Papers

My implementations of the seminal papers of machine learning.

Implementation queue:

• 3DGS https://arxiv.org/pdf/2308.04079.pdf
• LSTM (from scratch?)

LeNet-5

• MNIST recognition

3DGS

• 3D Gaussian splatting
• Goal 1 is to have a navigatable 3d model/scene locally
• Goal 2 is to upgrade it to 4d
• Goal 3 is to find a way to make it fast

Further reading:

• https://arxiv.org/pdf/2403.10242.pdf
• https://arxiv.org/pdf/2312.09147.pdf
• https://zouzx.github.io/TriplaneGaussian/
• https://ericpenner.github.io/soft3d/
• https://arxiv.org/pdf/2109.08857.pdf
• https://arxiv.org/pdf/2310.08528.pdf

ResNet

• Residual network
• Very vertical

Devlog/improvements:

• change line 273 to mx.array, changing the effects afterward

Bug log:

• the input channel issue: an extra reshape inside the __call__ messed the reshaping up
• residual operation shape issue bug: the stride was wrong, downsampled a bit too much
• hard coded the batch_size and matrix reshaping; changed it to the variable names and was fixed