SPSP 2013 Contributed Papers: Computation and Simulation

SPSP 2013 Contributed Papers: Computation and Simulation

Tweeted from the 4th biennial conference of the Society for Philosophy of Science in Practice in Toronto, Ontario, Canada, on June 29, 2013, during Concurrent Sessions VII

  1. First up, Catherine Stinson, “Computational models as experimental systems” #SPSP2013 #SPSP2013Toronto

  2. The problem: how can we learn (about brain & cognition) from (connectionist) computational models? #SPSP2013 #SPSP2013Toronto
  3. Scientists seem to think that they’re learning *something* from “computational experiments” #SPSP2013 #SPSP2013Toronto
  4. Maybe they’re wrong, or using the vocabulary differently. But hypothesis is that they know what they’re doing #SPSP2013 #SPSP2013Toronto
  5. What the scientists say they’re doing? Models closely tied to the physiology of the brain; #SPSP2013 #SPSP2013Toronto
  6. Models have a ” ‘physiological’ flavor” (why the scare-quotes) #SPSP2013 #SPSP2013Toronto
  7. Trying to get a handle on real causes; simplifying to identify the thing that’s more important to study #SPSP2013 #SPSP2013Toronto
  8. [As I’m late to the Storify on this session, there are some tweets that I must copy and paste. I’m not sure whether to blame Storify or Twitter for this, but the moral of the story is not to dawdle in Storifying one’s conference tweets.]
  9. Two goals: understand or *mimic* brain. RT : hypothesis..they know what they’re doing

    7:38 AM – 29 Jun 13 · Details
  10. Computer/Cog Sci criticized often by confusing the two goals. (Sometimes justifiably) RT @docfreeride: #SPSP2013 #SPSP2013Toronto

  11. @drugmonkeyblog @docfreeride There are problems with tuning complex models to produce recognizable patterns … “right” output wrong reasons

  12. Right…if trying to learn about human brain. If trying to improve information processing….can be fine. @N3OX @docfreeride

  13. @drugmonkeyblog @N3OX @docfreeride Useful distinction. We may find many interesting things trying to model the brain, even if models wrong.

  14. @drugmonkeyblog @docfreeride Yeah, and I think even if you ARE trying to understand the brain, it’s fine if skepticism levels stay high.

  15. @IdiotTracker @drugmonkeyblog @docfreeride Right. Lot of promise in new things that are “bio-inspired” even if they work differently.

  16. But of course the need to justify work for funding purposes leads to interesting claims… @N3OX @IdiotTracker @docfreeride

