revert

Author: antoine-galataud

_posts/2024-07-09-hvac-challenges.md

@@ -151,7 +151,7 @@ are also important parts of the engineering effort).
 The table below gives two arbitrary examples of unscalable and scalable solutions. Numbers give an order of 
 magnitude of the engineering effort required to handle a certain number of environments, and what we should aim for.
 
-![sources]({{ site.baseurl }}/assets/hvac_ai_challenges/scalability_resources.png){: .center }
+![sources]({{ site.baseurl }}/assets/hvac_ai_challenges/scalability_resources.svg){: .center }
 
 A first angle to approach the problem is by defining how scalability of the solution is impacted by the complexity of a project. 
 Intuitively, the larger the control perimeter, the more complex it gets, and the less likely it is that we'd be able to 
@@ -219,7 +219,7 @@ However, in industrial control, and typically in HVAC control, the problem is mo
 the cake (chocolate) and surroundings are unknown, other parts (very few) are known.
 
 ![sources]({{ site.baseurl }}/assets/hvac_ai_challenges/dall_e_marble_cake_anot-mh.png){: .center-m }
-<center class="image-foot"><i>Source: DALL-E</i></center>
+<center class="image-foot"><i>Generated by DALL-E</i></center>
 
 Why? Because the system dynamics (state space, transition function) is not a single, continuous distribution, but a mixture of different distributions, each corresponding to a 
 different operating mode of the system, under specific conditions (occupancy, weather). Moreover, the collected data is likely to reflect only steady-state conditions,