Title: Designing interventions to control the spread of epidemics
Speaker: Anil Vullikanti
Time: 2:30 PM
Abstract: A common problem in epidemiology is to design interventions for controlling the spread of a disease. There are diverse kinds of interventions, including vaccinations and social distancing, which turn out to be natural resource constrained optimization problems. Even under very simplified settings, developing effective algorithms is challenging, and there are few results
known, which give provable bounds relative to the optimum. We will discuss the simplest kind of vaccination problems, and discuss some rigorous results using methods from stochastic optimization. We will also touch upon practical heuristics for these problems, along with methods to evaluate their performance.
Title: Advanced analytics over heterogeneous datatypes in VERSA: A functional and declarative engine for socio-ecological agent-based system
Speaker: John Krulick
Time: 3:15 PM
Abstract: VERSA is a Python based framework to support agent-based workflows such modeling, interventions, analysis over experiment design output. The use of functional paradigm (separation of data from operators) and support of high (domain science) level operators allows users to express complex analytics using simple direct expressions. Overall, the framework enables users to write correct, robust code that is also easy to read and reuse.
In this talk, I will discuss extending support for heterogeneous data types within VERSA.
Most common operators over these data types are supported including aggregation, intersection, union and such. Various higher order types and custom operators
such as fishnet, interval join, points in fishnet etc. are defined.
In addition to providing type and operators, VERSA has been extended with features to facilitate building analytics workflow:
1) a wrapper layer over Python's visualization engine (matplotlib, seaborn, etc.)
2) Schema patterns are used to reduce data preparation steps
3) Materialization of intermediate results improves development time and fault tolerance by caching expensive operations.
4) support of multiple data engines (monetdb, dataframes, voltdb).