  17. #Modelorg RT @drugmonkeyblog: The need to justify work for funding purposes leads to interesting claims… @N3OX @IdiotTracker @docfreeride
  18. Generic mechanism (generic = a thing in the world, belongs to an abstract type) #SPSP2013 #SPSP2013Toronto
  19. models have explanatory power in virtue of being simplified; generalizations as a result (not an assumption) #SPSP2013 #SPSP2013Toronto
  20. Scale models as example: generic system of particular shapes, relative sizes. Do expt. to see how things fit #SPSP2013 #SPSP2013Toronto
  21. Another generic model: a sample from a target population. Sample based on generic of interest #SPSP2013 #SPSP2013Toronto
  22. (& work to mitigate effects of other generics that they are) #SPSP2013 #SPSP2013Toronto
  23. Connectionist models: interconnected nodes supposed to be like a brain in terms of connections of diff strengths #SPSP2013 #SPSP2013Toronto
  24. Experiments with the model (meant to show how past tenses of English verbs could be conjugated with one system) #SPSP2013 #SPSP2013Toronto
  25. (Am not tweeting the picture w/ the nodes, because I’m not that good, yo!) #SPSP2013 #SPSP2013Toronto
  26. Some of the idealizations might be actual problems (does training data help too much in getting the result?) #SPSP2013 #SPSP2013Toronto
  27. Advertised as showing patterns of language learning that young English speakers go through #SPSP2013 #SPSP2013Toronto
  28. You don’t use back-propagation in situations where that idealization would clearly interfere with answering the Q #SPSP2013 #SPSP2013Toronto
  29. Models as concrete objects we can manipulate to generate knowledge #SPSP2013 #SPSP2013Toronto
  30. Inferences from models needn’t involve the logic of representation (can use others kinds) #SPSP2013 #SPSP2013Toronto
  31. Q: how can we talk about idealization without still being in the neighborhood of representation? #SPSP2013 #SPSP2013Toronto
  32. Next: Miles MacLeod (& N Nersessian), “Integrating simulation & experiment: hybrid research in systems biology” #SPSP2013 #SPSP2013Toronto
  33. Actually, “Bimodal research in integrative systems biology” #SPSP2013 #SPSP2013Toronto
  34. DRawing on ethnographic study of model-building practice in 2 systems biology labs#SPSP2013 #SPSP2013Toronto
  35. Hypothesis: cellular redox environments affect drug efficacy #SPSP2013 #SPSP2013Toronto
  36. Lab G (computational lab): contains only modelers. Lab C (fully equipped wet-lab) w/ experimenters, modelers, …#SPSP2013 #SPSP2013Toronto
  37. … and researchers who do both expts & modeling (bimodal researchers)#SPSP2013 #SPSP2013Toronto
  38. Model-building in ISB characterized by complex problem-solving tasks #SPSP2013 #SPSP2013Toronto
  39. Complex nonlinear biological networks, model constrained by data, collaboration issues, and computation issues #SPSP2013 #SPSP2013Toronto
  40. Trying to understand insensitivity of diff cancer cell lines to chemo drugs #SPSP2013 #SPSP2013Toronto
  41. Hypothesis: cellular redox environments affect drug efficacy #SPSP2013 #SPSP2013Toronto
  42. Original model criticized as “very far from what occurs physiologically” #SPSP2013 #SPSP2013Toronto
  43. Refinement: not just one pathway, but whole redox pathway (ODE model built from lit.) #SPSP2013 #SPSP2013Toronto
  44. How does the drug alter the redox environment? Model didn’t match phenomena in cell lines #SPSP2013 #SPSP2013Toronto
  45. Isolated problem by experiment. The toxicity had to be from diff mech #SPSP2013 #SPSP2013Toronto
  46. Simulated hypothetical mechanisms, ran simulations to estimate unknown parameters, compared with experiments #SPSP2013 #SPSP2013Toronto
  47. Coupling experiment and simulation (bimodal work) as good modeling strategy #SPSP2013 #SPSP2013Toronto
  48. Simulation was main cognitive apparatus for interpreting data & dynamical relationships, structuring hypotheses #SPSP2013 #SPSP2013Toronto
  49. Managing complexity through triangulation, limits search space by localizing errors #SPSP2013 #SPSP2013Toronto
  50. Pure modelers rely more on mathematical/algorithmic techniques to search #SPSP2013 #SPSP2013Toronto
  51. Bimodal researchers can only handle relatively small scale systems w/ manageable # of unknowns #SPSP2013 #SPSP2013Toronto
  52. But bimodal researchers can produce more reliable “biologically grounded” models (mechs) #SPSP2013 #SPSP2013Toronto
  53. Last paper, Annamaria Carusi, “Computational models and medical science: the honeymoon is over” #SPSP2013 #SPSP2013Toronto
  54. How are new encounters, particularly in clinic, challenging systems biology? #SPSP2013 #SPSP2013Toronto
  55. Lessons for epistemology of modeling & simulating in a biomedical context? #SPSP2013 #SPSP2013Toronto
  56. “Systems Biology” is very broad, could point to many different kinds of approaches… #SPSP2013 #SPSP2013Toronto
  57. Talk focused on a particular example in systems physiology (modeling w/ PDEs & ODEs) #SPSP2013 #SPSP2013Toronto
  58. Disenchantment. Much funding on promise of building models that would be useful to medicine. #SPSP2013 #SPSP2013Toronto
  59. New collaboration in face of skepticism; new approach to validation #SPSP2013 #SPSP2013Toronto
  60. Major effort put into model construction, simulation techniques, for multi-scale integration #SPSP2013 #SPSP2013Toronto
  61. Not so many new experiments in response to/in conversation w/ models #SPSP2013 #SPSP2013Toronto
  62. Role of simulations defn’d mathematically: solving equations. Simulations are evaluated on those grounds #SPSP2013 #SPSP2013Toronto
  63. Philosophy of modeling/simulation? Simulations as downward, motley, and autonomous #SPSP2013 #SPSP2013Toronto
  64. Validity, legitimacy, justification, evaluation of simulations can be internal or external #SPSP2013 #SPSP2013Toronto
  65. Internal: relationship between the equations and the simulations #SPSP2013 #SPSP2013Toronto
  66. External: what makes this simulation a “model” of the target domain #SPSP2013 #SPSP2013Toronto
  67. Legitimacy grounded in certain features of model building practice. Trust in reliability of those practices #SPSP2013 #SPSP2013Toronto
  68. Modeler working too much in isolation from experimentalist, not perceptive to tinkering/tweaking relationship #SPSP2013 #SPSP2013Toronto
  69. Models are not single entities but hybrid systems, model-simulation-experiment (MSE) system #SPSP2013 #SPSP2013Toronto
  70. Makes personalized medicine a pipe dream for many kinds of systems biology #SPSP2013 #SPSP2013Toronto
  71. Challenge of variability: don’t know if simulation & experiment are in the same state when try to compare them #SPSP2013 #SPSP2013Toronto
  72. Modeling & simulating can (should?) be brought in closer contact w/ expts #SPSP2013 #SPSP2013Toronto
  73. Trying to see what counts as a comparison, set up conditions where it’s possible to make them #SPSP2013 #SPSP2013Toronto
  74. Coupling experiment and simulation (bimodal work) as good modeling strategy #SPSP2013 #SPSP2013Toronto

