AI4Research-seminarium: Neural Augmentation: Hybrid methods between generative white box and discriminative black box models
- Datum: –14.00
- Plats: Zoom
- Föreläsare: Max Welling (University of Amsterdam, Netherlands)
- Arrangör: AI4Research
- Kontaktperson: Cecilia Alsmark
Välkommen att delta i månadens AI4Research-seminarium. Föredraget hålls denna gång av professor Max Welling från University of Amsterdam.
Scientific modeling has almost exclusively been based on models that imitate the forward generative (physical) processes in nature. Machine learning has been following a somewhat inverse black box logic by learning models that predict the latent factors (causes) given the observations. Both methods are formally related through Bayes rule. In this talk I will discuss the pros and cons of these methods and argue that, depending on the amount of available data, we need to build hybrid methods that leverage the advantages of both. I will discuss a number of such neural augmentation models and their application in medical image analysis, wireless communication, optimization and causal discovery.