lrp eq

pages/_drafts/layer-wise-relevance-propagation.md

@@ -0,0 +1,55 @@
+---
+title: Layer-Wise Relevance Propagation
+tldr: 
+tags: 
+references: 
+aliases: 
+crossposts: 
+publishedOn: 
+editedOn: 
+authors:
+  - "[[Yoann Poupart]]"
+readingTime:
+---
+> [!caution] WIP
+> 
+> This article is a work in progress.
+
+> [!tldr] TL;DR
+> 
+> LRP is a method that produces pixel relevances for a given output which doesn't to be terminal. Technically the computation happens using a single back-progation pass. 
+
+
+> [!example] Table of content
+> 
+> - [Hot Takes](#hot-takes)
+> - [Short-term vs Long-term](#short-term-vs-long-term)
+> 	- [The Industry Pressure](#the-industry-pressure)
+> 	- [The Curse of AI Doomerism](#the-curse-of-ai-doomerism)
+
+## LRP Framework
+
+### Formulations
+
+With $R_j^{[l]}$ being the $j$-th neuron's relevance of the layer $l$, and the propagation mechanism is given by the equation $\ref{eq:aggregate}$.
+
+$$
+\begin{equation}
+%\label{eq:aggregate}
+R_{j}^{[l]}=\sum_{k}\dfrac{w_{jk}}{\sum_j w_{kj}}R_k^{[l+1]}
+\end{equation}
+$$
+### Different Rules
+
+### Technical Details
+
+## Classification Example
+
+### Network Decomposition
+
+### Interpretation
+
+> [!quote] References
+> 
+> [1] [[On Pixel-Wise Explanations for Non-Linear Classifier Decisions by Layer-Wise Relevance Propagation]]
+> [2] [[A Rigorous Study Of The Deep Taylor Decomposition]]

pages/_stories/my-approach-to-ai-safety.md

@@ -1,6 +1,6 @@
 ---
 title: My Approach to AI Safety
-tldr: Shortcoming issues should not be undermined because, if not tackled immediately, they will make [Alignment](https://en.wikipedia.org/wiki/The_Alignment_Problem) a lot more complex. I am convinced that interpretability will be the best tool for monitoring and control, and for that, I will pursue this agenda through research and entrepreneurship.
+tldr: Shortcoming issues should not be undermined because, if not tackled immediately, they will make Alignment a lot more complex. I am convinced that interpretability will be the best tool for monitoring and control, and for that, I will pursue this agenda through research and entrepreneurship.
 tags:
   - AIS
   - Agenda
@@ -20,7 +20,7 @@ readingTime: 11
 
 > [!example] Table of content
 > 
-> - [My Claims](#my-claims)
+> - [Hot Takes](#hot-takes)
 > - [Short-term vs Long-term](#short-term-vs-long-term)
 > 	- [The Industry Pressure](#the-industry-pressure)
 > 	- [The Curse of AI Doomerism](#the-curse-of-ai-doomerism)
@@ -34,7 +34,7 @@ readingTime: 11
 > 	- [Brief Agenda](#brief-agenda)
 > 	- [What's next?](#whats-next)
 
-## My Claims
+## Hot Takes
 
 - It is necessary to pursue research for short-term and long-term AIS in the meantime 
 - Entrepreneurship can be highly valuable to contribute to AIS