1 / 8

Non-competitive VM – Parting Shots

Non-competitive VM – Parting Shots . Where do we go from here? . Life isn’t always progressive . Product purchase, technology/innovation adoption, political campaigns are reasonably approximated by progressive models.

kalona
Download Presentation

Non-competitive VM – Parting Shots

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Non-competitive VM – Parting Shots Where do we go from here?

  2. Life isn’t always progressive • Product purchase, technology/innovation adoption, political campaigns are reasonably approximated by progressive models. • However, subscription of services (ISP, mobility, apps) is often non-progressive: subscribers may switch in and out. How do you model and what do you optimize?

  3. Spread isn’t the only thing that matters • (Expected) revenue, or better yet profit is what a business cares about. • Influence doesn’t always imply adoption. • What if we wanted to minimize the seeding expenses to achieve a given target spread? • What if we wanted to achieve a given target spread under a given seeding expense budget in the quickest possible time?

  4. Other Perspectives on Influence Propagation • Minimizing Budget: Given a target expected spread, find the smallest seed set that achieves the target. • Minimizing Propagation Time: Given a target expected spread and a budget on #seeds, find the best seed set under budget that achieves the target in the least possible time.

  5. Minimizing Seeds • Greedy-MinTss provides -approximation to the optimum, while suffering from a shortfall of from the target. • Better approx. unlikely to exist unless . • Formally proved for real-valued submodular set cover.

  6. Minimizing Propagation Time • Unless , there does not exist a PTIME algorithm for MINTIME that guarantees (for any α ≥ 1): • a (α, γ)-approximation, such that |S| ≤ k, R = α・and ) ≥ γ ・ η where γ = (1 − 1/e + δ) OR • a (α, β)-approximation, such that |S| ≤ β・k, R =α・ and (S) ≥ η where β = (1 − δ) lnη, for any fixed δ > 0.

  7. Minimizing Propagation Time • However, if we can tolerate a coverage (i.e., spread) shortfall of and a budget boost of , then Greedy achieves these objectives within the min. possible propagation time.

  8. Summary • Spread, revenue, profit, adoption – hard to optimize on two levels. • In addition to n/w structure, influence probs matter and need to be learned properly from data. • Greedy + MC simulation + CELF++ -- guaranteed approx. but doesn’t scale. • Replacing simulation by SimPath/PMIA etc. dramatic speedup with little loss of accuracy on various data sets. • Credit distribution model allows to predict spread from data directly. • MINTSS, MINTIME – very hard problems in general. Potential for further research. • Non-progressive – some recent progress!  • Competition under non-progressive models – largely unexplored territory.

More Related