Chizoba Obasi

2026-02-09 • information theory machine learning optimization deep learning mathematics resources

A deep dive into the mathematical foundations of machine learning, emphasizing the need for rigorous understanding over reliance on high-level abstractions.

2026-03-13 • reinforcement learning monte carlo methods algorithms eligibility traces td learning

Eligibility traces enhance reinforcement learning by bridging TD and Monte Carlo methods, enabling efficient online learning and immediate behavior updates.

2026-03-09 • reinforcement learning stability gradient descent approximation theory off-policy methods

The post examines the challenges and solutions of extending off-policy methods to function approximation in reinforcement learning, focusing on convergence issues and variance reduction techniques.

2026-03-09 • management machine learning reinforcement learning approximation theory sarsa

This post details the semi-gradient Sarsa method in reinforcement learning, highlighting its application in both episodic and continuing tasks while transitioning from discounted to average reward settings.

2026-03-01 • machine learning evaluation color fidelity nigerian languages asr bug

The evaluation of Facebook's omniASR-CTC-1B ASR model for Igbo shows a critical 75.5% loss in tonal diacritics, indicating reliance on orthographic rather than acoustic cues.

2026-02-24 • dynamic programming machine learning planning reinforcement learning algorithms

The post examines the distinctions between Model-Based and Model-Free Reinforcement Learning methods, focusing on planning techniques and the estimation of value functions.

2026-02-27 • neural networks machine learning reinforcement learning utility function approximation theory

Function approximation in reinforcement learning is crucial for estimating state-value functions, utilizing methods like linear models and neural networks to enhance generalization and performance.

2026-03-04 • metadata machine learning nigerian languages asr bug tonal languages

The evaluation of Meta's omniASR reveals a 75.5% loss of tonal accuracy in Igbo, highlighting significant shortcomings in language support for tonal languages.

2024-07-15 • neural networks machine learning python pytorch deep learning

This blog post details the implementation of Andrej Karpathy’s Neural Networks: Zero to Hero lecture series using Jupyter Notebook. It covers foundational concepts of neural networks, including backpropagation, multi-layer percept...

2025-05-30 • neural networks machine learning deep learning gradient descent backpropagation

This blog post provides a comprehensive overview of backpropagation, a fundamental algorithm for training deep neural networks. It explains the mechanics of backpropagation, including the forward pass, loss function, backward pass...

2025-02-24 • nuclear energy renewable energy decarbonization natural gas u.s. energy policy

In 'Power Hungry', Robert Bryce advocates for a transition from coal and oil to a combination of natural gas and nuclear energy (N2N) as a solution for U.S. energy autonomy. He argues that N2N meets essential criteria such as ener...

2020-09-20 • nuclear energy renewable energy environmental issues natural gas u.s. energy policy

The blog post critiques Robert Bryce's book 'Power Hungry,' which advocates for natural gas and nuclear energy as the next steps in the U.S. energy transition from coal and oil. It discusses the importance of considering geographi...

2026-01-22 • neural networks optimization deep learning stochastic gradient descent adam

This guide explores seven essential optimization techniques for training neural networks, detailing their mechanisms, advantages, and appropriate use cases.

2025-05-30 • machine learning python mathematics resources algorithms data science and machine learning

This blog post serves as a comprehensive guide to the essential equations in machine learning (ML), covering topics from probability and information theory to linear algebra and advanced ML concepts. It includes theoretical insigh...

2024-08-02 • neural networks python natural language processing deep learning rnn

This blog post provides a comprehensive guide to implementing a character-level language model using a Recurrent Neural Network (RNN) in Python. It covers the preparation of data, the architecture of the RNN, the forward and backw...

2026-03-16 • web development audiobooks text-to-speech personal projects inkdrop

Inkcast is a free, browser-based audiobook player developed to enhance the listening experience of EPUBs and PDFs, addressing personal frustrations with existing tools.