Colin Twomey's Research

About

I'm the executive director of the Data Driven Discovery Initiative — the University of Pennsylvania's hub for data science research and education in the School of Arts & Sciences. I'm also the founder and director of the University Atlas project (uAtlas) — an effort to map the research world at Penn and beyond (read more in Penn Today).

I started at Penn as a MindCORE Postdoctoral Research Fellow working with Joshua Plotkin on inference for cultural evolution. Prior to Penn, I did my Ph.D. on collective behavior at Princeton University with Iain Couzin. As an undergraduate, I studied computer science as a Goldwater Scholar at Colgate University and worked on combinatorial optimization as a Fulbright Fellow at the Université Libre de Bruxelles. In industry I've worked as a software developer at Sun Microsystems and as a data science consultant for the World Bank and agtech startup Arable Labs.

Contact

email: crtwomey{at}sas{dot}upenn{dot}edu
mail: Colin Twomey
Data Driven Discovery Initiative
University of Pennsylvania
Philadelphia, PA 19103

Research

What are the natural algorithms underlying collective behaviors? How do persistent patterns and properties of a group emerge from the repeated interactions of its constituents?

The collective motion of a fish school isn't determined by any one leader, but emerges from the attraction, repulsion, and alignment of individuals throughout the group. Common to all animals is the use of sensory systems to make informed decisions, and in my research I investigate in particular the role that sensory systems play in shaping and constraining collective behavior.

Behavioral cascades

Startle responses in fish schools are critical for avoiding predation. A single individual startling can induce a wave of startle responses across a group. What sensory information, social and non-social, makes an individual more or less likely to startle in response to the startle of a neighbor?

Answering this question allows us to uncover the network of sensory information in a group, and to say when and how information may propagate through a school at any given moment.
Within-group structure

What are the right levels of organization to look at to understand collective behavior? Collective properties at the group level may not always be best explained by interactions at the individual level. Instead, we should ask whether or not there may be a simpler, mesoscale descriptions that allow strongly interacting components to be considered as a unit, and investigate the weak but non-negligible couplings between these intermediate-scale and possibly ephemeral subgroups.
Collective cognition

Consensus decisions are a common feature of groups in the animal world. In humans, the words we use to communicate with others are themselves a product of a collective consensus process. Despite the apparent flexibility of such a process, there are surprising instances in which vocabularies of independent linguistic origin converge to very similar representations. The most famous of these are words for describing color. While languages vary in the number of words and the ways in which these words partition the space of visible light, they do so in a remarkably constrained way.

Why should there so often be simple and intelligible mappings between the color words of completely unrelated languages? In my work, I investigate how this may arise from a collective consensus process that is fundamentally constrained by the shared physiology of our perception of color. This has implications for how other animal groups may arrive at shared, learned representations of stimuli, even in the absence of a shared vocabulary.