Update documentation

tumaer · Jan 9, 2024 · 64a7bf9 · 64a7bf9
1 parent 8a82a81
commit 64a7bf9
Show file tree

Hide file tree

Showing 24 changed files with 482 additions and 297 deletions.
diff --git a/_images/lstm_block.png b/_images/lstm_block.png
diff --git a/_images/rnn_bidir.png b/_images/rnn_bidir.png
diff --git a/_images/rnn_causalcnn.png b/_images/rnn_causalcnn.png
diff --git a/_images/rnn_conv_rnn.png b/_images/rnn_conv_rnn.png
diff --git a/_images/rnn_deep.png b/_images/rnn_deep.png
diff --git a/_images/rnn_seq2seq.png b/_images/rnn_seq2seq.png
diff --git a/_images/rnn_seq2seq_translate.png b/_images/rnn_seq2seq_translate.png
diff --git a/_images/rnn_seq2vec.png b/_images/rnn_seq2vec.png
diff --git a/_images/rnn_sequences.jpeg b/_images/rnn_sequences.jpeg
diff --git a/_images/rnn_textcnn.png b/_images/rnn_textcnn.png
diff --git a/_images/rnn_vec2seq.png b/_images/rnn_vec2seq.png
diff --git a/_sources/about.md b/_sources/about.md
@@ -11,6 +11,26 @@ The course breaks down into an introduction to the topic, followed by 4 core con
 
 > Questions should preferably be posted in the Moodle, or else be sent to Artur Toshev.
 
+
+## Outline
+
+```{tableofcontents}
+```
+
+
+## Contributors
+
+Thanks to all contributors! Github names, if available.
+
+### Contributed content
+
+1. Armin Illerhaus - Notebooks on Windows
+
+### Content fixes
+
+1. Andreas Steger (AndSte01)
+
+
 ## Citation
 
 Please cite this work as:
@@ -24,11 +44,6 @@ Please cite this work as:
 }
 ```
 
-## Outline
-
-```{tableofcontents}
-```
-
 <br>
 <br>
 

diff --git a/_sources/faq.md b/_sources/faq.md
@@ -9,6 +9,7 @@ There exist a number of resources to run the tutorials online, and also work on
 2. [**AWS SageMaker Studio Lab**](https://aws.amazon.com/sagemaker/studio-lab/) new-ish competitor to Google Colab. Pretty much the same as Google Colab, just running on Amazon Web-Services.
 3. [**PaperSpace Gradient**](https://www.paperspace.com/gradient) Hosted JuPyter runtime on the cloud with pre-installed machine learning packages.
 
+Look at [](software.md) for more details on the software setup.
 
 ## 2. I have a question which is not suitable for the Moodle, who should I contact?
 

diff --git a/_sources/lecture/gmm.md b/_sources/lecture/gmm.md
@@ -251,7 +251,7 @@ In the E-step, we compute the posterior probability of $z^{(i)}_k$ given the dat
 $$
 \begin{aligned}
 p(z^{(i)}=k| x^{(i)},\pi,\mu,\Sigma) &= \frac{p(x^{(i)}|z^{(i)}=k, \mu, \Sigma)p(z^{(i)}=k,\pi)}{\sum_{l=1}^K p(x^{(i)}|z^{(i)}=l, \mu, \Sigma)p(z^{(i)}=l,\pi)} \\
- &= \frac{\pi_k \mathcal{N}(x^{(i)}|\mu_K, \Sigma_k)}{\sum_{l=1}^K \pi_l \mathcal{N}(x^{(i)}|\mu_l, \Sigma_l)}
+ &= \frac{\pi_k \mathcal{N}(x^{(i)}|\mu_k, \Sigma_k)}{\sum_{l=1}^K \pi_l \mathcal{N}(x^{(i)}|\mu_l, \Sigma_l)}
 \end{aligned}$$ (gmm_responsibilities)
 
 The values of $p(x^{(i)}|z^{(i)}=k, \mu, \Sigma)$ can be computed by evaluating the $k$th Gaussian with parameters $\mu_k$ and $\Sigma_k$. And $p(z^{(i)}=k,\pi)$ is just $\pi_k$. 
@@ -273,7 +273,7 @@ Most limitations of this approach arive from the assumption that the indivudual
 - If the data does not follow a Gaussian distribution, e.g. heavy-tailed ditribution with outliers, then too much weight will be given to the outliers
 - If there is an outlier, eventually one mode will focus only on this one data point. But if a Gaussian describes only one data point, then its variance will be zero and we recover a singularity/Dirac function.
 - The choice of $K$ is crucial and this parameters need to be optimized in a outer loop.
-- GMMs do now scale well to high dimensions.
+- GMMs do not scale well to high dimensions.
 
 
 ## Further References