Mostly Machine Learning

Seppo Enarvi

computer vision, language models, deep learning

Data format is crucial for good training speed

6 minute read

When your training runs start to take days, you might be tempted to throw more GPUs at the problem. It’s easy to underestimate the time needed for tasks that...

Using LLMs for writing Home Assistant components

6 minute read

Why I switched to Home Assistant

Kalman filter equations and extended Kalman filter

6 minute read

Kalman filter

Understanding convergence of SGD

10 minute read

Batch size, learning rate, weight averaging, and solutions that generalize better

REINFORCE

10 minute read

Improving sequence-to-sequence models with reinforcement learning

Seppo Enarvi

Recent Posts

Data format is crucial for good training speed

Using LLMs for writing Home Assistant components

Kalman filter equations and extended Kalman filter

Understanding convergence of SGD

REINFORCE