Seminar

The Cognitive Modeling Paradigm: An Experiment in Casual Inference

Monday, October 26, 2015 - 6:00pm to 7:00pm
Davis Auditorium, Room 412, Shapiro SEPSR
Columbia University
New York, NY 10027
United States

Alex Cosmas
Chief Scientist | Booz Allen Hamilton

The analytics community has invested significant resources in developing effective predictive analytical methods. However, even the most accurate predictive forecasts have limited value unless they can also provide clear action steps to bring about desired results.  In other words, the cause of the data is more important than the data itself.  Alex will introduce causal inference in the context of Bayesian Belief Networks (BBNs).  BBN’s produce accurate predictive forecasts, but with appropriate modeler input are also able to identify causal relationships between variables and pinpoint drivers of desired targets. With causal relationships identified, BBNs may be used in a prescriptive fashion in order to make actionable decisions.  Alex will dive into a case study in the aviation space which identifies causal drivers of daily flight operations on flight delays and allows us to prescribe delay-reduction plans by acting on controllable drivers.

Adventures in Little Data

Tuesday, May 5, 2015 - 4:00pm to 5:00pm
Costa Engineering Commons/750 CEPSR
530 West 120th Street, 7th Floor
New York, NY 10027
United States

Professor Paul Ginsparg, Cornell University

Abstract:

I will give a very brief sociological overview of the current metastable state of scholarly research communication, and then a technical discussion of the practical implications of literature and usage data considered as computable objects, using arXiv as exemplar.

Some of these algorithms scale to larger data sets.

Biography:

Institute-Industry-Innovation Seminar: Michael Schlee and Mayur Thakur, Goldman Sachs

Thursday, March 26, 2015 - 6:00pm to 7:00pm
Davis Auditorium, Room 412, Shapiro SEPSR
530 West 120th Street
New York, NY 10027
United States

Leveraging Big Data Analytics for Compliance in Financial Institutions

Leveraging Big Data Analytics for Compliance in Financial Institutions with Goldman Sachs

It is critical for a financial institution to comply with government regulations. The cost of non-compliance can result in criminal indictment, multi-billion dollar fines and loss of banking and other licenses. Employees of Compliance departments are responsible for implementation of proper policies, procedures and monitoring to ensure compliance with regulations. This discussion will focus on how Compliance leverages large quantities of data to establish monitoring controls. In particular, we will discuss specific business problems and show how they map into problems in natural language processing, outlier detection, and graph analytics. No prior knowledge of finance will be assumed.

Institute-Industry Innovation Seminar: Carme Artigas, Synergic Partners

Thursday, April 2, 2015 - 10:00am to 11:00am
Davis Auditorium, Room 412, Shapiro SEPSR
530 West 120th Street
New York, NY 10027
United States

Big Data and Data Science from a European Industry Perspective

Carme Artigas, Co-Founder and Partner | Synergic Partners

Big data and data science are not only revolutionizing the way in which companies process and analyze information but are also transforming business models. But are they developing at the same speed and in the same way all over the world and in all sectors? What is the maturity level of big data and data science in Europe?

A Conjugate Interior Point approach with applications to machine learning and robust inference for dynamic systems

Tuesday, February 17, 2015 - 1:10pm to 2:10pm
Columbia University
500 W. 120th St., Mudd Room 303
New York, NY 10027
United States

Many important applications can be formulated as large-scale optimization problems, including classification in machine learning, data assimilation in weather prediction, inverse problems, and medical and seismic imaging. While first-order methods have proven widely successful in recent years, recent developments suggest that matrix-free second-order methods, such as interior-point methods, can be competitive.

Uncertainty Quantification Framework for Modeling Prediction

Friday, October 10, 2014 - 11:00am to 12:00pm
233 Mudd
Columbia University
New York, NY 10027
United States

A methodology of uncertainty quantification developed in a series of studies and termed Bound-to-Bound Data Collaboration (abbreviated to B2B-DC) will be presented. B2B-DC is framework for combining models and training data from multiple sources to explore their collective information content. It is built on an underlying physical process and associated model, a collection of experimental observations with specified uncertainties, algebraic surrogate models (response surfaces) representing parametric dependence of the physical-model predictions of the experimental observables on the uncertain parameters, and specialized constrained-optimization algorithms. The methodology makes predictions on the true feasible set, transfers the uncertainties of both model parameters and training-set experiments directly into prediction, tests and quantifies consistency among data and models, explores sources of inconsistency, discriminates among differing models, and enables analysis of global sensitivities of uncertainty in prediction to the uncertainties in data and model. Applications of the approach include combustion science and engineering, atmospheric chemistry, and system biology.

 

Michael Frenklach
Department of Mechanical Engineering
University of California at Berkeley
 
Michael Frenklach is Professor in the Department of Mechanical Engineering of the University of California at Berkeley. He received his Diploma in Chemical Technology from the Mendeleyev Russian Chemical-Technological University (Moscow, Russia) in 1969 and his Ph.D. in Physical Chemistry at Hebrew University (Jerusalem, Israel) in 1976. Professor Frenklach’s faculty appointments began in 1979 in the Department of Chemical Engineering at Louisiana State University. He received the Alexander von Humboldt Research Fellowship and spent a year in the Institute of Physical Chemistry at Heidelberg University (Germany). In 1985 he joined the Materials Science Department of the Pennsylvania State University and in 1995 he accepted his current position at Berkeley. Professor Frenklach’s research interests are in the areas of soot formation, diamond synthesis, interstellar dust, kinetic modeling of complex reaction networks, and currently uncertainty quantification and cyber-automation of collaborative science.
 

Pages

Subscribe to RSS - Seminar
Back to Top