1
$\begingroup$

In a machine learning paper I'm reading, I came across the following notation:

In our experiments we parameterize these potential functions as an exponential of a linear scoring function

$\Phi(a_{i-1},a_i,\mathbf{x},\mathbf{y}) = \operatorname{exp}\Lambda\cdot\mathbf{f}(a_{i-1},a_i,\mathbf{x},\mathbf{y})$

where $\mathbf{f}$ is a vector of feature functions...

What does the "$\operatorname{exp}\Lambda\cdot$" notation mean?

$\endgroup$
1
$\begingroup$

Dot product of vectors. If $\Lambda = (\lambda_1,\lambda_2,\dots,\lambda_n)$ and the components of $\mathbf f$ are $f_1,f_2,\dots, f_n$, then the dot product is $$ \Lambda\cdot\mathbf{f}(a_{i-1},a_i,\mathbf{x},\mathbf{y}) = \lambda_1 f_1(a_{i-1},a_i,\mathbf{x},\mathbf{y})+ \lambda_1 f_1(a_{i-1},a_i,\mathbf{x},\mathbf{y})+\dots+ \lambda_n f_n(a_{i-1},a_i,\mathbf{x},\mathbf{y}) $$ which is what they called a "linear scoring function". So $\mathbf f$ is a "vector of feature functions", and $f_1, f_2, \dots, f_n$ are the "feature functions". I used $n$ for the number of features.

Then you take the exponential of that.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.