Petrus, Kashyap,, Pedram, Fabio, Katja, Carlos, Homayun, Zeinab, Aurelién, Sami.
|20 min||Active teaching||Carlos and Aurelién|
Having an active teacher that has a cognitive model of the user. The target is to minimize the number of iterations needed to learn a kanji by optimizing the order of presentation of teaching materials.
We need to have a cognitive model connected to a teacher model connected to a human that provides observational data.
There are different models that can be used for the cognitive model. ANN is a popular model for modelling the memory. The input of the network is a binary ID of the kanji at the moment but it may be better to have a better feature representation there (next steps).
Simulated expected results were promising.
|20 min||Modelling for modelling made easy||Pedram|
presented the modelling ideas for modelling made easy. The main focus of the presentation was about how to generate models. Two directions was proposed: 1) just consider a flexible model with many model-likelihoods attached to it. Domain knowledge affects this model through observations of this model-likelihoods. 2) define a set of model components (e.g., Gaussian or half-Gaussian distribution assumptions for weight) and then search in space of applying these components to a linear regression model.
The first solution is easier to implement and does not require Stan but it also has a focus application of linear regression while the second one can later be generalized to other probabilistic models.
- Next week: No one volunteered. Pedram will send an email next week to see if anyone is interested.