Agent based models allow to approximate the behaviour of complex systems under certain scenario conditions where other popular methods cannot - especially in case of non-linear systems displaying emergent, unpredictable behavior. ABM can also be used to approximate solutions to difficult problems occurring in complex systems.
ABM has been used successfully in various Data Science target fields, from predicting customer behavior, optimizing supply chain, to modelling social network effects and the human immune system.
6.30 Welcome, networking + free beer + pizza
7.00pm Talks start
"Technical Challenges of Real-World Agent-Based Modelling," Thomas French & Benjamin Herd, Software Engineers at Sandtable
Some developers will be aware of ABM through simple models created in NetLogo, or may have seen examples of iconic models such as boids. The technique is appealing because it allows us to see how complex behaviours can emerge from a few simple rules. It gives us hope that complex phenomena - such as the behaviour of markets and organisations - can be broken down into easily understandable, interacting, constituent parts.
Our experience is that when you start to build models using real world data about human behaviour and the rules that govern it, modelling itself becomes very complex, very quickly. This complexity manifests itself through a number of challenges:
- How can we trust models of this degree of complexity
- How can we understand what is going on in our models - and validate them?
- How do we deal with performance issues for complex simulations - both in terms of running them and processing the data they produce?
In this talk Thomas and Benjamin from Sandtable, a specialist ABM company based in London, share their experience of addressing these challenges in a demanding commercial context.
"Agent Based Methods. Modelling Consumer Choice" by Adi Andrei, Data Scientist & Co-Founder at Technosophics
Markets, economies, societies and networks are inherently complex dynamic systems.These systems involve a collection of individual agents (consumers, organizations, interest groups, computers etc.) which interact with each other following some overall well known rules of engagement.
Traditional economic methods used to study such populations (macro, micro, and more recently machine-learning approaches) do not capture the heterogeneity and intra-population interactions very well, nor do they allow for emergent phenomena. Also, whenever their models attempt to be “realistic” - in order to take account of real-world interactions - they become analytically intractable.
Agent based modelling applied to populations and processes modeled as a dynamic system of interacting agents - has the potential of dealing with these issues in an effective manner. We will present this in the context of modelling consumer choice. You can learn more about Adi's research on ABM here
9.30pm-ish session ends