1: \begin{abstract}

2: Recommender systems can be formulated as a matrix completion problem, predicting ratings from user and item parameter vectors.

3: Optimizing these parameters by subsampling data becomes difficult as the number of users and items grows.

4: We develop a novel approach to generate all latent variables on demand from the ratings matrix itself and a fixed pool of parameters.

5: We estimate missing ratings using chains of evidence that link them to a small set of prototypical users and items.

6: Our model automatically addresses the cold-start and online learning problems by combining information across both users and items.

7: We investigate the scaling behavior of this model, and demonstrate competitive results with respect to current matrix factorization techniques in terms of accuracy and convergence speed. 

8: \end{abstract}

9: