Add chatgpt and sora examples

reveal-md.json

@@ -1,3 +1,4 @@
 {
-    "theme": "white"
+    "theme": "white",
+    "css": "slides/assets/css/slide-style.css"
 }

slides/01_introduction_to_deep_learning.md

@@ -36,16 +36,6 @@ The Chinese board game is considered to be a much more complex challenge for a c
 
 ---
 
-### Impersonating celebrities
-
-<iframe width="560" height="315" src="https://www.youtube.com/embed/cQ54GDm1eL0" frameborder="0" allow="accelerometer; autoplay; encrypted-media; gyroscope; picture-in-picture" allowfullscreen></iframe>
-
-Note:
-Deepfakes are synthetic media[2] in which a person in an existing image or video is replaced with someone else's likeness.
-
-Obviously dangerous but there is also ongoing research into using deep learning to detect deep fakes.
----
-
 ### Mars mission
 
 <span class="medium-note">Terrain classification, navigation, autonomous rover, etc.</span>
@@ -60,6 +50,31 @@ Terramechanics plays a critical role in the areas of ground vehicles and ground
 This research applies state-of-the-art algorithms in deep learning to two key problems: estimating wheel slip and classifying the terrain being traversed by a ground robot
 ---
 
+### Text Generation
+
+<div>
+<img src="assets/img/chatgpt-investigator-award.png" style="width:50%"/>
+</div>
+<span class="small-note">ChatGPT 3.5 - OpenAI</span>
+
+---
+
+### Image and Video Generation
+
+<iframe width="560" height="315" src="https://www.youtube.com/embed/HK6y8DAPN_0?si=BR3NSJVezn_GwRR7" frameborder="0" allow="accelerometer; autoplay; encrypted-media; gyroscope; picture-in-picture" allowfullscreen></iframe>
+<div><span class="small-note">Sora - OpenAI</span></div>
+---
+
+### Impersonating celebrities
+
+<iframe width="560" height="315" src="https://www.youtube.com/embed/cQ54GDm1eL0" frameborder="0" allow="accelerometer; autoplay; encrypted-media; gyroscope; picture-in-picture" allowfullscreen></iframe>
+
+Note:
+Deepfakes are synthetic media[2] in which a person in an existing image or video is replaced with someone else's likeness.
+
+Obviously dangerous but there is also ongoing research into using deep learning to detect deep fakes.
+---
+
 ### What else?
 
 * Cancer detection

slides/03_convolution_neural_networks.md

@@ -163,16 +163,6 @@ Especially when there are big differences between the magnitude of your features
 
 ---
 
-### Image Data Differences
-* Tabular data specifies features but in images, these are the pixels.
-* Image features are spatially related, i.e. ears are located on a head.
-* The 2-D (greyscale) or 3-D (colour) input can be flattened to work with `Dense` NN's, similar to tabular data.
-
-<object type="image/svg+xml" data="assets/img/cat-flatten.svg" style="background: white; width: 60%; height: auto;">
-</object>
-
----
-
 ### Hierarchical feature representaion 
 Deep neural networks learn larger and more abstract features deeper into the network.
 
@@ -205,15 +195,6 @@ Each node in layer is connected to **every node** in the **previous** layer.
 
 ---
 
-### Images in Dense Networks
-* `Dense` NN's are unable to effectively utilise spatial information.
-* This is because a node in the next layer is connected to all previous nodes, not just local nodes like in convolutional layers.
-
-<object type="image/svg+xml" data="assets/img/cat-px-relations.svg" style="background: white; width: 60%; height: auto;">
-</object>
-
----
-
 ### Convolution Layer
 
 <object type="image/svg+xml" data="assets/img/convolution-1d.svg" style="background: white; width: auto; height: auto;">
@@ -234,6 +215,16 @@ Each node in layer is connected to **every node** in the **previous** layer.
 * Each node only connects to a subset of nodes in the previous layer
 ---
 
+### Dense vs Convolution for Images
+* Image features are spatially related, i.e. ears are located on a head.
+* `Dense` layers are unable to effectively utilise spatial information.
+* This is because a node in the next layer is connected to all previous nodes, not just local nodes like in convolutional layers.
+
+<object type="image/svg+xml" data="assets/img/cat-px-relations.svg" style="background: white; width: 60%; height: auto;">
+</object>
+
+---
+
 ### Convolution in 2D
 
 We arrange our input nodes as a grid to match our input image.