fix: add research questions and make minor fixes

Author: Boolean-true

presentation_final/sections/05_novelty.tex

@@ -12,7 +12,7 @@
 
     \vspace{0.2cm}
     \item \textbf{Comprehensive empirical study} covering 2~SOTA models
-      (DeepSeek~V3, Kimi~K2) $\times$ 5~grid regions $\times$ 4~years of replay data
+      (DeepSeek~V3, Kimi~K2) $\times$ 5~grid regions of replay data
       --- the first systematic Pareto frontiers for carbon-aware LLM training.
 
     \vspace{0.2cm}

presentation_final/sections/06_objectives.tex

@@ -1,32 +1,14 @@
-\begin{frame}[t]{Project Objectives}{What will the simulation answer?}
-  \textbf{Obligatory:}
+\begin{frame}[t]{Research Questions}{What do we want to answer?}
   \begin{itemize}
     \item \textbf{Carbon Intensity \& Temporal Trade-offs}\\
-      Quantify how CO$_2$ savings depend on the pause threshold
-      (gCO$_2$eq/kWh) versus added training time
-      $\Rightarrow$ Pareto frontier for carbon-aware scheduling.
-    \item \textbf{Impact Assessment on SOTA Architectures}\\
-      Benchmark absolute and relative CO$_2$ reduction potential for
-      different large models (e.g.\ DeepSeek V3, Kimi K2).
-  \end{itemize}
-
-  \vspace{0.3cm}
-  \textbf{Wishful:}
-  \begin{itemize}
+      How much CO$_2$eq can be saved by dynamically pausing the
+      pretraining when CO2 intensity is high and how much training
+      time is added?
     \item \textbf{Spatiotemporal Selection}\\
-      Identify favourable \emph{time windows} and \emph{seasons}
+      What are the favourable \emph{time windows} and \emph{seasons}
       to run training in a given region.
     \item \textbf{Geospatial Grid Analysis}\\
-      Compare regions to find the most sustainable locations
-      for pretraining, given their grid mix.
-  \end{itemize}
-
-  \vspace{0.3cm}
-  \textbf{Optional:}
-  \begin{itemize}
-    \item \textbf{Computational Granularity \& Resumption Logic}\\
-      Study how checkpoint frequency (batch vs.\ epoch, etc.)
-      impacts the trade-off between CO$_2$ savings and overhead.
+      What region provides the best carbon savings for the pretraining of a given model?  
   \end{itemize}
 
 \end{frame}

presentation_final/sections/08_parameters.tex

@@ -1,6 +1,6 @@
 \subsection{Specification}
 
-\begin{frame}[t]{Parameters I}{}
+\begin{frame}[t]{Parameters}{}
 
   \small
 

presentation_final/sections/18_results.tex

@@ -106,7 +106,7 @@ \subsection{Threshold Space Analysis}
 
   \vspace{0.2cm}
   \begin{alertblock}{Central finding}
-    $\theta_p - \theta_r < 16$~gCO$_2$eq/kWh \textbf{in every region and model}.
+    $\theta_p - \theta_r \leq 16$~gCO$_2$eq/kWh \textbf{in every region and model}.
     Wide hysteresis degrades the savings-to-overhead ratio.
   \end{alertblock}
 

presentation_final/sections/19_conclusion.tex

@@ -14,7 +14,7 @@
   \vspace{0.2cm}
   \begin{alertblock}{Central finding}
     The optimal hysteresis margin is consistently narrow
-    ($\theta_p - \theta_r < 16$~gCO$_2$eq/kWh).
+    ($\theta_p - \theta_r \leq 16$~gCO$_2$eq/kWh).
     Wide hysteresis \textbf{never} Pareto-optimal.
     Pausing is a binary clean/dirty decision, not a multi-level filter.
   \end{alertblock}