Skip to content

CNS 2012 workshops

2012/07/30

This was my first time at CNS (computational neuroscience conference, not to be confused with the cognitive neuroscience one with the same acronym). I was invited to give a talk at the “Examining the dynamic nature of neural representations with the olfactory system workshop” organized by Chris Buckley, Thomas Nowotny, and Taro Toyoizumi. I presented my bursting olfactory receptor neurons can form instantaneous memory about the temporal structure of odor plume encounter story and a bit of related Calcium imaging study. Below is my summary of the workshop talks I went to (system identification workshop, information theory workshop on the first day, and olfactory workshop on the second day).

Garrett Stanley talked about system identification of the rat barrel cortex response from whisker deflection. He started by criticizing the white-noise Volterra series approach; it requires too much data. Instead, by designing a sequence of parametric stimuli that will directly show 2nd order and 3rd order interactions, he could fit a parametric form of firing rate response with good predictive powers [1]. As far as I can tell, it seemed like a rank-1 approximation of the 3rd order Volterra kernel. However, this model was lacking the fine-temporal latency, as well as stimulus intensity dependent bimodal responses, which was later fixed by a better model with feedback [2].

Vladimir Brezina talked about modeling of feedback from muscle contractions onto a rhythmic central pattern generator in the crab heart. He used LNL and LN models to fit the response of 9 neurons and muscles in the crab heart. For the LNL system, he used a bilinear optimization of the squared error. However, for the spiking response of the LN model, instead of using the Bernoulli or Poisson likelihood (the GLM model), he used least squares to fit the parameters.

Matthieu Louis gave a talk about optogenetically controlling drosophila larva’s olfactory sensory neurons. They built an impressive closed loop system that can control the larva’s behavior as if it were in an odor gradient. They modeled the system as a black box with odor input and behavior as output, skipping the model of the nervous system, and successfully predicted the behavior and control it [3].

Daniel Coca talked about how fly photoreceptors can act as a nonlinear temporal filter that is optimized for detecting edges. He fit a NARMAX (nonlinear ARMA-X) model and analyzed it in the frequency domain and found that the phase response is consistent with phase congruency detection model for edge detection. Also, he explained how the system “linearizes” when stimulated with white Gaussian noise, although I couldn’t follow the details due to my lack of knowledge in nonlinear frequency domain analysis.

Tatyana Sharpee talked about sphere packing in the context of receptive fields of retina, and conditional population firing rates of song birds. For the receptive fields, she showed that to maximize the mutual information per unit lattice between a point source of light and the (binary) neural response of ganglion cells, if the lattice is not-perfect, elliptical shapes of receptive fields can help. For the song bird case, she showed that the noise correlation can change with training to improve separation (classification performance) of the conditional distributions while the irrelevant stimuli became less separable.

Rava Azeredo da Silveira talked about how finely tuned correlation structure can immensely increase performance. Given two population of neurons, each tuned to a class weakly (slightly higher firing rate for the preferred class), if cross-population correlation is slightly higher than otherwise, the population response as a whole can be very certain about the class identity. He also talked about many other related things such as asymptotics on required population size vs noise.

Shy Shoham talked about Linear-Nonlinear-Poisson (LNP) and Linear-Nonlinear-Hawkes (LNH) models, and how to relate spike train (output) correlations to gaussian (input) correlation [4,5]. LNH has a similar form to GLM but the feedback is added outside the nonlinearity. He referred to the procedure of inferring the underlying latent AR process as correlation-distortion, and proposed to use it for studying neural point processes as AR models; hence apply Granger causality, and other signal processing tools. He also talked about semi-blind system identification where the goal is to infer the linear kernel of the model given the autocorrelation of the input and the autocorrelation of the population spike trains are given (the phase ambiguity of the filter is resolved by choosing the minimal phase filter.)

Maxim Bazhenov talked about modeling the transient synchronization in the locust olfactory system as a network phenomena (interaction between projection neurons (PNs) and local inter-neurons (LNs)). The pattern of synchronization of PNs over multiple LFP cycles is repeatable, and his model reproduces it. He showed an interesting illustration of the connectivity between LNs posed as the graph coloring problem [6]. Each cluster of LNs targets everybody outside their cluster, enabling synchrony within. The connectivity matrix is effectively a block diagonal of zeros, and the off-diagonals are ones, because they are inhibitory neurons.

