Gaussian distribution using Bayes' rule

147 Views Asked by At

I have this exercise:


An explosion was detected by two sensors. Each sensor is only able to output a noisy estimate of the location of the explosion due to measurement noise. Assuming the two sensor outputs are y1 and y2, and the likelihood of the exact location x given the sensor outputs is $$ p(y_1|x)p(y_2|x) = \mathcal{N}(y_1;x,σ_1^2)\mathcal{N}(y_2;x,σ_2^2) $$ where $σ_1^2$ and $σ_2^2$ are the measurement noise variances. Assuming a prior distribution over the location: $$ p(x) = \mathcal{N}(x; 0, σ_0^2), $$ where $σ_0^2$ is the prior variance.

1: Find the posterior distribution $p(x|y_1, y_2)$


When deriving the expression for the posterior, do you first find a new Gaussian Pdf for the likelihood $(p(y_1|x)p(y_2|x) = \mathcal{N}((y1,y1); μ3, σ3)$ and then multiply this with the prior to find yet another Gaussian? And if yes how does one go about doing that?

1

There are 1 best solutions below

0
On

One important thing to note when working with posteriors and priors is that we assume they belong to the same class of probabilities. In other words, given a prior we already know the general form of the posterior.

Expanding on your problem here we have that (denoting the prior by $\pi(x)$) that; $$ \pi(x|y_1,y_2) \propto f(y|x)\pi(x)\propto \prod_{i=1}^{2}f(y_i|x)\pi(x) $$ which simplifies to $$\pi(x|y_1, y_2)\propto \frac{1}{\sqrt{2\pi}\sigma_1}e^{-\frac{(y_1 - x)^2}{2\sigma_1^2}}\frac{1}{\sqrt{2\pi}\sigma_2}e^{-\frac{(y_2 - x)^2}{2\sigma_2^2}}\frac{1}{\sqrt{2\pi}\sigma_0}e^{-\frac{x^2}{2\sigma_0^2}}$$ This is possible because densities which are proportional have the same distribution.

You will find that this is again a normal distribution with $N(x;\mu',\sigma'^2)$ where $\mu', \sigma'^2$ are identified by examining how $\pi(x|y_1,y_2)$ depends on x up to some constant of proportionality not depending on x. (really recommend doing this step yourself as being comfortable with manipulating expressions in probability is soo useful)