[cs-talks] Upcoming CS Seminars: Data Group (Tues) + IVC (Tues) + Student Sem (Thurs) + Theory Sem (Fri)
fgreen1 at bu.edu
Tue Oct 6 11:34:12 EDT 2015
Data Group Seminar
Practical and Optimal LSH for Angular Distance
Ilya Razenshteyn, MIT
Tuesday, October 6th, 2015 at 11am in MCS 148
ABSTRACT: Locality-Sensitive Hashing (LSH) is a powerful technique for
the approximate nearest neighbor search (ANN) in high dimensions. In this talk, I will describe a recent result that gives an *optimal* LSH family for Cosine Similarity (aka Angular distance, aka Euclidean distance on a sphere). Unlike known optimal constructions [Andoni, Indyk, Nguyen, Razenshteyn 2014] [Andoni, Razenshteyn 2015], the new one
is practical, which is demonstrated by experiments on real and synthetic data.
Along the way, I will try to debunk two popular myths about LSH:
* LSH-based data structures consume too much memory and are thus impractical;
* Optimal LSH constructions are too complicated to be made practical.
No prior knowledge about LSH is required: I will start the talk with defining LSH and explaining how it can be used to solve ANN.
The talk is based on a NIPS 2015 paper joint with Alexandr Andoni, Piotr
Indyk, Thijs Laarhoven and Ludwig Schmidt. The preprint is available at
Supporting Collaborative Innovation at Scale
Pao Siangliulue, MIT
Tuesday, October 6, 2015 at 2pm in MCS 148
Abstract: Emerging online innovation platforms have enabled large groups of people to collaborate and generate ideas together
in ways that were not possible before. However, these platforms also introduce new challenges, such as finding inspiration from a large number of ideas. Prior research suggests that seeing particularly creative or diverse ideas suggested by others can inspire ideators, but few scalable mechanisms exist to assess idea diversity. We developed a scalable crowd-based method that can select sets of diverse ideas.
Noting circumstances where outsourcing the task to crowd workers is not feasible, we explored an “organic” human computation approach, where community members—instead of external workers—provide information about ideas while engaging in activities that help them with idea generation. The results of our experiment to evaluate our organic approach show that it creates a more accurate model of semantic relationships between ideas than an existing crowdsourced approach at a minimal cost to the quality of interactions.
We see this work as a step toward building more effective online platforms for supporting large scale collective ideation. We envision an intelligent self-sustainable ideation platform that integrates this approach to provide more effective inspiration to community members and thus enhance creative experience for users at both individual and community level.
Bio: Pao Siangliulue is a Ph.D. candidate in Computer Science at Harvard University, focusing on Human-Computer Interaction. She is broadly interested in how to technology can support people to express themselves creatively. Her current project focuses on designing and studying support tools for creative collaboration on open innovation platforms.
Before joining the Intelligent Interactive Systems group at Harvard, Pao received her B.S. in Electrical Engineering and M.S. in Computer Science from Stanford University.
Turing Testing Turing
Thursday, October 8, at 12pm in MCS 148
Topic: "The Imitation Game" is a movie<http://www.imdb.com/title/tt2084970/> about Alan Turing and his role in cracking the Enigma code.
If you have seen the movie and didn't like it, join us cause we are going to point out some stuff in the film that didn't actually happen (i.e. most of the stuff). If you have seen the movie and did like it, join us as we talk about how the movie pays tribute to Alan Turing. And if you have not seen the movie, just stop by to hear about this brilliant scientist.
Friday, October 9, 2015 at 2pm in MCS 148
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the cs-talks