 |
 |
 |
 |
 |
|
|
|||||||||
 |
||||||||||
|
|
Neural Networks
WEEK 1 READINGS: Pick one from each of the following sets
WEEK 2 READINGS: Pick one from each of the following sets
Introduction: As most of you know, the general theme of “the mind” was probably farthest from what I typically work on and think about. As such, I’ve been conflicted about what direction I’d end up taking with this all year – other than that it would have something to do with “neural networks.” While I have been to a couple of conferences where people have presented work on neural networks, I previously had not tried too hard to interpret what they were doing. They spoke a different language of networks, and even visually displayed them differently than what I am used to. So, the nature of the first week of these two is going focus a little more on establishing the state of a literature than we have done in previous weeks. While prepping for this, I can’t believe how much of this stuff I’ve ended up reading because I found it interesting – hopefully you do to.
As a self-indulgent side-note – I spent most of my life trying my best to avoid learning anything about biology (much to the disappointment of my dad & step-mom who are both veterinarians). Also, I failed one math class in undergrad (much to the disappointment of my mom, who was a math major in college) – one on linear algebra (all of the matrix algebra that underpins the primary “stuff” of networks – you know…the only class I took as an undergrad that has any bearing on what I do today - Go figure!). So, in adopting the theme of “the mind” then somehow agreeing on the topic of neural networks – preparing for this seminar has successfully forced me to revisit some of my most painful educational experiences. As such, my sense of efficacy is at an all-time low. Hopefully you all can help our discussion overcome this – especially in the first week.
Week 1 – What is a “neural network” and (why) do we care? My main interest in this week is trying to figure out why things happened the way they did in the rat study that Maria had us read (if you need/want a refresher on that one, there was a summary of it from Albert 2007, NEJM – available here). I wanted to stop our conversation those weeks several times & ask “But, how does the brain work?” Instead, I saved those questions for my seminar. So we will spend part of this week figuring out what a neural network is & the basics of how we think they function. It was unclear to me if there is a consensus model, and I am ill-equipped to derive one. Either way, to avoid turning this session into something entirely didactic, I think it may be useful to not focus solely how we think brains work, but to spend more of our time working out how some of the things we have talked about in previous sessions may be altered given the various possibilities available currently. If nothing else, this will be helpful in setting up week 2’s discussion.
To kick things off, I’ll probably spend a little bit of time explaining the concept of “block-models” as they are generically used in networks research. Though I haven’t yet found a direct analogue in what I’ve read on neural nets, I think there is something there & I will try to draw some analogues to possibilities in the brain that we can then use to launch the rest of the discussion.
The aim will be to build from a few models of how we think the brain organizes what we “learn” and how we “use” things we’ve learned. In essence the question has two parts – do we care about the differences in the varying models (for the things we study – the “in this room” we, not the generic we); and if so, how would they vary the way we tackle the things we care about. In particular, it may be helpful to think about what these models (may) mean for some of the other things that we have talked about in the course of the year (I’m open to using it to discuss other topics that are of interest to you if you think they’ll fit).
Since Maria’s seminar largely sparked the way I have been thinking about this week – in particular the rat study she had us read – I’ll use that for an example of what direction I see this potentially taking. One of the specific questions we addressed that week was how convinced we were of the “beneficial effect of cognitive engagement on cognitive aging.” For the moment, let’s assume we were convinced, and let’s think explicitly about the implicit model of memory we were building on. Given the model we saw that week, what are the some of the “down the chain” implications if we vary the assumptions of “how the brain works” in the model. Specifically, let’s suppose for a moment that what their intervention did was to target the “deadening” of particular individual neurons. How do we think this matters for the way we subsequently interpreted the usefulness of this study? What about if they were inhibiting communication between specific neuronal pairs. How about “knocking out” particular clusters of neurons? Or inhibiting communication between two paired “classes” of neurons? Same question – but networks within/across clusters/regions? In essence, I want to revisit another question Maria asked – “What exactly might cognitive engagement do (i.e. structural changes versus how you “use” your brain? Neurogenesis, more synapses, or more “efficient” use of the networks you already have)?” (now that I feel like I have some information about the brain to actually discuss this some) – but instead of focusing on what it may do, talk about whether and how differences would matter.
This is just one way I have been orienting my thinking about this week’s seminar. If you have other directions you’d like to take things in – feel free. If you want to email me those, or questions pertaining to what I have here in advance, that may be helpful.
(The background readings listed for this week may be helpful if you don’t have much sense at all of what a neural network is. I didn’t. The Russell piece is probably the more helpful on that (it’s written towards an undergrad audience, a feature that I found really helpful). The other is a little more partial, and a bit dated, but I found it a fun read. There is also a “What have we learned since 1943” companion piece to it, directly following it’s reprint in the 1990 issue, if you’re curious.)
Week 2 – Applying the Neural Network model to Scientific Production By now you’ve probably looked at the readings list, and have wondered what these have to do with what we did last week? This week is going to be a complete shift in gears topically. But the ways that neural network models have been applied are about as wide-ranging as you can imagine, and I’d like to take advantage of that fact to turn this week into a bit more of a meta-commentary on the way knowledge is produced, consumed and implemented in the study of population health. While – given the nature of the H&SS program – we may all value some level of interdiciplinarity, the aim here is to explicitly spend some time working out the implications that disciplinary/inter-disciplinary/trans-disciplinary approaches can have – both positive and negative – on what we study and what we find in that study.
In essence, my aim this week is for us to talk about our perspectives of the “ought”s of scientific production. As a launching point for this discussion - rather than just opening it up to a free-for-all - let’s at least start by thinking of population health research in a neural network frame. (The optional reading explicitly addresses this link, if you’d like it spelled out a bit.)
In this framework, I think that we can conceptualize the addition or removal of nodes and edges – which were alluded to in the first week – a little more concretely. If we start from our perceptions of the existing state of health research (unfortunately I didn’t come across a study that looked at this explicitly – though if someone knows of one before we get to week 2, we can substitute it for some of the others I included), we can talk about adding/removing nodes (whether individual, grouped, or “blocked”) that correspond to authors, papers, journals, schools, etc. Or we can talk about the adding/removing edges, which may correspond to citations, co-authorships, faculty hires, recommendation letters etc. There are also things that fit into this setting, but it’s a little less clear which role they fill – e.g., funding calls, conferences, etc.
I am open to this conversation going wherever you’d like it to go as a group, but think that the network frame gives us a good place to start from, and at least some orientation to so,e questions of interest. A few potential starters: The nature of the field, w/respect to:Current state, past incarnations, future projections How “learning” happens – from the networked field perspective How perturbations (think additions and subtractions of nodes/blocks/regions and edges) affect this, if at all. Building from that, how it informs where we’d like to position ourselves (in terms of what we produce/consume/”gate keep”, etc.) w/regard to: Our “home disciplines” Disciplinary/transdiciplinary/interdisciplinary approaches
|
|||||||||
 |
||||||||||
|
||||||||||
 |
||||||||||
|
||||||||||
