diff --git a/slides/information-theory/slides-info-kl-ml.tex b/slides/information-theory/slides-info-kl-ml.tex
index 0e8e7151..7421d76b 100644
--- a/slides/information-theory/slides-info-kl-ml.tex
+++ b/slides/information-theory/slides-info-kl-ml.tex
@@ -31,7 +31,7 @@
     \item \textbf{Probabilistic model fitting}\\
 Assume our learner is probabilistic, i.e., we model $p(y| \mathbf{x})$ for example (for example, ridge regression, logistic regression, ...).
 
-\includegraphics[width=0.4\linewidth]{slides/information-theory/figure_man/kl_ml_prob_fit.png}
+\includegraphics[width=0.4\linewidth]{figure_man/kl_ml_prob_fit.png}
 
 We want to minimize the difference between $p(y \vert \mathbf{x})$ and the conditional data generating process $\mathbb{P}_{y\vert\mathbf{x}}$ based on the data stemming from $\mathbb{P}_{y, \mathbf{x}}.$
 
@@ -122,7 +122,7 @@
  \end{itemize}
  \framebreak
 
-\includegraphics[width=0.6\linewidth]{slides/information-theory/figure_man/kl_ml_fkl_rkl.png} \\
+\includegraphics[width=0.6\linewidth]{figure_man/kl_ml_fkl_rkl.png} \\
 The asymmetry of the KL has the following implications
 \begin{itemize}
     \item The forward KL $D_{KL}(p\|q_{\bm{\phi}}) = \E_{\xv \sim p} \log\left(\frac{p(\xv)}{q_{\bm{\phi}}(\xv)}\right)$ is mass-covering since $p(\xv)\log\left(\frac{p(\xv)}{q_{\bm{\phi}}(\xv)}\right) \approx 0$ if $p(\xv) \approx 0$ (as long as both distribution do not extremely differ)
diff --git a/slides/regularization/slides-regu-softthresholding-lasso-deepdive.tex b/slides/regularization/slides-regu-softthresholding-lasso-deepdive.tex
index 5be7817d..b8bed098 100644
--- a/slides/regularization/slides-regu-softthresholding-lasso-deepdive.tex
+++ b/slides/regularization/slides-regu-softthresholding-lasso-deepdive.tex
@@ -3,7 +3,7 @@
 \input{../../latex-math/basic-math}
 \input{../../latex-math/basic-ml}
 
-\newcommand{\titlefigure}{slides/regularization/figure/th_l1_pos.pdf}
+\newcommand{\titlefigure}{figure/th_l1_pos.pdf}
 \newcommand{\learninggoals}{
   \item Understand the relationship between soft-thresholding and L1 regularization
 }
@@ -54,7 +54,7 @@
 1) $\hat{\theta}_{\text{Lasso},j} > 0:$ \\
 \lz
 \begin{minipage}{0.4\textwidth}
-    \includegraphics[width=5cm]{slides/regularization/figure/th_l1_pos.pdf}
+    \includegraphics[width=5cm]{figure/th_l1_pos.pdf}
 \end{minipage}
 \hfill
 \begin{minipage}{0.49\textwidth}
@@ -70,7 +70,7 @@
 2) $\hat{\theta}_{\text{Lasso},j} < 0:$ \\
 \lz
 \begin{minipage}{0.4\textwidth}
-    \includegraphics[width=5cm]{slides/regularization/figure/th_l1_neg.pdf}
+    \includegraphics[width=5cm]{figure/th_l1_neg.pdf}
 \end{minipage}
 \hfill
 \begin{minipage}{0.49\textwidth}
@@ -85,7 +85,7 @@
 
 
 \begin{minipage}{0.4\textwidth}
-    \includegraphics[width=5cm]{slides/regularization/figure/th_l1_zero.pdf}
+    \includegraphics[width=5cm]{figure/th_l1_zero.pdf}
 \end{minipage}
 \hfill
 \begin{minipage}{0.49\textwidth}