<?xml version="1.0" encoding="utf-8" standalone="yes"?><rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom"><channel><title>Elbo on mlbot.blog</title><link>https://mlbot.blog/tags/elbo/</link><description>Recent content in Elbo on mlbot.blog</description><generator>Hugo</generator><language>en-US</language><lastBuildDate>Thu, 30 Apr 2026 13:45:04 +0530</lastBuildDate><atom:link href="https://mlbot.blog/tags/elbo/index.xml" rel="self" type="application/rss+xml"/><item><title>Repairing a Nonlinear Strict Filter Without Reference Targets</title><link>https://mlbot.blog/posts/nonlinear-strict-filter-objective-repair/</link><pubDate>Thu, 30 Apr 2026 13:16:00 +0530</pubDate><guid>https://mlbot.blog/posts/nonlinear-strict-filter-objective-repair/</guid><description>&lt;p&gt;After the scalar benchmark, the work moved to a nonlinear sine-observation model:&lt;/p&gt;
\[
z_t = z_{t-1} + w_t,\quad w_t \sim \mathcal{N}(0,Q)
\]\[
y_t = x_t \sin(z_t) + v_t,\quad v_t \sim \mathcal{N}(0,R)
\]&lt;p&gt;The strict filtering contract stayed the same:&lt;/p&gt;
\[
q^F_t = \operatorname{update}(q^F_{t-1}, x_t, y_t)
\]&lt;p&gt;No hidden sequence state was allowed in the headline rows. The filter had to export an explicit online filtering marginal at each time step.&lt;/p&gt;</description></item></channel></rss>