Hugo Larochelle, Chelsea Finn, Sachin Ravi
Learn a generative model in the few-shot learning regime to generate MIDI sequences or lyrics.
Rajat Mani Thomas, Paul Zhutovsky, Guido van Wingen, Max Welling
Develop a manifold of structural/functional brain scans in an embedding space that clusters mental disease states into clinically recognized classes.
Yoshua Bengio, William Fedus
Experiment on a proposal for a new kind of prior for top-level abstract representations, inspired by our understanding of consciousness.
Benchmark and build RL architectures that can do multitask and transfer learning.
Develop a specialized visual segmentation model to help cytogeneticists conduct research.
Improve segmentation models and generative models by using a subpixel CNN as the upsampling operation.
Analyze political botnet activity on Twitter and develop effective counter-measures.
Use GANs and other generative models to develop better data augmentation techniques for computer vision models.
Develop a model (similar to a super-resolution model) capable of enhacing the realism of 3D-rendered scenes.
Use neuroscience and deep learning to generate music that pushes the right buttons in our brains.
Explore techniques for developing models that can perform well on data that significantly differs from the training data.
Explore techniques for training supervised residual networks in a layer-by-layer fashion, rather than end-to-end.
Develop a model to analyze the content of new biomedical articles to determine the likelihood of fraud or scientific error.