AI•ON
Artificial Intelligence
Open Network

The AI•ON collection of open research problems


Our problems are divided in two groups: applied research, and fundamental research.

Applied research problems are ideal for students looking to develop their expertise in machine learning and deep learning while doing novel work and having a meaningful impact on the world.

Fundamental research problems are ambitious and important problems for which no solution exists today, and which may not even be fully solvable in the near future.


Applied research problems

Cardiac MRI Segmentation

Develop a system capable of automatic segmentation of the right ventricle in images from cardiac magnetic resonance imaging (MRI) datasets.

Identifying biomedical articles at risk for retraction

Develop a model to analyze the content of new biomedical articles to determine the likelihood of fraud or scientific error.

Photorealistic post-processing of rendered 3D scenes

Develop a model (similar to a super-resolution model) capable of enhacing the realism of 3D-rendered scenes.

Smart data augmentation with generative models

Use GANs and other generative models to develop better data augmentation techniques for computer vision models.

Social media botnet detection and analysis

Analyze political botnet activity on Twitter and develop effective counter-measures.

Subpixel CNN in Upsampling Applications

Improve segmentation models and generative models by using a subpixel CNN as the upsampling operation.

Chromosome Segmentation

Develop a specialized visual segmentation model to help cytogeneticists conduct research.


Fundamental research problems

Layer-wise supervised incremental training of residual networks

Explore techniques for training supervised residual networks in a layer-by-layer fashion, rather than end-to-end.

Machine learning in non-stationary environments

Explore techniques for developing models that can perform well on data that significantly differs from the training data.

Multitask and Transfer Learning

Benchmark and build RL architectures that can do multitask and transfer learning.

Music generation based on surprise optimization

Use neuroscience and deep learning to generate music that pushes the right buttons in our brains.