this is for holding javascript data
jBillou edited Waveform optimization.tex
over 9 years ago
Commit id: 06e6cca4b1d61d78b216351b999142b961019d11
deletions | additions
diff --git a/Waveform optimization.tex b/Waveform optimization.tex
index 20b9431..aefad55 100644
--- a/Waveform optimization.tex
+++ b/Waveform optimization.tex
...
We can now expand the integral over the hidden states:
$\idotsint \prod_j^N \prod_u^{T_j} $$\idotsint \prod_{j=1}^N \prod_{u=1}^{T_j} dx_u^j \sum_i [ \sum_t (d_t^i - m(x_t^i))^2
]P(X|D,\Theta_{k-1})$ ]P(X|D,\Theta_{k-1})$$
$ $$ = \sum_{i,t} \idotsint
\prod_j^N \prod_u^{T_j} \prod_j \prod_u dx_u^j (d_t^i - m(x_t^i))^2
P(X|D,\Theta_{k-1})$ P(X|D,\Theta_{k-1})$$
$ $$ = \sum_{i,t} \iint dx_t^i (d_t^i - m(x_t^i))^2 \idotsint \prod_{j\neq
i}^N i} \prod_{u\neq
t}^{T_j} t} dx_u^j
P(X|D,\Theta_{k-1})$ P(X|D,\Theta_{k-1})$$
$ $$ = \sum_{i,t} \iint dx_t^i (d_t^i - m(x_t^i))^2
P(x_t^i|D,\Theta)$ P(x_t^i|D,\Theta)$$
Where $P(x_t^i|D,\Theta)$ is the posterior distribution of our hidden states at time $t$ for a trace $i$, and is computed as a discrete table indexed by integer $l \in [1,..N_\lambda]$ and $m \in [1,..N_\phi]$: $P_t^i[l,m]$ . Since we work in discretized space we rewrite the above integral as: