Sunday, 8 June 2014



Bursts, Cascades, and Time Allocation




ADILSON MOTTER
Northwestern University
Dynamics of Complex Systems and Networks Group

VIDEO

OVERVIEW: In this talk, I will present recent results on three distinct but related problems concerning Web Science and the Mind: bursts in the temporal distribution of words, cascading dynamics in diverse network systems, and human allocation of time. In each case I will discuss key properties, the principles governing these properties, and opportunities their modeling offers for monitoring and controlling complex behavior.

READINGS:
    Cornelius, S. P., Kath, W. L., & Motter, A. E. (2013). Realistic control of network dynamicsNature communications4:1942
    Altmann, E. G., Pierrehumbert, J. B., & Motter, A. E. (2009). 
Beyond word frequency: Bursts, lulls, and scaling in the temporal distributions of wordsPLoS One4(11), e7678.
    Motter A. E.. &  Albert R. (2012), Networks in motion  Physics Today 65(4), 43-48
.

33 comments:

  1. Fascinating talk. I found the idea of exploiting the system’s endogenous attractor dynamics in order to obtain a target state very interesting. My question for Dr. Motter concerns complex systems whose attractor dynamics vary over time dynamically. Are the state space analysis techniques described in the talk still applicable to such systems?

    ReplyDelete
    Replies
    1. This comment has been removed by the author.

      Delete
    2. The analysis applies to fixed-point attractors, limit cycles, and chaotic attractors. With minor modifications, it also applies to situations in which parameters change slowly as a function of time. More work is needed, however, to address cases in which the parameter changes are not small and induce bifurcations (e.g., changes that create, destroy or significantly modify attractors). On the other hand, parameter changes can also be exploited to control systems by modifying the basins of attraction and/or attractors, as shown in this paper: S. Sahasrabudhe and A.E. Motter, Rescuing ecosystems from extinction cascades through compensatory perturbations, Nature Communications 2, 170 (2011).

      Delete
    3. Thank you for the response and reference!

      Delete
  2. Here is an interesting article exploring the topic of how the brain recovers
    consciousness after a significant perturbation (i.e., anaesthesia).

    Recovery of consciousness is mediated by a network of discrete metastable activity states
    http://www.pnas.org/content/early/2014/06/04/1408296111.full.pdf+html?with-ds=yes

    ReplyDelete
    Replies
    1. Thanks for this interesting recent reference.

      Delete
  3. What about systems which are 'at the edge of chaos', meaning that they do not have one stable attractor, but rather are moving from one to another all the time? Are these methods presented applicable for these systems?

    ReplyDelete
    Replies
    1. Although this was not covered in my presentation, we have developed an approach to control noise-induced transitions between attractors as a tool to control network dynamics.

      Delete
  4. Dear Adilson, It was a great presentation! Do you have advices to build efficient algorithms for text mining to study dynamic of scientific terms structured in OWL-DL and that it’s possible to link with formal syntactic grammars?

    ReplyDelete
    Replies
    1. I believe Jennifer Golbeck and Jim Hendler would be able to provide you with more useful information.

      Delete
  5. This comment has been removed by the author.

    ReplyDelete
  6. Really nice talk. I asked the word meaning question; I had a follow-up question: in terms of bursts, are abstract words indifferenciable to functional words (stopwords) ?

    ReplyDelete
    Replies
    1. The best is to consider some examples:

      Class 1 words (Entities): Africa, Bible, Darwin, etc.
      Class 2 words (Predicates and Relations): blue, die, in, religion, etc.
      Class 3 words (Modifiers and Operators): believe, everyone, forty, etc.
      Class 4 words (Higher Level Operators): hence, let, supposedly, the, etc.

      Class 1 words tend to be more bursty than class 2 words, which in turn tend to be more bursty than class 3 words, which are generally more bursty than class 4 words.

      For details, I refer to Fig. 2 in E.G. Altmann, J.B. Pierrehumbert, and A.E. Motter,
 Beyond word frequency: Bursts, lulls, and scaling in the temporal distributions of words, 
PLoS One 4(11), e7678 (2009).

      Delete
    2. The concept of burstiness of word usage reminds me of the phenomenon of "vogue words" where certain terms like "paradigm" in the mid 90's become very popular and fashionable. I imagine that they follow the drawn out decline in usage the way your model describes. Interestingly, the vogue word phenomenon is also associated with semantic broadening.

      Delete
  7. I would like to know more about how we can use the techniques of analyzing complex networks on the burstiness of words. As Dr. Motter said, burstiness may be explained largely by context; the conclusion of a paper will have many more conclusion-y words (like "thus," "therefore," etc.) and thus seems pretty predictable to me-- or at least much less complex than these structurally complex and dynamic systems discussed in the presentation.

    ReplyDelete
    Replies
    1. I think I agree with you Rachel. At the moment, I don't see the importance in understanding the burstiness of word usage. This pattern of word usage seems intuitive. What can we use this knowledge for?

