Statistical models for social networks struggle with the tension between scalability - the ability to effectively and efficiently model networks with large numbers of nodes - and fidelity to social processes. Recent developments within the field have sought to address these issues in various ways, including algorithmic innovations, use of scalable latent variable models, and clever use of covariate information. This workshop will provide both a forum for presenting new innovations in this area, and a venue for debating the tradeoffs involved in differing approaches to social network modeling. The workshop will consist of a combination of invited speakers and contributed talks and posters, with ample time allowed for open discussion. Participating invited speakers include Ulrik Brandes, Carter Butts, David Eppstein, Mark Handcock, David Hunter, and David Kempe.

This workshop will be of interest to researchers working on analysis of large social network data sets, with a focus on the development of both theoretical and computational aspects of new statistical and machine learning methods for such data. Case studies and applications involving large social network data sets are also of relevance, in particular, as they impact computational and statistical issues.

Examples of specific questions of interest include:
- What are the key computational challenges involved in scaling up statistical network modeling techniques such as exponential random graph models to large networks?
- Do other techniques such as latent variable models offer useful and tractable alternatives to exponential random graph models?
- Are there new ideas from the algorithms community (in areas such as data structures and graph algorithms) that can be leveraged within network estimation algorithms?
- Can techniques from machine learning (e.g., approximate inference methods such as variational inference) be applied to statistical social network modeling?
- How can side-information (actor and edge covariates, temporal information, spatial information, textual information) be incorporated effectively into network models? How does this side-information impact computational tractability?