OUTTA Basic — Organizing BERT, Hugging Face, and Gemini Safely
I focused on tokenization, attention masks, BERT output shapes, and environment-based credential handling for Gemini.
Read noteBlog Category
AI · Deep Learning · Computer Vision
Separate learning paths for discriminative AI, vision, and generative models.
Subcategories
OUTTA Basic — Following the Time Axis in RNNs and Seq2Seq
I revisited hidden states, sequence shapes, train-test divergence, and teacher forcing through retained RNN learning curves.
Read note
OUTTA Basic — Connecting Language Models, Embeddings, and Autoencoders
I connected TF-IDF cosine similarity, dense embeddings, and an autoencoder latent space as different representation-learning tools.
Read note
OUTTA Basic — PyTorch CNN and Handwritten-Digit Error Analysis
I traced convolution shapes and read retained loss, accuracy, correct-sample, and misclassification plots together.
Read note
OUTTA Basic — Neural Networks for Function Approximation and Nonlinear Classification
I compared sigmoid-basis function approximation, spiral decision boundaries, and a stalled classifier to understand network capacity and training failure.
Read note
OUTTA Basic — From Data Exploration to Autodiff and Linear Regression
I revisited scatter plots, loss surfaces, gradient descent, and normalization using the linear-regression outputs retained in my notebook.
Read note
From DDPM to Latent Diffusion and DiT
Notes connecting forward noise, reverse denoising, latent space, and a transformer backbone.
Read note
From GANs to Pix2Pix
Notes connecting alternating generator and discriminator updates with Pix2Pix conditioning and L1 loss.
Read note
VAE ELBO and Reparameterization
Notes on why the encoder predicts a mean and variance and how reconstruction and KL losses interact.
Read note
Autoregressive Factorization and MLE
Notes connecting joint-probability factorization, teacher forcing, log-likelihood, and cross entropy.
Read note
Starting with Generative Models and Data Distributions
My notes comparing discriminative and generative models before checking tensor shapes and visualizing data.
Read note
Preserving evaluation conditions for AUC and F1 in PPG-HRV classification
Why subject splits, folds, preprocessing, confusion matrices, and paper-reported metrics are not merged into one performance value.
Read note