      Delete
    2. First please see the reply to Louis Chartrand (just above), which might clarify some of the issues. I believe "context" more strongly influences the usage of entities as compared, say, to operators. But to a smaller extent, it is expected that operators too would be influenced by "context."

      Concerning your example of the conclusion of a paper, note that the stretched-exponential inter-event distribution is observed at scales much longer than few hundred words (see Fig 1 of the paper mentioned just above).

      Delete
  8. Controlling dynamics systems to reach stability is possible according to this nice talk, but there is two things to know before this, the variables and the parameters that could be used to reach a basin attractor. The problem is: in most cases we don’t know all variables ( in very complex systems) that can influence the network state, what can we do to reduce this uncertainty!!!

    ReplyDelete
    Replies
    1. Good question. It is not essential to know all variables and parameters. Models usually identify the dominant variables and parameters, and that is all that is needed if the approach is robust, which is the case of the one I presented.

      Delete
  9. I found Dr. Motter’s comment regarding physics, chemistry, and biology interesting. I used to think that if we knew all about physics we would in turn know everything about chemistry. Dr. Motter pointed out, however, that this logic is false. We need to understand physics network interactions before fully understanding chemistry.

    I do believe, however, that upon understanding all of physics (something that will never happen) we then have the potential to uncover all to know about chemistry. I think knowing all of chemistry depends on knowing all of physics.

    ReplyDelete
    Replies
    1. I suspect people will agree or disagree with your statement depending on what exactly is meant by “all”. For example, one does not need to account for everything about elementary particles to explain most phenomena in condensed matter physics. More generally this relates to the issue of whether one expects to describe things from first principles.

      Delete
    2. Good points. We mastered classical mechanics before understanding much about quantum mechanics. Physics, however, may be particular in this regard because laws differ on quantum, classical, and astronomic scales. Chemistry and especially biology may not confront the same problem.

      Delete
  10. Thank you ADILSON MOTTER. One question: So, We can use a simple dynamic systems for time allocations. That it make sense?

    ReplyDelete
    Replies
    1. I did not have enough time to discuss time allocation in my presentation (ironically), even though this was alluded to in the title. We can discuss in person, however.

      Delete
  11. Dear Professor Motter:

    1. Could you elaborate a little on the definitional basis of the concrete/abstract distinction -- as well the functional implications of the burst dynamics?

    2. What is the basis for the criterion of "preservation of biomass?

    3. Is there a burst dynamic account of the reaching a Pareto Equilibrium (80/20).

    4. Most of your control dynamic examples were of human interventions is artificial or biological systems. Do similar processes happen naturally, e.g., in evolution?

    ReplyDelete
    Replies
    1. 1. See the reply above to Louis Chartrand and the reference therein, which should clarify your question.

      2. You might be referring to the minimization of biodiversity loss (at least if in the context of ecological networks). The motivation is that this is an objective function of interest in practice, which nevertheless can be modified to, e.g., assign different weights to different species.

      3. This may depend on the context, but was not a point discussed in my work. I did mention, however, that the empirically-determined hazard function follows a power law, leading to a stretched exponential for the inter-event distribution.

      4. The short answer is yes, as discussed in person after your posting. In the case of intracellular networks, this problem is addressed in the following references: S.P. Cornelius, J.S. Lee, and A.E. Motter, Dispensability of Escherichia coli's latent pathways, Proc. Natl. Acad. Sci. USA 108, 3124 (2011); T. Nishikawa, N. Gulbahce, and A.E. Motter, Spontaneous reaction silencing in metabolic optimization, PLoS Computational Biology 4(12), e1000236 (2008).



      Delete
  12. I thought it was very poignant how Professor Motter articulated that different network dynamics come into play at different scales. I think this type of thinking is particularly relevant to the brain, where it's not individual neurons that we need to study for understanding brain behaviour, but the dynamics of the whole system. I wonder whether it makes sense to think of brain states as tending towards stable attractors, too.

    ReplyDelete
    Replies
    1. Good question, particularly in view of population coding and collective dynamics involving macroscopic regions. I believe useful insights are provided by the literature on attractors and metastable states in the context of brain/neuronal networks.

      Delete
  13. I understood that according to DR. Motter, to know control theory is to know mind control, is this correct? Could we do an analogy such as the control theory consists in a controller manipulating a system as brain is the controller of the body? If this is incorrect what kind of analogy could be done?

    ReplyDelete
    Replies
    1. The brain controlling the body is a good analogy to the extent that the sensory system collects information which is processed to, e.g., implement motion control. That said, I think there is a big distance between my presentation and implications for mind control.

      Delete
  14. This comment has been removed by the author.

    ReplyDelete
  15. Thank you for a very interesting talk. In the first part of the talk you described a method of intervention based on forecasting adjacent trajectories that seem to get closer to a basin. Is this something that can be done in real time? What is the knowledge which is necessary to apply such method? Do you need an estimation of the boundaries of basins of convergence? Do you assume knowledge of all the attractors of the system?

    ReplyDelete