wassname/discovering_latent_knowledge
1
0
Fork 0
mirror of https://github.com/wassname/discovering_latent_knowledge.git synced 2026-09-11 12:10:11 +08:00
Code Issues Packages Projects Releases Wiki Activity

70% acc

Browse Source
This commit is contained in:
deep1
2023-10-12 17:06:39 +08:00
parent 9223d18bb0
commit 3b4e034ea2
2 changed files with 29 additions and 26 deletions
Download Patch File Download Diff File Expand all files Collapse all files
mjc_notes.md
+4
View File
@@ -1705,3 +1705,7 @@ We use the following linear models during evaluation:
3. K-Means - We take an unlabelled dataset D and perform K-Means clustering with K = 2, hoping to separate high-concept and low-concept samples. We take the difference between the centroids of the two clusters as the concept direction.
4. Mean Difference - We take the difference between the means of high-concept and low- concept samples of the data:
5. Logistic Regression - The weights of logistic regression trained to separate Xhighl and Xlowl on some training data can be used as a concept direction as well.
# 2023-10-12 13:07:28
So I want to try not applying a random but by fitting to a few samples. Probobly mean difference is the easiest. Of course this would be on weights not embeddings.