various updates

stanton119 · Jul 18, 2024 · 11c8870 · 11c8870
1 parent 6a3cb3d
commit 11c8870
Show file tree

Hide file tree

Showing 5 changed files with 273 additions and 229 deletions.
diff --git a/README.md b/README.md
@@ -194,6 +194,8 @@ Use the `settings.json` file in the repo
 *   Beta Bernouli bandit vs logistic regression with no features
 *   NN multi-row vs multi-column - do they perform similarly?
 *   Multi horizon forecasting direct method - with shared NN architecture - compare separate models for each horizon with a NN that shares layers. Compare with sequence to sequence models.
+*   Gaussian process from scratch
+    *   ref - https://www.youtube.com/watch?v=HA-VHNVbvwQ&list=WL&index=26
 *   Probabilistic neural networks
     *   Normalizing flows - model complex distributions with transformations of gaussians
     *   Can we train an output layer as a gaussian mixture to model complex distirbutions via gradient descent

diff --git a/neural_networks/unfinished-classification_uncertainty/dirichlet_prior_networks.ipynb b/neural_networks/unfinished-classification_uncertainty/dirichlet_prior_networks.ipynb
diff --git a/nlp/unfinished-text_embeddings/text_embedding.ipynb b/nlp/unfinished-text_embeddings/text_embedding.ipynb
diff --git a/paper_list.md b/paper_list.md
@@ -243,4 +243,10 @@ Read 2024/03
   * Using a learning rate decay of $\alpha_t=\alpha/\sqrt{t}$ is common.
 * Results
   * Performs similarly to SGD with momentum on dense MNIST
-  * Performas similarly to AdaGrad on sparse problems (best of both worlds)
+  * Performas similarly to AdaGrad on sparse problems (best of both worlds)
+
+### The LambdaLoss Framework for Ranking Metric Optimization [2018]
+https://dl.acm.org/doi/10.1145/3269206.3271784
+Read 2024/03
+* Summary
+  * Formulates a general framework for optimise learn to rank problems
diff --git a/unfinished/GaussianProcesses/gaussian_processes_from_scratch.ipynb b/unfinished/GaussianProcesses/gaussian_processes_from_scratch.ipynb