsible for creating such invariant representations, but the exact nature of this algorithm remains in the dark. If this hypothesis is true, though, there must be certain commonalities between all cognitive tasks that lead to the formation of invariant representations - whether mediated via visual, auditory, tactile or any other senses.
Based on background information and the following Research Challenges the aim of this project is to generate insights from various disciplines that can lead to progress towards an understanding of invariant representation.
You can take one out of four tracks to shape the future of AI!
Take a look at the given resources: Can you find connections between algorithms which may create invariant representations and neural circuits which are actually implemented in biological brains?
You´ll find all the information you need and more precise Research Questions that could lead your elaboration in the attached Research Paper
Research Challenge: Cortical Algorithm and Grid Neurons
Take a look at the given resources: Can you elaborate the model and its assumption? Or can you even disprove the stated approach?
You´ll find all the information you need and more precise Research Questions that could lead your elaboration in the attached Research Papers
Research Challenge: A Theory for Representations of Hierarchically Structured Entities
Research Challenge: Vector Quantization and Entity Decomposition
Take a look at the given resources: Can you come up with (concepts for) an unsupervised learning algorithm or parts of it that are aimed at understanding invariant representations?
Research Challenge: Prototypes for Invariant Representations
Read everything you can find and let your creativity be the key!
Background Information and intro to the topic
Presentation and Book recommendations
© 2018-2023 ekipa GmbH. All rights reserved.