Scientific arbitrage

In many of my research projects, I spend a nontrivial amount of my time wondering if I am reinventing the wheel.   I try  to make sure that what I’m trying to do hasn’t already been done but this is not always simple because  a solution from another field may be hidden in another form using unfamiliar jargon and concepts.  Hence, I think that there is a huge opportunity out there for scientific arbitrage, where people can look for open problems that can be easily solved by adapting solutions from other areas.  One could argue that my own research program is a form of arbitrage since I use methods of applied mathematics and theoretical physics to tackle problems in biology. However, generally in my work, the problem comes first and then I look for the best tool to use rather than specifically work on problems that are open to arbitrage.

I’m certain that some fields will be more amenable to arbitrage than others. My guess is that fields that are very vertical like pure mathematics and theoretical physics will be less susceptible  because many people have thought about the same problem and have tried all of the available techniques. Breakthroughs in these fields will generally require novel ideas that build upon previous ones, such as in the recent proofs of the Poincare Conjecture and the Kepler sphere packing problem.   Using economics language, these fields are efficient. Ironically, economics itself may be a field  that is not as efficient and be open to arbitrage since many of the standard models, such as for supply and demand, are based on reaching an equilibrium.  It seems like a dose of dynamical systems may be in order.


I think a nice example of scientific arbitrage is the Hopfield network.  John Hopfield was already a very famous physicist when he published his two famous papers on the topic in 1982 and 1984.  People had already been studying neural networks based on the McCullough-Pitts neuron for several decades before Hopfield came along and changed the field.  Basically, these networks consist of two-state neurons that can be either “on” or “off”.   Each neuron is connected to other neurons with some pair dependent weight.  A neuron is on when the  weighted sum of active neurons connected to it exceeds a threshold and is off otherwise.  The goal was to figure out how to make such networks perform functions such as pattern recognition, input classification and associative memory.

Hopfield’s brilliant move was to recognize that a neural network looks very much like the Ising model of statistical mechanics.    The Ising mode is a simplified model of a magnetic system where two-state spins (up or down) interact.  The energy of the system is lower when the spins are aligned then when they are opposite.  The global ground state of the system can then be found by finding the state of minimal energy.  If the connections between the spins are symmetric then the minimum energy is bounded and there is always a ground state (i.e. global attractor).  Hopfield realized that the neurons in a neural network were like spins in an Ising model so that a network with symmetric connections also had a bounded energy  or Lyapunov function as it is known in dynamical systems theory.  The Hopfield network is then guaranteed to have stable attractors and all initial conditions will flow to them.  This is why it acts like a content addressable or associative memory.  Any initial state that is near an attractor  (i.e. incomplete or partial memory) will flow to the memory state, e.g. smelling apple pie makes you remember Grandma.  Hopfield’s arbitrage revolutionized neuroscience and it led to an influx of ideas  and people from physics.



7 thoughts on “Scientific arbitrage

  1. I’ve been thinking a lot about this very thing lately, and wondering whether one might use computer search to find such linkages. Specifically, with all the growing amounts of searchable structured and unstructured data out there, whether it might be possible to model the features of a problem in one’s field in a particular way so as to use the model to search for analogues to it in the data of other fields, hoping for a connection or breakthrough that solves one’s problem or sets a new direction or extends ones path. It would require some creativity as to how to build a model for a search of this kind (for instance, it may well not be at all the same as a simulation model).


  2. It would require some creativity as to how to build a model for a search of this kind

    Oh! Dear!
    You guys are stumbling on the “ontology problem”.

    To relate different documents pertaining to the same domain or topic you need ontologies to describe the concepts within, and you actually need more you need A common ontology for the whole realm you are interested in.
    Remember the Semantic Web?
    Why didn’t it really took off?

    [PDF] Abstract. Despite the potential of domain ontologies to provide consensual representations of domain-relevant knowledge, the open, distributed and decentralized nature of the Semantic Web means that individuals will rarely, if ever, countenance a common set of terminological and representational commitments during the ontology design process. More often than not, differences between ontologies are likely to occur, and this is the case even when the ontologies describe identical or overlapping domains of interest. Differences between ontologies are often referred to as ontology mismatches and there is an extensive research literature geared towards the technology-mediated reconciliation of such mismatches. Our approach in the current paper is not to comment on the relative merits or demerits of the various technological solutions that could be used to resolve ontological differences; rather, we aim to explore the reasons why such differences may arise in the first place. In addition to a review of the various factors that contribute to ontology mismatches on the Semantic Web, we also discuss a number of focus areas for future research in this area. An improved understanding of the origins of ontology mismatches will, we argue, complement existing research into semantic integration techniques. In particular, by understanding more about the complex cognitive, epistemic and socio-cultural factors associated with the ontology development process, we may be able to develop knowledge acquisition and modeling tools/techniques that attenuate the impact of ontology mismatches for large-scale information sharing and data integration on the Semantic Web. Keywords: semantic web, ontology, owl, semantic integration, ontology alignment, human cognition, ontology mismatches, ontology reconciliation, conceptual processing, knowledge acquisition techniques.

    I think ontology reconciliation is plainly impossible, because even during the lifetime of a single ontology any update destroys some information: the meaning of the concepts drift (like in any natural language) and the (current) logic of ontology usage is too crude to cope.
    Ontologies aren’t an AI solution they are an AI problem.


Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s