Hassan-Roland Nasser

• Professor
• Researcher at Agroscope

We are developing a project based on LIPUS and we need to buy a Pizo Crystal with the following properties:

Anyone can help telling us about a website from which we can buy such crystals?

Thanks in advance.

In the literature, it is not clear what is the difference between those terms and there seem to be a very big misuse of them.

Online resources and organizational reports seems to use them interchangeably. In the same time, research papers suggests several definitions for each of them, which are sometimes very "un-related'.

Could anyone help to clarify the difference between them?

Say we have a set of alpha_i (1<i<8) and and real number Z that belongs to R, where Z is constant and predefined (say Z = 4).

Finding all possible combination of alpha_i would be an NP problem. However, the task could be possible if we discretize. If we discretize at 0.01, then we would have 400^8 possible combination. It is possible to scan all of them and then test, but computationally, this would be very costly.

Is there an efficient algorithm to do this task?

Can we generalize this relationship on other multidimensional distributions?

After my Phd, I started a private company with 2 colleagues in software development. I am looking to participate in a research project that can lead to developing an innovative software that can add values to the market place. We would like also that the software create job vacancies together with the market value.

We can participate at many levels:

- Studying the economical impact of the idea/software.

- Developing the code, graphical user interface.

- Design and software ergonomics.

How could such scenario happen?

I have a set of underground parkings and I am searching for a way to make a navigator tool that allows to create the 3D shape of the parkings and then create a tool with the 2D (?) or 3D (?) data in order to navigate virtually.

A challenging issue in today's visual neuroscience community is to understand the activity of a network of Retinal Ganglion Cells (RGC) jointly with a stimulus. In this spot, we have several mathematical and computational tools which are based on several frameworks (empirical statistics, encoding and decoding, probabilistic modeling).

In the spirit of our ultimate goal (understanding the response/stimulus), we are faced with two problematic questions which are really important (and very correlated) to set a clearer vision and to develop the suitable tools:

1- What framework (and mathematical tools) should we use? Do we build a framework based on all relevant literature works? Are there any guidelines to decide what are the suitable tools?

2- How can we bring a common tool to both biologists (neurobiologists, experimentalists) and programmers (computational neuroscientists, computer scientists)? Clearly, the best tool for a biologist is a graphical user interface (but they will be limited to available tools) and for programmers is an IDE or programming language (but their works will be hardly accessible by biologists). So, is there a way to build a tool that they both use and be flexible with a common vision and two different working textures?

I am looking for people who work in labs for neuroscience (mostly computational neuroscience and probabilistic modeling) and in the same time have relationship at the industrial level, i.e., they contributed in a spin-off or any of the technology transfer actions.

I have a Maxent probability distribution that I want to fit to empirical data. I implemented the algorithm described in this paper (http://citeseer.ist.psu.edu/viewdoc/summary?doi=10.1.1.89.6189). The author claims and proves mathematically the convergence.

I have a very strange behavior DURING THE PARALLEL UPDATE (when all the parameters are updated at one): After a very nice convergence, where the error decreases during few (5-10) iterations (some times 1-2), the error increases dramatically.

I had several the following explanation in mind:

Convergence oscillations. When the problem arrives at the bottom, it begins to oscillate around the convergence point since the problem is relaxed and we permit some error on the estimated features.

However, The amplitude of the oscillations are very big!!! And hence, I think that it could not come from the oscillations.

Has anyone experienced a similar problem? Any interpretation?

Thanks you very much.

Please consider also to indicate the application. Is it for retinal spike trains or cortex acquisitons ... ?

I am trying to fit spatio-temporal maximum entropy parameters (lagrange multipliers) to real spike train data. I would like to know if anyone can recommend a convergence criterion other than the one I'm currently using.

The averages of features with respect to empirical distribution are: <emp_i> and those of the target distribution are <tar_i>. Each time I update the target distribution, I compute the error with the following assumption:

error^2 = max ( (<empi_i> - <tar_i>)/ SigmaMax).

Where SigmaMax is the maximum mean squared deviation measured on the empirical distribution and it is equal to: (<emp_i>(1-<emp_i>))/DistributionLength.

The problem with the criteria is that it's not really reliable because of the sparsity of some features which sometimes make division near zero. The absolute difference of error is also not reliable.

Any suggestions?