To subscribe to the bd-ucla mailing list for seminar announcement, please visit this page

BD-UCLA (Big Data - UCLA, formely DB-UCLA) Seminar : Current Schedule

Time: 12:00pm-1:00pm Fridays; Room: 3551P Boelter Hall

*To invite a guest speaker or to schedule a talk, contact Youfu Li (youfuli at cs dot ucla dot edu)

Winter 2017

Date	Speaker	Title
01/06
01/13
01/20	Manoj Reddy	Constrained Clustering
01/27
02/03
02/10
02/17
02/24
03/03
03/10
03/17

Spring 2017

Date	Speaker	Title
03/31
04/07	Prof. Dennis Shasha	Reducing Errors by Refusing to Guess (Occasionally)
04/14
04/21
04/28
05/05
05/12
05/19
05/26
06/02
06/09

Fall 2017

Date	Speaker	Title
09/22
09/29
10/06
10/13
10/20
10/27
11/03
11/10
11/17
11/24
12/01
12/08
12/15

Constrained Clustering

Speaker:

Manoj Reddy

Abstract:

Clustering is a commonly used technique for exploratory data analysis. In this talk, I will focus on improving clustering quality using constraints. Constraints are represented in the form of must-link and cannot-link which represents when two data points need to be in the same cluster and in different clusters respectively. I will motivate the reason for constraints through a real-world application. This problem is modeled as a non-negative matrix factorization task and is tackled using a convex optimization technique known as Alternating Direction Method of Multipliers (ADMM). This is ongoing work and I shall present some preliminary results.

---

Reducing Errors by Refusing to Guess (Occasionally)

Speaker:

Prof. Dennis Shasha

Abstract:

We propose a meta-algorithm to reduce the error rate of state-of-the-art machine learning algorithms by refusing to make predictions in certain cases even when the underlying algorithms suggest predictions. Intuitively, our new Conjugate Prediction approach estimates the likelihood that a prediction will be in error and when that likelihood is high, the approach refuses to go along with that prediction. Unlike other approaches, we can probabilistically guarantee an error rate on predictions we do make (denoted the {\em decisive predictions}). Empirically on seven diverse data sets (chosen for their size), our method can probabilistically guarantee to reduce the error rate to 1/4 of what it is in the state-of-the-art machine learning algorithm at a cost of between 11% and 58% refusals. Competing state-of-the-art methods refuse at roughly twice the rate of ours (sometimes refusing all suggested predictions).

Bio:

Dennis Elliot Shasha is a professor of computer science at the Courant Institute of Mathematical Sciences, a division of New York University. His current areas of research include work done with biologists on pattern discovery for microarrays, combinatorial design, network inference, and protein docking; work done with physicists, musicians, and professionals in finance on algorithms for time series; and work on database applications in untrusted environments. Other areas of interest include database tuning as well as tree and graph matching.

---