Fix outdated info and typos

slides/00_overview.md

@@ -2,7 +2,9 @@
 title: "0. Overview"
 ---
 
-# RSES: Introduction to Deep Learning Course
+# Deep Learning Demystified
+## Foundations for Non-Computer Scientist
+
 
 Twin Karmakharm
 
@@ -19,16 +21,15 @@ Twin Karmakharm
     * Grant support
 * Software optimisation, GPU and HPC
 * Training, outreach and education activities
-* Led by Dr. Paul Richmond
+* Led by Dr. Romain Thomas
 * Visit us at [https://rse.shef.ac.uk](https://rse.shef.ac.uk)
-* [Apply for up to 12 days of RSE time on AI/HPC related project](https://rse.shef.ac.uk/collaboration/tier2)
 ---
 
 ### Course Materials
 
-All course materials can be found at:
+Course materials can be found at:
 
-[https://rses-dl-course.github.io/](https://rses-dl-course.github.io/)
+[https://rse.shef.ac.uk/dl-demystified-course](https://rse.shef.ac.uk/dl-demystified-course)
 
 
 ---
@@ -141,7 +142,7 @@ A platform to run lab code interactively
 `Ctrl+Space` to get code suggestion
 
 ---
-
+<!-- .slide: data-visibility="hidden" -->
 
 ### Practical Labs Using Jupyter Notebooks
 A platform to run lab code interactively

slides/01_introduction_to_deep_learning.md

@@ -375,7 +375,7 @@ There's likely a framework for a language that you know e.g.:
 * Matlab - Deep Learning Toolbox
 
 Or import from python
-* 'keras' package for R - Used for R version of the course
+* 'keras' package for R
 
 ---
 

slides/02_neural_networks.md

@@ -189,7 +189,7 @@ Note:
 
 ### Error and Loss
 
-We need a way good way to quantify how far our prediction is from our goal.
+We need a good way to quantify how far our prediction is from our goal.
 
 <object type="image/svg+xml" data="assets/img/neuralnetwork-initialisation.svg" style="background: white; width: 80%; margin-top: 1em">
 </object>
@@ -415,7 +415,7 @@ Too small learning rate ($\eta$) means our network takes longer to converge.
 
 How do we know the 'optimal' learning rate?
 
-We won't know until actually training the network! <!-- .element: class="fragment" -->
+We won't know until we train the network! <!-- .element: class="fragment" -->
 
 Start with the default values provided. <!-- .element: class="fragment" -->
 
@@ -460,7 +460,7 @@ The algorithm for computing the gradients efficiently and adjusting the weights
 1. Calculate loss - compare prediction with target
 1. Back propagation - find gradients for each weight and bias
 1. Repeat for all samples
-1. Average gradients of weights and bias then update <!-- .element: class="fragment" -->
+1. Average the delta of weights and bias then update <!-- .element: class="fragment" -->
 1. Start again from 1 <!-- .element: class="fragment" -->
 
 ---