commit 7c2a9ca36cebdfa27a8fa752d81b6d2a10d27992
parent 33978c4f87c919066cd1a299b31528e288e01d6e
Author: AndrewLockVI <andrewlaack1@gmail.com>
Date: Thu, 9 Jan 2025 09:22:59 -0600
Refactor
Diffstat:
1 file changed, 17 insertions(+), 0 deletions(-)
diff --git a/lecture-notes/DeepLearning/01.09.25.md b/lecture-notes/DeepLearning/01.09.25.md
@@ -0,0 +1,17 @@
+- KnowledgeBaseApproach
+- PolarCoordinates
+- RepresentationLearning
+- FactorsOfVariation
+- ComputationalGraph
+- ProbabilisticGraph
+- DistributedRepresentation
+
+Today I read general thoughts about deep learning.
+
+The most interesting points made were the importance and definition of factors of variation, distinctions between probabilistic and computational graphs, and distributed representations.
+
+While lots of general stuff, the phrase 'factors of variation' is one I have thought a lot about and had not considered the term for it. This encompasses all information that impacts predictions about samples.
+
+The distinction between probabilistic and computation graphs is interesting because the former focuses on concepts such as finding faces then eyes whereas the computation graph is simply focused on the longest path of computation to reach the end. This second approach seems more appropriate, but I had not considered how some might thing something like linear regression has a 'depth' anything other than 1.
+
+As far as distributed representation, this is the idea that we handle representations using multiple neurons instead of just one. The case in the reading was made that instead of having 9 neurons to represent blue, red, and white, cars, trucks, or SUVS, we can simply have 6 of which 3 represent colors and 3 represent the vehicle type.