Nitin Gupta gave a talk on lateral horn (LH) cells. The normative model has been that the inhibitory neurons in LH acts as feed-forward inhibition to limit the integration time within the Kenyon cells (KCs). He identified a heterogeneous population of neurons in LH (see [7] for beautifully filled neurons). Among the ones that project to mushroom body (where KCs are), he found no evidence of GABA co-location, suggesting that there is no feed-forward inhibition through LH. He proposed an alternative model for limiting integration time in KCs, namely the feedback inhibition through (non-spiking) GGNs.

Thomas Nowotny talked about how odor plume structure can help in separating mixture of different sources, based on the the results of [8]. He proposed a simple model of lateral inhibition circuit among the glomeruli. The model showed counter-intuitive results for temporal mixtures of odor when linear decoding is used.

Kevin C. Daly gave a data packed talk on Manduca sexta (moth) olfactory system [9]. The oscillation he observed had a frequency modulation; starts at a high frequency and quickly falls, and it is odor dependent. He criticized the use of continuous odor application which may result in pathological responses (my wording), and instead he showed response to odor-puffs. (Interestingly, the blank puffs decreased the response.) He also emphasized the importance of not cutting the head of the animal, which preserves a pair of histamine neurons.

Aurel A. Lazar talked about precise odor delivery system using laminar flows that can produce a diverse temporal pattern of odor concentration with around 1% of error. Using this system, they showed that the firing response of the first two stages of drosophila; receptor neurons and projection neurons are both temporally differentiating. This was not simultaneously recorded, but thanks to the repeatable stimuli and response, it is well supported.

References:

  1. R. M. Webber and G. B. Stanley. Transient and steady-state dynamics of cortical adaptation, J. Neurophys., 95:2923-2932, 2006.
  2. A. S. Boloori, R. A. Jenks, Gaelle Desbordes, and G. B. Stanley. Encoding and decoding cortical representations of tactile features in the vibrissa system, J. Neurosci., 30(30):9990-10005, 2010.
  3. Gomez-Marin A, Stephens GJ, Louis M. Active sampling and decision making in Drosophila chemotaxis. Nature Communications 2:441. doi: 10.1038/ncomms1455 (2011).
  4. Michael Krumin, Shy Shoham. Generation of Spike Trains with Controlled Auto- and Cross-Correlation Functions. Neural Computation. June 2009, Vol. 21, No. 6, Pages 1642-1664
  5. Michael Krumin, Inna Reutsky,  Shy Shoham. Correlation-Based Analysis and Generation of Multiple Spike Trains Using Hawkes Models with an Exogenous Input.  Front Comput Neurosci. 2010; 4: 147
  6. Assisi C, Stopfer M, Bazhenov M. Using the structure of inhibitory networks to unravel mechanisms of spatiotemporal patterning. Neuron. 2011 Jan 27;69(2):373-86.
  7. Nitin Gupta, Mark Stopfer. Functional Analysis of a Higher Olfactory Center, the Lateral Horn. Journal of Neuroscience, 13 June 2012, 32(24): 8138-8148; doi: 10.1523/​JNEUROSCI.1066-12.2012
  8. Paul Szyszka, Jacob S. Stierle, Stephanie Biergans, C. Giovanni Galizia. The Speed of Smell: Odor-Object Segregation within Milliseconds. PLoS ONE, Vol. 7, No. 4. (27 April 2012), e36096, doi:10.1371/journal.pone.0036096
  9. Daly KC, Galán RF, Peters OJ and Staudacher EM (2011) Detailed characterization of local field potential oscillations and their relationship to spike timing in the antennal lobe of the moth Manduca sexta. Front. Neuroeng. 4:12. doi: 10.3389/fneng.2011.00012
Advertisements
No comments yet

Leave a Reply

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

WordPress.com Logo

You are commenting using your WordPress.com 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 )

Google+ photo

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

Connecting to %s

%d bloggers like this: