neuroscience - The concept of backpropagation in neural networks actually occurs in the brain?

I am aware the Human brain has many functionally distinct components, but let us specifically consider the Human visual cortex: could Artificial Neural Networks (ANNs) be "trained" (through, e.g. backpropagation) in an analogous way to how the visual cortex "learns"?

Is the concept of backpropagation in ANNs a phenomena actually observed in the Human brain?

Related

Answer

I recommend Yoshua Bengio's recent works. E.g.: https://arxiv.org/abs/1502.04156 and his slides from the NIPS 2016 Brains and Bits workshop.

Also, Timothy Lillicrap's work: http://www.nature.com/articles/ncomms13276

This is still a big open question. In short, we (the neuroscience community) have little idea on how the brain learns at a circuit/systems level in general. We know a thing or two about how individual synapse changes under specific experimental protocols, but those are phenomenological models (e.g. STDP; there has been many many normative models of learning that concludes their learning rule is STDP-like). (Some systems such as cerebellum, we know relatively more, but the learning seems to be more specific to motor timing.)

The main technique we use in machine learning and artificial neural network are mostly gradient descent (or some sort of stochastic optimization algorithm), and people have been trying to find if the biological neural network learns in a similar way. There are several issues:

1. 
   

   There's no error gradient being propagating according to chain rule. Information gets mushed together within a neuron, so there's no plausible way to trace exactly the forward-path during an backpropagation.

   
   



2. 
   

   There's a lot of recurrent connections but they are not symmetrical.

   
   


3. 
   

   Temporally, it is implausible that there's a clocked signal to backpropagate after a each forward path. Neuronal network doesn't seem to operate with clocks (some thing there are precise clocks, but I think there's little evidence).

   
   

Recent efforts have made some progress with random backprojection, decoupled updates using synthetic gradients, target prop, autoencoder, weight/signal quantization, deep belief networks, etc. But there seems to be several more jumps needed to make it biologically plausible such that it agrees with the neural architecture and information flow. Even then, there's no guarantee that neocortex is learning with a similar principle.

Can molecular genetics make a boolean variable from a continuous variable?

In the same kind of idea than this question. Gene expression are regulated through complex interactions. The concentration of enhancers and repressors is an important aspect that dictate the level of expression of a given gene. These concentrations can take different value on a continuous scale.

Imagine a case where fitness is maximized when a given gene produce n proteins per minute if the concentration of a given protein is greater than x. If the concentration of the protein was to be lower than x, then the gene should not be expressed (0 proteins per minute are produced). In such case, it would be great if a bunch of reactants were to be able to simulate a "switch function" that would switch from "NO EXPRESSION" to "EXPRESSION" at x.

It seems to me that such switch function should be very complicated to evolve. I would suspect that all chemical reactions, including the binding of enhancer to promoter region should follow the law of Michaelis-Menten and the Michaelis-Menten function is not at all switch function. So, I have been thinking about cooperative binding. Hill's equation describes a function that is effectively a switch function given that the hill coefficient is high enough. However, seeking a bit in the literature, it's seems that the Hill coefficient never really overpass 3 (or 5 for extreme estimates). A Hill coefficient of within this range gives a logistic function but still looks quite suboptimal compared to what a perfect switch function could do.

Are there switch functions in molecular genetics that could translate a concentration into a TRUE/FALSE signal? How well do they simulate the perfect switch function? Are they based on cooperative binding or on some other mechanism?

---

References for estimate of Hill coefficients:

Answer

Positive feedbacks can be one alternative. Positive feedbacks exhibit bistability and can therefore adopt one of the two stable states depending on the initial condition. A famous example of a positive feedback switch would be that between cI and Cro in λ-phage, which repress each other. Positive feedbacks also display hysteresis: if the state of the system is dependent on an inducing ligand X such that system attains state S2 at high levels of the ligand, then to switch from S2 to S1 the concentration of X would have to be reduced below the level that was required for the S1 to S2 switch; and vice-versa. In other words, the system tries to remain in its current state.

If the transcription factors that constitute the genetic positive feedback have a co-operative behaviour then you can expect a steep switching curve.

For details see this review.

Is it alright to reapply to a PhD program you got rejected from if you work on your application?

The main weakness in my application was my weak research experience in the field I applied to. Say I got two new writing samples and greatly improved my research experience. Would it be acceptable for me to reapply to a school I got rejected from? I really want to go to that program.

conference - Default poster size

What is the typical size for a conference poster? Is it "ask the organizers", or can I safely assume that I'll get a default size, and that there will be enough space to hang it at the poster session?

Answer

You'll have to look at the conference announcement or contact the organizers. The space allotted to each poster is usually clearly stated in the conference announcement, and there are a number of different poster sizes being used, common ones being e.g. A0, A1. However, sometimes you are free to use the poster boards as you see fit (i.e. only provided a maximium limit).

If you are interested, here are also some resources/guidelines for constructing posters that I've used earlier:

mammals - Do dolphins really sleep?

I have always wondered, if dolphins sleep with one eye opened, do they really sleep at all?
According to this article dolphins indeed shut down their brains.

I already googled it. No help from wikipedia or other articles and publications found.

publications - What is the standard/recommended font to use in papers?

I looked around but did not find that anyone has asked this before, but what are the fonts that are standard/recommended while writing academic reports/papers?

Answer

If there's no template, then the choice is yours. However, you should make sure to pick a font that's easy to read. The usual standards in academia tend to be the Times, Helvetica/Arial, and Computer Modern families. This doesn't restrict you from using fonts like Book Antiqua, Myriad Pro, Goudy Old Style, or Garamond, but they're definitely not standard.

career path - How much of advantage do US citizens have in American academia?

How much of a disadvantage are you if you are not a US citizen, but want to enter American academia? It seems Americans have lots of advantages starting early, including NSF graduate fellowships, continuing with more access to DARPA money, US-citizen specific awards like PECASE, etc.