Did you find this story interesting? Be the first to
or comment.

Liked!
facebooktwittergoogle_pluslinkedinmail
Posted in Biology, Conferences, Mathematics, Medicine, Methodology, Minds and/or brains, Philosophy.

3 Comments

  1. wtf…..a computational bashing session @SPSP????…as one of these trying to be a computational biologist peoples, i wonder if anyone at the symposium actually had any insight as to how one is to make and test models of neural systems without futzing around with a connectivity matrix and trying to do closed-loop neurophysiologic experiments???….a single neuron’s dynamics are simple to describe quantitatively, but the emergent properties of the network lie in the details of the connectivity matrix that ties any model together…..

    it seems really obvious (to me at least) that those who have existed in this space thus far haven’t figured it out yet because of the enormous gap between knowing how to do the ‘magic’ of biology and the theory and practice of being a computationalist……so most of the computational work out there thus far seems to fall into the extremes of being overly-parameterized or overly-generalized…..both being of little use until we can really measure these properties precisely and directly from the biologic preparations…..

    but imo simply having a bash session doesn’t seem like it would lead to any meaningful solutions……i hope this is just a consequence of the twitt-world and the symposium had more substance than im aware of….

  2. I love your blog, but please consider putting VERY long tweet streams after a jump, for those of us who read the blog on the website (not via RSS) and otherwise have to scroll endlessly to the next post. =)

  3. As a “bimodal” neuroscientist, I want to point out that just as with wet experiments, there is good computational practice and there is bad computational practice. Testing the model for robustness to parameter variation is good practice and limits over-parameterization. Validating the model against experimental data is also good practice and limits over-generalization.

Leave a Reply

Your email address will not be published. Required fields are marked *