Writing

Welcome to my blogs! Here you will find peeks into my mind and thoughts. Most posts here are related to AI and Machine Learning, but you may also find some other topics.

Between the Fire and the Sun

A study of the similarity between the representations of diffusion models and primates’ ventral visual stream

Representational Alignment

The field of Artificial Intelligence has drawn much inspiration from biology. In the particular case of Artificial Neural Networks (ANNs), however, this influence is starkly…

Diffusion Models Should Just Go for the Data

Exploration of ‘Back to Basics: Let Denoising Generative Models Denoise’ using Molab

deep generative models

diffusion models

Released earlier this year, “Back to Basics: Let Denoising Generative Models Denoise” (Li & He, 2026) challenged what has been the norm for years for building diffusion…

The Variational Autoencoder

A Deep Dive into Latent Variable Modeling and Latent Space Continuity on the CelebAMask-HQ Dataset

Deep Generative Models

Variational Autoencoders

Latent Variable Models

Informally, a generative model is one that is able to generate new data points where the data could be anything: text, images, audio, etc. This is done through building a…

Organizing Code, Experiments, and Research for Kaggle Competitions

Lessons and tips learnt while earning a Kaggle Competition Medal

Code Organization

Experimentation

Tell me and I forget. Teach me and I remember. Involve me and I learn.

Generating Structured Outputs from LLMs

An overview of popular techniques to confine LLMs’ output to a predefined schema

Structured Output

Today, the most common interface for interacting with LLMs is through the classic chat UI found in ChatGPT, Gemini, or DeepSeek. The interface is quite simple, where the…

What is LoRa and when should it be used?

A look at a memory and storage efficient technique to fine-tune large models

With the recent advancements in LLMs, the size of models continues to grow, and fine-tuning models is becoming increasingly expensive. The rise of this issue led to an…

Implementing the Word2Vec Paper

A look at a classic NLP paper

Paper Implementation

In 2013, the Efficient Estimation of Word Representations in Vector Space paper was released introducing two new model architectures for generating vector representations…

Origins of The Cross-Entropy Loss Function

Gaining Intuition on the Popular Loss Function

The cross-entropy loss function is one of the most widely used loss functions in deep learning, particularly for classification tasks. I’ve used it before and understood the…

Using BERT to predict disaster Tweets

In 2019, Kaggle hosted the Natural Language Processing with Disaster Tweets competition. It was made for contestants to test their ability to build a machine learning model…

Building a Receipt Understanding Model

Fine-tuning LayoutLMv3 on the SROIE dataset

document understanding

image processing

Receipts are a common way to store information about the items purchased. They are used by customers to keep track of their expenses and by businesses to keep track of their…

Exploring Egypt’s College Admission Test

2022 Sanawya Amma Results Analysis

data visualization

Every year, high school seniors rise to one of their largest life challenges: College Admission. Each country has its system for college admission shaping the career journey…

2024 Olympic Medalists

A Nation-Level Analysis

data visualization

On July 26, 2024, the Paris 2024 Olympics commenced, sending lights of hope into the hearts of every athlete. As the world watched in anticipation, dreams were forged and…